1 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Bio-Medical Computing (6.872/HST.950)
Peter Szolovits, PhD Gil Alterovitz, PhD + guest lecturers
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 2 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Medical Informatics
• Intersection of medicine and computing • Plus theory and experience specific to this combination • =Medical Computing, ~Health Informatics
• Science • Applied science • Engineering
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 3 www.onlineeducation.bharatsevaksamaj.netTypes of www.bssskillmission.in Bio-Medical Informatics • Cellular level: Bioinformatics, Systems Biology • Patient level: Clinical Informatics, Health I., Medical I., … • Population level: Public Health I. • Imaging Informatics
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 4 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Bio-Medical Informatics
• Phenotype = Genotype + Environment • In humans, we rely on “natural experiments” • Measurements – Genotype: sequencing, gene chips, proteomics, etc. – Environment: longitudinal surveys, etc. – Phenotype: clinical records, assembled to longitudinal data
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 5 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Outline (today) • What is biomedical informatics? • BMI is defined by goals and methods of health care • The genomic revolution • The science of health care – Genotype, phenotype, environment – From associations to mechanisms • What is health? • Practice of health care • Challenges
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 6 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Outline (semester) • Clinical and Genomic Data • Methods of modeling • Combining clinical and genetic data • “Translational medicine” • Engineering the health care system • Decision support to improve health care • Personalized medicine • Public health • The developing world • Your projects www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 7 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in The Medical Cycle
interpret information data
patient formulate
therapy diagnosis plan initial presentation www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 8 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Care Processes
• Data: instrumentation, monitoring, telemetry • Information: interpretation, filtering, sampling, smoothing, clustering • Diagnosis: inference, model-based reasoning, classification • Prognosis: prediction, natural course, experience • Therapy: planning, predicting effects, anticipating www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 9 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Meta-level processes
• Acquisition and application of knowledge • Education • Quality control and process improvement • Cost containment • Reference (library)
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 10 www.onlineeducation.bharatsevaksamaj.netEnterprise-level Clinical www.bssskillmission.in Process Automation...
Health Episode Authorization Schedule Membership Status Mgt. Activation
Visit Refer Activation
Plan Design Health Mgt. Health Self Care Mgt. Team Record Assess
Measure Care Community Evaluate Plan Team care
Account Discharge Act Dismiss
Rad Surgery
Lab
Int. Med Pharm
Image by MIT OpenCourseWare. Adapted from figure by David Margulies.
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in figure from David MarguliesWWW.BSSVE.IN 11 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in The “Learning Health Care System”
Model Plan Intervene
Health Episode Authorization Schedule Analyze Membership Status Mgt. Activation
Visit Refer Activation
Plan Design Health Mgt. Health Self Care Mgt. Team Record Assess Observe/ Measure Measure Care Community Evaluate Plan Team care
Account Discharge Act •Process Dismiss •Medical content Rad Surgery Lab
Int. Med Pharm
Image by MIT OpenCourseWare. Adapted from figure by David Margulies. www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 12 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Dogma
Phenotype = Genotype + Environment Traits Gene sequence Diet, smoking, drugs, … Diseases SNP’s Insults and injuries Behaviors Expression data Exposures … … …
• What is the functional form? • How do we investigate these relationships? • Can we take advantage of the exponential growth of genomic data? www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 13 www.onlineeducation.bharatsevaksamaj.netGrowth in Gene www.bssskillmission.in Expression Omnibus Measurements Fall 2004: ~30,000 submissions, ~. 5B measurements
Today (9/2010): ~472,929 samples
http://ncbi.nlm.nih.gov/geo/ Figure 4 from Edgar, Ron and Alex Lash. "Chapter 6 The Gene Expression Omnibus (GEO): A gene expression and hybridization repository." The NCBI Handbook, National Library of Medicine, 2003. www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 14 www.onlineeducation.bharatsevaksamaj.netWhere are the www.bssskillmission.in Phenotype and Environment-related Data? Phenotype = Genotype + Environment
Traits Gene sequence Diet, smoking, drugs, Diseases SNP’s … Behaviors Expression data Insults and injuries … … Exposures … • Perform Controlled Experiments? – Unethical using human subjects!!! – OK on rats.
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 15 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Experimental Subjects
Image by Randall McIlwaine on CartoonStock.com has been removed due to copyright restrictions. Researcher holding hat with Mickey Mouse ears says to another researcher, "We've run out of lab rats, Henderson... Put this on and come with us."
Image by DakotaPrarieNova on Flickr.
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 16 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in High-throughput phenotyping at Medical College of Wisconsin
Image of laboratory at Medical College of Wisconsin have been removed due to copyright restrictions.
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 17 www.onlineeducation.bharatsevaksamaj.netWhere are the www.bssskillmission.in Phenotype and Environment-related Data? • Environment – (Hardest to get) – Questionnaires, • e.g., Nurses’ Health Study, Framingham Heart Study – Monitoring • e.g., LDS hospital infectious disease monitors • Phenotype – “Natural Experiments” – ∴ Clinical Data
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 18 www.onlineeducation.bharatsevaksamaj.netThe fantasy: Informatics www.bssskillmission.in for Integrating Biology & Bedside
Family Hx l2b2: l. Kohane, et al.
Consent Create new versions of each of the images
Create new versions of each of the images
Create new versions of each of the images
Create new versions of each of the images Text
Create new versions of each of the images Text Text Text Text
Annotation
Banking Genomics
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in WWW.BSSVE.INImage by MIT OpenCourseWare. Adapted from Peter Szolovits. 19 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Plausibility Butte & Kohane, Nature Biotech 2006 • Phenome-Genome Network – Gene Expression Omnibus • expression data • annotations: tissue, disease, expt. conditions, … – Interpret annotations to UMLS – Differential expression vs. condition – Interesting relations: • 11 genes & aging • DDX24 and leukemia • 2 genes & injury
Reprinted by permission from Macmillan Publishers Ltd: Nature Biotechnology. Source: Butte, Atul J. and Isaac S. Kohane. "Creation and implications of a www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.inphenome-genome www.bsslifeskillscollege.in network." Nature Biotechnology 24 (2006). © 2006. 20 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Clinico-Genomic Research
• Identify a highly specific clinical population, and controls • Gene-wide association studies (GWAS) • Hope that notable differences appear between those with/those without disease • Disease models: – Mendelian – Single-nucleotide polymorphisms – Private variation 20 –? www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 21 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Time scale in medicine
• Cure—usually acute illness • Manage—long-term, chronic illness • Prevent • Predict (especially based on genetics)
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 22 www.onlineeducation.bharatsevaksamaj.netWHO Constitution www.bssskillmission.in defines “health” “a state of complete physical, mental and social well-being and not merely the absence of disease or infirmity” • Physical • Mental • Social —very hard to measure
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 23 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Distribution of Ages
• Life table deaths by year (Japan, 1989)
Graph showing distribution of ages in Japan has been removed due to copyright restrictions. The graph shows a far greater number of elderly people than younger people.
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 24 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Life table death rates by age
1.0000
0.1000
male p(death) female p(death) 0.0100
US SSA 1997 0.0010
0.0001 1 2 3 4 5 6 7 8 9 11 10 12 13 14 15 16 17 18 19 10 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 111 1 1 1 1 1 1 1 1 1 www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in100 101 102 103 104 105 106 107 108 109 120 25 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Life table cohort survival
100,000
75,000
males alive 50,000 females alive
25,000
0
1 2 3 4 5 6 7 8 9 11 12 13 14 15 16 17 18 19 10 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 111 110 1 1 1 1 1 1 1 1 100 101 102 103 104 105 106 107 108 109 US120 SSA 1997 www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 26 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Measures of Health
• Longevity at birth (CIA World Fact Book, 2001) Country Male Female Rwanda 38.35 39.65 Kenya 46.57 48.44 South 47.64 48.56 CambodiaAfrica 54.62 59.12 Brazil 58.96 67.73 Russia 62.12 72.83 Albania 69.01 74.87 USA 74.37 80.05 Japanwww.bsscommunitycollege.inWWW.BSSVE.IN 77.62 www.bssnewgeneration.in 84.15 www.bsslifeskillscollege.in 27 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Causes of death (industrialized countries, 1989)
Circulatory 48% system Malignant 19% neoplasms Accidents 7%
Others 26%
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 28 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Quality of life
• Value of a total life depends on – Length (assume now is N) – Quality (q) over time – Discounts (g) for future or past (depends very much on what the value is to be used for)
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 29 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Modeling life quality
Figure showing four hypothetical survival scenarios showing survival from death, onset of disease, and onset of disability has been removed due to copyright restrictions.
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 30 www.onlineeducation.bharatsevaksamaj.netMortality, Disability, www.bssskillmission.in Morbidity
100000 100000
80000 80000
60000 60000
40000 40000
20000 20000
0 0 0 10 20 30 40 50 60 70 80 90 100 110 0 10 20 30 40 50 60 70 80 90 100 110
100000 100000
80000 80000
60000 60000
40000 40000
20000 20000
0 0 0 10 20 30 40 50 60 70 80 90 100 110 0 10 20 30 40 50 60 70 80 90 100 110
Mortality Disability Morbidity
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in WWW.BSSVE.IN Image by MIT OpenCourseWare. 31 www.onlineeducation.bharatsevaksamaj.netTop 10 Chronic www.bssskillmission.in Conditions Persons aged ≥ 65
Condition Both Male Female Arthritis 49.6 40.7 55.7 Hypertension 39.0 33.0 43.2 Hearing impairment 30.0 35.2 26.3 Heart disease 25.7 26.9 24.9 Orthostatic impairment 16.8 15.7 17.8 Cataracts 15.5 11.3 18.4 Chronic sinusitis 15.2 13.7 16.2 Visual impairment 10.1 12.0 8.8 Genitourinary 9.9 11.3 8.9 Diabetes 8.9 7.8 9.7 U.S. Nat’l Ctrwww.bsscommunitycollege.inWWW.BSSVE.IN Health Stat, Vital www.bssnewgeneration.in and Health Statistics www.bsslifeskillscollege.in, 1985 (1982 data) 32 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Societal quality of life
• Aggregation of individual qualities + Equity (distributions)
• Is more better? (Population control.) • Is less better? • How much to spend?
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 33 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Who makes decisions? “In those days there was no bureaucratic regimentation, there were few forms to fill out, malpractice premiums were affordable, and the overhead costs of running a practice were reasonable. Our bills were simple, spelled out so anybody could understand them without the use of codes. Patients usually paid their own bills, promptly too, for which an ordinary receipt was given. Hospital charges were set by the day, not by the aspirin. Medical care was affordable to the average person with rates set by the laws of the marketplace, and care was made available to all who requested it regardless of ability to pay. Doctors were well respected; rarely were we denigrated by a hostile press for political reasons. Yes, in the days before government intervention into the practice of medicine, doctor’s fees were low, but the rewards were rich; those were truly the ‘golden years’ for medicine.” Edward Annis, past President of AMA Code Blue, 1993 www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 34 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Aggregation
• Trend: social aggregation leads to decisions at a larger scale – Multi-specialty providers – Government guarantees and mandates – Risk sharing – Oregon-wide spending “optimization”; – British NHS
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 35 www.onlineeducation.bharatsevaksamaj.netChanging Context www.bssskillmission.in of Health Care • Fee-for-service • HCFA (Health Care Financing Agency) pays for Medicare • Capitation – HMO’s (Health Maintenance Organizations) take overall responsibility to care for patient for fixed fee – Pushing risk down to the physician or group
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 36 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Determining Factors: $£€¥R
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 37 Exponentiallywww.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in growing expense of health care • More healthcare than steel in GM cars • Increased demand – Much more possible – Better tests, therapies – High human motivation • No pushback • Waste – Unnecessary procedures • ½ of health expenses in last year of life – Marginally useful procedures • Defensive medicine – Bad Medicine • IOM: 48-98K “unnecessary” deaths/year www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 38 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Managed Care
“Decisions that were once the exclusive province of the doctor and patient now may be examined in advance by an external reviewer—someone accountable to an employer, insurer, health maintenance organization (HMO), or other entity responsible for all or most of the cost of care. Depending upon the circumstances, this outside party may be involved in discussions about where care will occur, how treatment will be provided, and even whether some treatments are appropriate at all.” Controlling Costs and Changing Patient Care IOM, 1989 www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 39 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in How is care managed?
• Active case management: – Preadmission review – Continued-stay review – Second surgical opinion • Selective case management—high-cost cohorts • Institutional – Capitation – Institutional arrangements (referrals, hospitals, pharmacies, …) – Control “leakage”
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 40 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Managed Care Scorecard
• “U.M. has helped to reduce inpatient hospital use and to limit inpatient costs…” • “The impact of U.M. on net benefit costs is less clear. Savings on inpatient care have been partially offset by increased spending for outpatient care and program administration.” • “U.M. … does not appear to have altered the long-term rate of increase in health care costs.” IOM, 1989
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 41 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Obama Proposals
• Universal coverage: everyone must get insurance – Employer – Insurance collaborative – Government (?) — rejected • Insurance companies cannot deny insurance, cancel coverage, impose reimbursement limits based on illness, past or present • Government assistance to poor people, small companies • Health Information Technology (HIT) to smooth info flow • Cost savings from avoiding billing disputes, ceasing to reimburse only procedures, evidence-based medicine. www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 42 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Quality Improvement
• IOM Study: 96,000 US deaths/year from medical error (perhaps half preventable?) • Information intervention at the point of decision making can improve decisions • DPOE: Direct Physician Order Entry allows such intervention • Leapfrog Group: Large employers ($$$) require DPOE from providers • Patient Involvement: Indivo Health, Google Health, Microsoft Healthvault
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 43 www.onlineeducation.bharatsevaksamaj.netImplications www.bssskillmission.in of Health Care Organization for Informatics • Money determines much – Medicine spends 1-2% on IT, vs. 6-7% for business overall, vs. 10-12% for banking – “Bottom line” rules, therefore emphasis on • Billing • Cost control • Quality control, especially if demonstrable cost savings • Retention and satisfaction (maybe) – Management by accountants
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 44 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Challenges
• Computerized Medical Records (EMR/ CPR/…) • Usability of systems in the workflow of health care • Large-scale “Engineering Systems” problem • Genomic Medicine
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 45 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in MIT OpenCourseWare http://ocw.mit.edu
HST.950J / 6.872 Biomedical Computing Fall 2010
For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 46 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Nature of Medical Data
6.872/HST950
Peter Szolovits
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 47 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Outline
• Recall context of current medical practice • History of medical record keeping • Organization of medical records • Computerized medical records – Why – Key issues – Failures and successes • Current approaches
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 48 www.onlineeducation.bharatsevaksamaj.netImplications www.bssskillmission.in of Health Care Organization for Informatics • Money determines much – Medicine spends 1-2% on IT, vs. 6-7% for business overall, vs. 10-12% for banking – “Bottom line” rules, therefore emphasis on • Billing • Cost control • Quality control, especially if demonstrable cost savings • Retention and satisfaction (maybe) – Management by accountants
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 49 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Why Keep Records?
• Basis for historical record • Communication among providers • Anticipate future health problems • Record standard preventive measures • Identify deviations from the expected • Legal record • Basis for clinical research
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 50 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Who Keeps Records?
• Doctor • Nurse • radiologist • Office staff, • pharmacist admissions • patient • Administrator • physical therapist • lab personnel
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 51 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Forms of Clinical Data
• Numerical Measurements • Coded (?) discrete data – Lab data – Family history – Bedside measurements – Patient’s medical history – Home instrumentation – Current complaint • Symptoms (patient) • Recorded signals (e.g., • Signs (doc) ECG, EEG, EMG) – Physical examination • Images (X-ray, MRI, CAT, – Medications Ultrasound, Pathology, • Narrative text …) – Doctor’s, nurse’s notes • Genes (SNPs, expression – Discharge summaries arrays, pedigrees, …) – Referring letters
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 52 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Organization of Data
• Doctor’s journal (traditional) • Time order of collection, per patient (Mayo) • Source of data • Problem-Oriented Medical Record (POMR) (L. Weed, 1969) – Notes organized by problems – SOAP: subjective, objective, assessment, plans
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 53 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in POMR
Data Base Problem List
Progress Notes Plans (by problem) (by problem)
diagnostic, therapeutic, patient education www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 54 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in The Data Base
• Identifying information (name, age, sex, race, religion, insurance info, etc.) • Patient profile (occupation, education, marital status, children, hobbies, worries, moods, sleep patterns, habits, etc.) • Medical history – Chief complaints – History of present illness – Past medical history – Review of systems – Family history – Medications • Physical examination • Laboratory data and physiologic tests (complete blood count, electrocardiogram, chest x-ray, creatinine, urinalysis, vital capacity, tonometry,www.bsscommunitycollege.inWWW.BSSVE.IN etc.) www.bssnewgeneration.in www.bsslifeskillscollege.in 55 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in The Problem List
• “those features in the patient’s psychobiological makeup that require continuing attention” – Social history – Risk factors – Symptoms – Physical findings – Lab tests • Causally organized; e.g., GI bleeding caused by duodenal ulcer appears under the ulcer
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 56 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Example Problem List No Active Date Inactive Date 1 Hypertension 1953 2 Recurrent bronchitis 1958 3 Penicillin allergy 1958 4 S/P pyelonephritis 1960 5 Gallstones Oct 1972 Cholecystectomy Mar 1973 6 Arthralgias Mar 1973 #9 June 1973 7 Pleurisy Mar 1973 #9 June 1973 8 Proteinuria Apr 1973 #9 June 1973 9 SLE June 1973 10 Unemployment Nov 1973 www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 57 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Problem-Related Plans
• Diagnostic: lab tests, radiology studies, consultations, continued observations, … • Therapeutic: medications, diet, psychotherapy, surgery, … • Patient education: instruction in self-care, about goals of therapy, prognosis, …
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 58 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Plans per problem
1. Diarrhea Dx: • stool for occult blood, culture, ova, and parasites, microscopic fat; and muscle fibers • Sigmoidoscopy • Barium enema if persistent Rx: Avoid foods that exacerbate Ed: Informed that more info is needed to make a diagnosis, will aim for symptomatic therapy for now.
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 59 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Plans per problem (cont.)
2. Pyuria Dx: • BUN • Repeat urinalysis • Urine culture 3. Obesity Rx: 1500 kcal diet, Weight Watchers Ed: Dangers of obesity cited. Goal: 170 lbs.
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 60 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Progress Notes
• Subjective: interval history, adherence to program • Objective: physical findings, reports of lab, x-ray, other tests • Assessment: Appraisal of progress, interpretation of new findings, etc. • Plan: Dx, Rx, Ed.
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 61 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Example SOAP Note
#3 RHD with mitral stenosis S: 2 flight dyspnea, mild fatigue. No orthopnea, hemoptysis, ankle edema. Child has strep throat. O: BP 120/70. P 78 regular Neck veins normal, lungs clear. Grade iii diastolic rumble, wide opening snap, P2 slightly ↑ A: Stable. Catheterization still not indicated. Risk of strep throat present. P: Dx: Cardiac fluoroscopy Rx: Continue chlorothiazide and penicillin V 250mg b.i.d.—2 weeks Ed: Reinstructed about antibiotic coverage for tooth extractions, sched. for next month. (Will contact oral surgeon.)
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 62 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in POMR characteristics
• Augment with data flow sheets • Importance of clinical judgment • Benefits: – Communication among team members, explicitness – Education and audit – Clinical research
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 63 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in POMR evidence
• Difficult adoption • Some duplication • Some doctors liked it • Paper-based POMR slow, computer- based maybe faster • Demand-oriented MR: by time, by source, by problem, etc. Dynamic arrangement.
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 64 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Mayo experience
• Paper records, mostly • Pneumatic tube delivery, therefore limited size • Formal procedures for reaping and organizing records at discharge • Comprehensive index
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 65 www.onlineeducation.bharatsevaksamaj.netThe Computer-based www.bssskillmission.in Patient Record • IOM Study: Dick, R. S. and Steen, E. B., Eds. (1991). The Computer-Based Patient Record: An Essential Technology for Health Care. Washington, D.C., National Academy Press. • Made strong case for CPR • Recommended CPRI (Institute), but it never caught on • Today’s standards grow more out of communication standards: HL7 (labs) and DICOM (digital images) www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 66 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Paper record: Strengths
• Familiar; low training time • Portable to point of care • No downtime • Flexibility; easy to record subjective data • Browsing and scanning – Find information by unanticipated characteristics (e.g., Dr. Jones’ handwriting)
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 67 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Paper record: Weaknesses
• Content: missing, illegible, inaccurate – E.g., one hospital study: 11% of tests were repeats to replace lost information – Too thick (1.5 lbs avg.) – Fail to capture rationale – Incomprehensible to patients and families
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 68 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Sample paper record defects
• 75% of face sheets had no discharge disposition, 48% no principal Dx • Agreement between encounter (witnessed) and record: 29% med hx, 66% Rx, 71% info re current illness, 72% tests, 73% impression/Dx, 92% chief complaint • 20.8% of Medicare discharges coded incorrectly (DRG inflation)
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 69 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in More paper record defects
• Unavailable at up to 30% of patient visits – Two clinic visits in a day – Docs keep records in their office – Failure to deliver – Misfiled in file room • Discontinuity across institutions – In/outpatient records separate
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 70 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Ethnographic Design
• Xerox PARC analysis of office work – Sociologists, Anthropologists, Engineers – Much of work is • communication, • assignment of responsibilities, • problem solving
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 71 www.onlineeducation.bharatsevaksamaj.netMedicine is www.bssskillmission.in an Information Industry • 35-39% of hospital operating costs due to professional and patient communications • Physicians spend 38%, nurses 50% of their time charting • Exponential growth of medical knowledge and literature
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 72 www.onlineeducation.bharatsevaksamaj.netIndividual www.bssskillmission.in Users of Patient Records • Providers • Management – Chaplains – Administrators – Dental hygienists – Financial managers and accountants – Quality assurance managers – Dentists – Records professionals – Dietitians – Risk managers – Lab technicians – Unit clerks – Nurses – Utilization review managers – Occupational therapists • Reimbursement – Optometrists – Benefit managers – Insurers (Fed, State, private) – Pharmacists • Other – Physical therapists – Accreditors – Physicians – Gov’t policymakers, legislators – Physician assistants – Lawyers – Podiatrists – Health care researchers, clinical investigators – Psychologists – Health Sciences journalists and editors – Radiology technologists – Patients, families – Respiratory therapists – Social workers www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 73 www.onlineeducation.bharatsevaksamaj.netInstitutional www.bssskillmission.in Users of Patient Record • Healthcare Delivery • Reimbursement – Alliances, associations, networks, – Business Health coalitions systems of providers – Employers – Ambulatory surgery centers – Insurers – Donor banks (blood, tissue, organs) • Research – Disease registries – HMO’s – Health data organizations – Home care agencies – Health care technology developers and – Hospices manufacturers – Hospitals – Research Centers – Nursing homes • Education – Allied health professional schools, – PPO’s medical, nursing, public health schools – Physician offices, group practices • Accreditation – Psychiatric facilities – Accreditation organizations – Public Health Departments – Inst. licensure agencies – Substance abuse programs – Prof. Licensure agencies • Management and Review • Policymaking – Fed, State, Local gov’t agencies – Medicare peer review organizations – Quality assurance companies – Risk management companies www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 74 www.onlineeducation.bharatsevaksamaj.netPrimary www.bssskillmission.in Uses of Patient Record
• Patient care delivery (Patient) • Patient care management – Document services received – Document case mix – Analyze severity of illness – Constitute proof of identity – Formulate practice guidelines – Self-manage care – Manage risk – Verify billing – Characterize use of services – Basis for utilization review • Patient care delivery (Provider) – Perform quality assurance – Foster continuity of care • Patient care support – Describe diseases and causes – Allocate resources – Analyze trends and develop forecasts – Support decision making about Dx and Rx – Assess workload – Assess and manage risk – Communicate between departments – Facilitate care via Clin. Practice Guidelines • Billing and reimbursement – Document patient risk factors – Document services for payment – Bill for services – Assess and document patient expectations and satisfaction – Submit insurance claims – Adjudicate insurance claims – Generate care plans – Determine disabilities (workmen’s comp) – Determine preventive advice – Manage & report costs – Remind clinicians – Perform actuarial analysis – Support nursing care – Document services provided www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 75 www.onlineeducation.bharatsevaksamaj.netSecondary www.bssskillmission.in Uses of Patient Record • Research • Education – Develop new products – Conduct clinical research – Document health care professional experience – Assess technology – Study patient outcomes – Prepare conferences and presentations – Study effectiveness and cost- – Teach students effectiveness of care • Regulation – Identify populations at risk – Evidence in litigation – Develop registries and databases – Foster postmarketing surveillance – Assess cost-effectiveness of record systems – Assess compliance with standards • Industry – Accredit professionals and hospitals – Conduct R&D – Compare health care organizations – Plan marketing strategy • Policy – Allocate resources – Conduct strategic planning – Monitor public health www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 76 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in User Requirements
• Record Content • Record Format – Uniform core data – “Front-page” problem list – Ability to “flip through” the elements record – Standardized coding – Integrated among systems and formats disciplines and sites of care • System Performance – Common data – Rapid retrieval dictionary – 24/7 – Information on – Available @ convenient outcomes of care and places functional status – Easy data input
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 77 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in User Requirements (cont.)
• Linkages • Intelligence – To other info systems (e.g., – Decision support radiology, lab) – Clinician reminders – Transferability of information – “Alarm” systems, customized among specialties and sites • Reporting – With relevant literature – “Derived documents”, e.g., insurance forms – Other registries and institutional – Easily customized output, UI databases – Standard clinical reports, e.g., – To records of other family discharge summary members – Custom and ad hoc reports – E-billing – Trend reports and graphics • Training and Implementation • Control and Access – Easy patient access – Minimal training required – Safeguards of confidentiality – Graduated implementations
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 78 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Why is this hard?
• Characterize edema: – Where? • Thousand diseases, – When? syndromes, clinical states • Few thousand symptoms, – How often? signs, observables – Temporal variation? • Few thousand specific lab – Severity tests – Symmetry • Thousands of meds, – What other variations, combinations, characteristics? routes, dosage schedules, … • Uncertainties in all of • ??? Treatments the above
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 79 www.onlineeducation.bharatsevaksamaj.netNot just www.bssskillmission.in database, knowledge representation • “Sometime before his 5th birthday, Johnny had scarlet fever, which caused changes in his heart sounds.” • LEG WEAKNESS PROXIMAL ONLY • (EDEMA with LOCATION = FACIAL or PERI-ORBITAL, PAINFULNESS = not PAINFUL, SYMMETRY = not ASYMMETRICAL, ERYTHEMA = not ERYTHEMATOUS)
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 80 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in What is the “Right” representation?
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 81 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Inadequate Coding Systems
• Low degree of refinement – E.g., ICD-9’s categories for Chronic Bronchitis • Simple • Mucopurulent • Obstructive • Other • Unspecified • Poor coverage of symptoms • Difficulty of automatic coding – Gabrieli’s 10M-phrase thesaurus
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 82 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Lotte Record
For relevant article, see Kohane, Isaac et al. "Building National Electronic Medical www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in Record Systems via the WorldWWW.BSSVE.IN Wide Web." JAMIA 3, no. 3 (1996). (PDF) 83 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Immunizations www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 84 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Labs Summary www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 85 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Lab Studies
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 86 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Thyroxine
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 87 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Weight
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 88 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in TCH Database
•Documents •DOC_STORE •DOC_ATTRIBUTES •DOC_DESCRIPTION •CHILD_DOCS •Doctors •PERSNL_PUBLIC •PPR •Patients •PAT_DEMOGRAPH •PAT_FIN_ACCT •… www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 89 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Database Demo
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 90 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 91 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 92 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in What Have We Learned?
• Real world is ugly! – Poor (inchoate) design – Non-adherence to design (+historical debris) • Standards desperately needed: – Terminology & Concepts – Structure of relationships – Communication • But, world is quite complex, and different complexity is appropriate for different uses
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 93 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Current Status of EMR
• Fully computerized in many hospitals – billing, labs, pharmacy, medication administration • Some computerization – Physician orders, visit histories, discharge summaries, vaccination records, emergency dept notes, pathology & radiology notes • Little computerization – Anything outside hospitals & large clinics – History, physical, plans, rationale, …
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 94 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Current Ideas
• Improved Coding • Data Capture – Dictation to text, or speech understanding – Text to meaningful code extraction – Comprehensive instrumentation – Capture at point of generation • Integration to Workflow – Direct physician order entry, protocols, expert systems • “Aware” environments www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 95 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in MIT OpenCourseWare http://ocw.mit.edu
HST.950J / 6.872 Biomedical Computing Fall 2010
For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 96 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Databases
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 97 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Why?
• Abstraction of logical from physical structure • Allows separation of a program’s “business logic” from concerns about traversal of the data
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 98 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Types of databases • Object – direct representation of programming language objects • Relational (<== dominant) – Tables – Operations • Select, project, union • Join (natural, inner, outer, left, right, …) – Indexes • Hierarchical (e.g., XML) – Parent-child • Network • Flat files (e.g., spreadsheet, text file) www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 99 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Hierarchical Model
Figure from Association for Computing Machinery removed due to copyright restrictions. See Levin, Michael. "An introduction to DIAM: levels of abstraction in accessing information." Association for Computing Machinery, 1978.
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 100 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Network Model
Figure from Association for Computing Machinery removed due to copyright restrictions. See Levin, Michael. "An introduction to DIAM: levels of abstraction in accessing information." Association for Computing Machinery, 1978.
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 101 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Relational Database Underlying Concepts • Individual entities • Their properties • Relations among them – 1-1 – 1-n (or n-1) – n-n • Integrity • Transactions
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 102 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Relational Model
Figure from Association for Computing Machinery removed due to copyright restrictions. See Levin, Michael. "An introduction to DIAM: levels of abstraction in accessing information." Association for Computing Machinery, 1978.
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 103 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Relational Algebra Operations
• Select—subset of rows with conditions • Project—subset of columns • Join A and B – Outer: cross product of all rows in A and B, result includes all columns of each – Natural: select rows of cross-product in which matching columns have same values – Join on specific column relations (=, >, <, …) • Grouping operations (partition by criteria) • Summarization (count, max, min, average) 8 www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 104 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in MySQL SELECT syntax
SELECT [ALL | DISTINCT | DISTINCTROW ] [HIGH_PRIORITY] [STRAIGHT_JOIN] [SQL_SMALL_RESULT] [SQL_BIG_RESULT] [SQL_BUFFER_RESULT] [SQL_CACHE | SQL_NO_CACHE] [SQL_CALC_FOUND_ROWS] select_expr, ... [FROM table_references [WHERE where_condition] [GROUP BY {col_name | expr | position} [ASC | DESC], ... [WITH ROLLUP]] [HAVING where_condition] [ORDER BY {col_name | expr | position} [ASC | DESC], ...] [LIMIT {[offset,] row_count | row_count OFFSET offset}] [PROCEDURE procedure_name(argument_list)] [INTO OUTFILE 'file_name' export_options | INTO DUMPFILE 'file_name' | INTO @var_name [, @var_name]] [FOR UPDATE | LOCK IN SHARE MODE]] 9 www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 105 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 106 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in MySQL SELECT examples
select * from persnl_public where last_name=‘Bird’;
select pat_num from persnl_public, ppr where persnl_public.persnl_id=ppr.provider_id and persnl_public.last_name=‘Bird’;
select d.last_name,d.first_name from persnl_public as p, ppr, pat_demograph as d where p.persnl_id=ppr.provider_id and ppr.pat_num=d.pat_num and p.last_name=‘Bird’;
select p.last_name,p.first_name,count(*) as c from persnl_public as p join ppr on p.persnl_id=ppr.provider_id group by p.persnl_id having c>1 order by c desc; 11 www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 107 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in MIT OpenCourseWare http://ocw.mit.edu
HST.950J / 6.872 Biomedical Computing Fall 2010
For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 108 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Medical Data, Standard Vocabularies, Communication Standards
6.872/HST950
Peter Szolovits (with some material from Chris Cimino)
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 109 www.onlineeducation.bharatsevaksamaj.netRecall Children’s www.bssskillmission.in Clinicians’ Workstation Database • Demographics • Problems • Allergies • Medications – Immunizations • Lab Data • Clinical Measurements – Growth Charts • Visit History • Reports and Letters
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 110 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in The Database
•Documents •DOC_STORE •DOC_ATTRIBUTES •DOC_DESCRIPTION •CHILD_DOCS •Doctors •PERSNL_PUBLIC •PPR •Patients •PAT_DEMOGRAPH •PAT_FIN_ACCT •PAT_TEST_HISTV •General Information •REMOTE_TEST •CPT_CODE •PHARMACY_TABLE •ICD9 •PROBLEMS •ICD9_PROCDR •PROBLEM_NOSOLOGY www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 111 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Vocabularies and Terminology
• Why? – Surrogate for “messy reality” – Uses • How? – Flat list – Taxonomy (Hierarchy, Nosology, …) – Heterarchy – Combinatorial Language • Derivation rules • Inference • … knowledge representation
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 112 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in “Ontology” for Computer Folks
• An organization of concepts (hierarchy or heterarchy) • (Some) concepts are defined in terms of others – A triangle is a polygon with exactly 3 sides – A dachshund is a dog (with ???) • Automatic classification – If P is a 3-sided polygon with …, it is recognized automatically as a triangle
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 113 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in OWL: “Semantic Web”
• Description Logic – Concepts and Instances – Is-a in virtue of • Primitive assertion: “A dog is a mammal.” • Definition: “A triangle is a three-sided polygon.” – Limited logical power of definition language assures tractable inference • Slot restriction • Number restriction • But, no negation or disjunction – Subsumption inferences are central – Other logical assertions may be made, but are typically not enforced or utilized in DL www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 114 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Definitions
• Word – a set of characters including punctuation delimited by white space. • Term – one or more words used as a unit. • Concept – an idea, action, or thing. • Synonym – two terms for the same concept.
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 115 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Vocabulary Uses
• Indexing – Finding what you want • Cataloging – Putting away what you have – E.g., WHO, DRGs • Knowledge Representation – Representing the facts – Blurring the facts – Creating new shades of meaning
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 116 www.onlineeducation.bharatsevaksamaj.netDescribe www.bssskillmission.in a term for a Laboratory Test • Where was it done? • How was it done? • Under what conditions was it done? • How many minutes after eating carbohydrate was it measured?
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 117 www.onlineeducation.bharatsevaksamaj.netDescribe a Vocabularywww.bssskillmission.in for a Gene • Whose gene? • Gene fragment? • Open Reading Frame? • Promoter + all exons and introns • Promoter + all exons + all introns + other binding sites affecting function? • Final/draft/species/SNP/Alternative splicing?
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 118 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Knowledge vs. Language
• Get two or more people to enumerate terms to describe the same set. – Do any terms match exactly? – Do terms differ by word order? – Do terms differ by word suffix or prefix? – Are there terms that some people think are synonyms that other people think are not?
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 119 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in History of 3 Vocabularies
• MeSH — Index • ICD — Precoodinated • SNOMED — Post-coordinated
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 120 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in History
• The modern history of medical controlled vocabularies begins with the U.S. Army General Surgeon who petitioned Congress to fund a medical library. (~Civil War) • The position eventually became “The US Surgeon General” and the library the National Library of Medicine – http://www.nlm.nih.gov/
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 121 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in History
• Library collection was indexed with Index Medicus (created by NLM) which is published in book form. • Index Medicus was extended to index medical literature articles. • Index Medicus was extended further to provide on-line indexing (1960). This became the Medical Subject Headings (MeSH).
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 122 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in MeSH
• Purpose is to index the medical literature. • Content of MeSH is driven by publications. • Who “owns” MeSH? • What impact do vocabulary changes have?
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 123 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in MeSH – Structure http://www.nlm.nih.gov/mesh/ • MeSH is organized into a series of “trees”. (e.g. physical findings, diseases, chemicals) • A MeSH main heading is a “concept”. (e.g. “Neurologic Disease”, “Epilepsy”) • Main Heading (MH) is often called a term. (Try to avoid doing this.)
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 124 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in MeSH – Structure
• Each MH has a unique identifier. • Each MH may have multiple synonyms. • Each MH may have multiple locations in multiple trees. Each of these “contexts” has a unique tree address. The concept of “context” is synonymous with “multiple inheritance”.
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 125 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in MeSH – Structure
• There is a small set of subheadings (50) that “modify” MH based on tree address. (e.g. “diagnosis” applies to MH in the “Disease” tree but not to the “Chemical” tree). • There is a small set of tag terms (15) which exist unrelated to the rest of MeSH. (e.g., “Review Article”, “Human”, “Animal”)
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 126 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in MeSH – Structure
• Every article is indexed with tag terms. • Every article is indexed with MH terms for focus (main index term) and mention (minor index term). • Every index term is checked for subheadings. • This is all done by trained reviewers. • The MeSH Vocabulary is revised annually.
http://www.nlm.nih.gov/mesh/ www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 127 www.onlineeducation.bharatsevaksamaj.netMESH www.bssskillmission.in Redux— The Genome “Ontology” biological process molecular function cell Image showing genome ontology removed due to copyright restrictions. components
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 128 www.onlineeducation.bharatsevaksamaj.netInternational www.bssskillmission.in Classification of Disease (ICD) • Any agency that dispenses funds for health care needs a way to assess needs and effectiveness. • The United Nations World Health Organization (WHO) funds health care prevention projects world wide and gathers statistics for member nations. • Who “owns” ICD? • What impact will changes have?
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 129 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in ICD – Structure
• ICD is divided into categories based on a 5-digit numeric code. (e.g., “133.21”) • Usually round numbers are more general concepts (e.g., “100” subsumes “130” which subsumes “133”) • The fourth and fifth digit is called a modifier but it isn’t really.
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 130 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in ICD – Structure
• The code is both the concept and the unique identifier. Multiple terms are linked to the same code. • Every patient is coded with as many terms as possible. • Terms should be the most specific one to describe a particular problem.
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 131 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in ICD – Structure
• Coding scheme limits the size of the vocabulary. • Obsolete codes must be reused. • Base ten results in limited flexibility and the need for “other”, “NOS”, and “NOC” terms.
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 132 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in ICD – Structure
• Lack of multiple contexts or multiple inheritance results in duplicate terms. • Lack of overall organization results in ambiguous terms.
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 133 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in ICD – Structure
• ICD has been adopted by most insurance companies as a method for controlling billing and payment. • Economic forces drive how the vocabulary is used which drives how ICD is modified which drives changes in reimbursement which drives how the vocabulary is used… • Who “owns” ICD? • The Vocabulary is revised sporadically.
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 134 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in SNOMED – Structure
• Developed by the American College of Pathologists to overcome the faults of ICD. • Really describes 6 [now 12] different vocabularies, one for each “axis” of a concept (e.g., anatomy, environment, history). • Every concept is built up from a term from each “axis” (e.g., “surgery of” “blue” “nevus” “of left” “forearm”).
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 135 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in SNOMED – Structure
• There is some overlap of the axes so it is possible to form two different versions for the same concept (e.g., “blue nevus” “nevus colored blue”). • There are few rules for how to combine axes terms so it is possible to form valid nonsense terms (e.g., “nevus” “of left” “esophagus”). • Who “owns” SNOMED?
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 136 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in SNOMED Axes
• D – Diseases • M – Morphology • C – Drugs • J – Occupations • F – Function • A – Physical Agents • L – Living Organisms • P – Procedures • X – Manufacturers • S – Social Context • G – Modifiers • T -- Topography
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 137 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in “Postcoordination”
• D5-46210 Acute Appendicitis • G-A231 Acute, D5-46100 Appendicitis NOS • G-A231 Acute, M-40000 Inflammation, T-59200 Appendix • M-41000 Acute Inflammation, T-59200 Appendix • T-59200 Appendix, M-41000 Acute Inflammation
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 138 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Semantics of Postcoordination
T-59200 Appendix
D5-46100 Appendicitis NOS
D5-46210 Acute Appendicitis
G-A231 M-40000 Acute Inflammation
M-41000 Acute Inflammation … in progress www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 139 www.onlineeducation.bharatsevaksamaj.netSNOMED-CT www.bssskillmission.in (Clinical Terminology) • Combined SNOMED + Reed Codes – SNOMED for diseases – Reed for symptoms • Licensed by NLM for anyone in US to use royalty-free – Attempt to encourage standardization – New international organization to maintain it
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 140 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in History
• For everyone who wants to “own” a medical vocabulary, there is a set of terms which are likely to overlap but be inconsistent with every other vocabulary. • Read, CPT, COSTART, ChemAbstracts, … • In theory they are all describing agreed upon concepts. A single standard vocabulary would improve the automated flow of medical information.
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 141 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Ideal Vocabulary
?
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 142 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Ideal Vocabulary
• Boundary • Organization • Completeness • Absence of ambiguity • Growth • Aging
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 143 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Ideal Vocabulary – Definitions
• String – a unique sequence of characters. The same set of characters may represent different concepts. • Lexical variants – synonyms with “minor” differences. Word order, capitalization, and punctuation are usually included. Suffixes (plural) and prefixes may be included. • One man’s lexical variant is another’s synonym.
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 144 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Ideal Vocabulary – Definitions
• Related terms – distinct terms whose concepts overlap in some way. The most used relations are “broader” and “narrower” (e.g., “Neurologic Disease” includes but is broader than “Epilepsy”.) • One man’s related term is another’s synonym.
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 145 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Ideal Vocabulary – Definitions
• Controlled versus “free” text – Freedom of expression – Automatic indexing accuracy • Atomic versus enumerated (Pre vs Post) – Handle the unexpected – Predict what to expect • Definitions – “Free” text versus semantic
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 146 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Unified Medical Language
• The Unified Medical Language System (UMLS) started as an NLM collaborative program with 7 centers around the country. • Early years: – Explore ideas (1986) – One “winner” selected and developed (1988) – Usage Testing (1991)
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 147 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in UMLS – Structure
• Three components – Metathesaurus (META) – Semantic Network – Information Sources Map (ISM) • dropped – Specialist Lexicon
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 148 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in META – Structure
• NOT a controlled vocabulary. • Database of information about other controlled vocabularies. • Contains sufficient info to recreate most of the component vocabularies. • Basic unit is the concept. A concept is linked to multiple strings from multiple vocabularies.
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 149 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in META - Structure
• Each concept-string pair is either a preferred term, synonym, or lexical variant. • The same string may be linked to multiple concepts but a term, synonym, or lexical variant will only link to one concept each. • Other links exist based solely on the existence of those links in a source vocabulary.
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 150 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in META – Structure
• Each concept has only one preferred term chosen from all linked terms based on order of precedence of source vocabularies. With a few exceptions, MeSH is number one. • Each concept is linked to semantic types in the semantic network. • NOT a controlled vocabulary – or is it?
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 151 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
http://www.ncbi.nlm.nih.gov/bookshelf/br.fcgi?book=nlmumls&part=ch02
From UMLS Reference Manual, National Library of Medicine, 2009. 44 www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 152 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
From UMLS Reference Manual, National www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.inLibrary of Medicine, 2009. 45 153 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Semantic Network – Structure
• Small vocabulary that attempts to implement an ideal vocabulary • Terms defined with free text definitions and by linkage.
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 154 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in UMLS Semantic Network
"T001" "Organism" "A1.1" "Generally, a living individual, including all plants and animals." "Homozygote; Radiation Chimera; Sporocyst“
"T002" "Plant" "A1.1.1" "An organism having cellulose cell walls, growing by synthesis of inorganic substances, generally distinguished by the presence of chlorophyll, and lacking the power of locomotion. Plant parts are included here as well." "Pollen; Potatoes; Vegetables"
"T003" "Alga" "A1.1.1.1" "A chiefly aquatic plant that contains chlorophyll, but does not form embryos during development and lacks vascular tissue." "Chlorella; Laminaria; Seaweed“
… 188 terms www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 155 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in UMLS Sem Net: Relations
H: isa R3: functionally_related_to R4: temporally_related_to R: associated_with R3.1: affects R4.1: co-occurs_with R1: physically_related_to R3.1.1: manages R4.2: precedes R1.1: part_of R3.1.2: treats R5: conceptually_related_to R1.2: consists_of R3.1.3: disrupts R5.1: evaluation_of R1.3: contains R3.1.4: complicates R5.10: method_of R1.4: connected_to R3.1.5: interacts_with R5.11: conceptual_part_of R5.12: issue_in R1.5: interconnects R3.1.6: prevents R5.2: degree_of R1.6: branch_of R3.2: brings_about R5.3: analyzes R1.7: tributary_of R3.2.1: produces R5.3.1: assesses_effect_of R1.8: ingredient_of R3.2.2: causes R5.4: measurement_of R2: spatially_related_to R3.3: performs R3.3.1: carries_out R5.5: measures R2.1: location_of R3.3.2: exhibits R5.6: diagnoses R2.2: adjacent_to R3.3.3: practices R5.7: property_of R2.3: surrounds R3.4: occurs_in R5.8: derivative_of R2.4: traverses R5.9: developmental_form_of R3.4.1: process_of R3.5: uses R3.6: manifestation_of R3.7: indicates R3.8: result_of www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 156 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Definition of relations
("RL" "T132" "physically_related_to" "R1" "Related by virtue of some physical attribute or characteristic." "" "" "" "PR" "physically_related_to") ("RL" "T133" "part_of" "R1.1" "Composes, with one or more other physical units, some larger whole. This includes component of, division of, portion of, fragment of, section of, and layer of." "" "" "" "PT" "has_part") ("RL" "T134" "contains" "R1.3" "Holds or is the receptacle for fluids or other substances. This includes is filled with, holds, and is occupied by." "" "" "" "CT" "contained_in") ("RL" "T135" "location_of" "R2.1" "The position, site, or region of an entity or the site of a process." "" "" "" "LO" "has_location") ("RL" "T136" "temporally_related_to" "R4" "Related in time by preceding, co-occuring with, or following." "" "" "" "TR" "temporally_related_to") ("RL" "T137" "co-occurs_with" "R4.1" "Occurs at the same time as, together with, or jointly. This includes is co-incident with, is concurrent with, is contemporaneous with, accompanies, coexists with, and is concomitant with." "" "" "" "CW" "co-occurs_with")www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 157 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Semantic Network
Brain Treated Phenytoin Cortex Effect with Organ
Epilepsy Associated Is a with
Head Trauma Disease
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 158 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in State of the Art
• UMLS is not sufficient. – META is not complete. Still weak for clinical terms (sign and symptom terms). – META has superficial organization. Links between vocabularies is based primarily on lexical matches. Inter-vocabulary links growing slower than total size. – Ambiguous sources mean META is ambiguous.
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 159 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in State of the Art
• Semantic typing does scale up so META and the semantic network can form a starting point. • Semantic rules are being added to SNOMED which may remove its ambiguity problem. This would greatly strengthen SNOMED and META. • Who “owns” the rules?
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 160 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in State of the Art
• Semantic tools are being developed to provide end user management of vocabularies. The same tools would allow users to add (or nominate) new terms and help the user understand the semantic definition of existing terms. • Links back to META allow institutions to “own” a vocabulary while complying with other organizations’ requirements.
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 161 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Representation Languages
• Capabilities – Naming: intensional as well as extensional • E.g., “people who provide me healthcare” – Definitions – Assertions • Examples – K-Rep – GALEN and GRAIL – W3C: RDF, OWL, …
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 162 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Communication Standards
• HL7, DICOM, CorbaMED, XML-based … • HL7: – Messages (unit of transfer, type=“purpose”) • Segments (e.g., header, event-type, pat. id, …) – Fields (character string) • HL7 Details – Optional and repeated fields – Character encoding: •
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 163 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in HL7 Examples
• AD – Address Components:
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 164 www.onlineeducation.bharatsevaksamaj.netHL7 (3.0) www.bssskillmission.in Reference Implementation Model • Attempt at complete specification of health-care related communications (and, by derivation, records) • Top-level types: – Entities – Roles – Acts – Infrastructure (communications) – Other • http://www.hl7.org/ (see RIM documents) www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 165 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in MIT OpenCourseWare http://ocw.mit.edu
HST.950J / 6.872 Biomedical Computing Fall 2010
For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 166 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Medical Natural Language Processing 6.872/HST950
WWW.BSSVE.IN
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 167 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
The Dream
• Develop a comprehensive, precise language of expression for all clinical data • It’s the language that is precise.Thus, it must be able to state imprecision, uncertainty, etc. • Translate all actual clinical text into this language • Develop reasoning/inference methods to draw consequences WWW.BSSVE.INwithin this language • Get clinicians (and others) to use this
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 168 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
The Reality
• Most clinical records of observations, interpretations and procedures are stated in free-form natural language • There are many sources of error and ambiguity • Language is infinitely varied • Computers are still poor at doing most text analysis tasks But, with significant exceptions, especially for narrow tasks WWW.BSSVE.IN• • Different approaches work best for different tasks -- no universal methods
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 169 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Structure in a CHB ED Note
Patient seen: 11:45 AM 21 year old male patient injured his right knee. The injury occurred when he was tackled while playing football 2 days ago. He complains of pain and swelling along the medial aspect of the right patella, medial collateral ligament of the right knee and medial collateral ligament of the right knee. He has been able to bear weight. His symptoms are exacerbated by bending his knee.. He has used a knee immobilizer. With some relief. CURRENT MEDICATIONS: None. ALLERGIES: Denies known allergies. IMMUNIZATIONS: Up to date. PE: Alert. In no acute distress. Well-developed. Well-nourished. Right Knee: Positive for tenderness and swelling involving the medial condyle of the distal right femur. There is no effusion or ecchymosis. Full range of motion. Slight limp. Normal bulk, tone, and strength. Sensation intact. The examinationWWW.BSSVE.IN of the other knee is unremarkable. There is no evidence other trauma. Other PE: No other injuries. TREATMENT & COURSE: Knee immobilizer applied. DISPOSITION/PLAN: Discharged in good condition. ASSESSMENT: 1. Sprain of the medial collateral ligament 844.1. ATTENDING NOTE: Discussed with me agree with plan.
4 www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 170 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Bulk of Valuable Data are in Narrative Text
Mr. Blind is a 79-year-old white white male with a history of diabetes mellitus, inferior myocardial infarction, who underwent open repair of his increased diverticulum November 13th at Sephsandpot Center. The patient developed hematemesis November 15th and was intubated for respiratory distress. He was transferred to the Valtawnprinceel Community Memorial Hospital for endoscopy and esophagoscopy on the 16th of November which showed a 2 cm linear tear of the esophagus at 30 to 32 cm. The patient’s hematocrit was stable and he was given no further intervention. The patient attempted a gastrografin swallow on the 21st, but was unable to cooperate with probable aspiration. The patient also had been receiving generous intravenous hydration during the period for which he was NPO for his esophageal tear and intravenous Lasix for a question of pulmonary congestion. On the morningWWW.BSSVE.IN of the 22nd the patient developed tachypnea with a chest X-ray showing a question of congestive heart failure. A medical consult was obtained at the Valtawnprinceel Community Memorial Hospital. The patient was given intravenous Lasix. A arterial blood gases on 100 percent face mask showed an oxygen of 205, CO2 57 and PH 7.3. An electrocardiogram showed ST depressions in V2 through V4 which improved with sublingual and intravenous nitroglycerin. The patient was transferred to the Coronary Care Unit for management of his congestive heart failure , ischemia and probable aspiration pneumonia. www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 171 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Some Typical Tasks
• Information retrieval -- usually, find an article relevant to x • Question answering -- answer specific questions from information represented in text • Learn and generalize -- find and categorize all protein-protein interactions reported in research literature • Case selection -- find patients based on their clinical characteristics; e.g., find asthmatics who don’t smoke • Extract diagnoses, symptoms, tests, results, medications, WWW.BSSVE.INoutcomes, etc., from clinical records • Extract relations among the above: e.g., x was done to rule out y • Find (and suppress) identifying information to make data safe for public release
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 172 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Methods
• grep • Search for specific words, simple patterns • Good for some things: smok.*, • 25 mg Lasix PO QD • \d+ [um]g [-A-Za-z]+ (PO|IV|IM) (QD|BID|TID|Q6H|Q4H) • dictionary + rules • E.g., names of people, towns, streets, hospitals, clinics, wards, companies; Mr. xxx. • WWW.BSSVE.INsupervised training using single word, bigram, etc., features • mostly leads to probabilistic models that recover the most likely interpretation • parsing to recover syntactic structure of sentences • semantic interpretation in terms of medical vocabularies, taxonomies • discourse analysis for resolution of pronouns, anaphora www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 173 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Example: Simple text matching
• UMLS contains >1M medically meaningful phrases • vocabularies from ~150 sources • e.g.,“heart attack”,“myocardial infarction”,“acute MI”, etc. • synonym, antonym, generalization, specialization, co- occurrence links • 189 semantic types in taxonomy of entities and relations WWW.BSSVE.IN• normalizer, all terms indexed by their normalized versions • Search each of n2 substrings for match in UMLS; then search for best cover by resulting matches
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 174 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Example: Tawanda Sibanda’s MEng thesis, 2006 http://groups.csail.mit.edu/medg/ftp/tawanda/THESIS.pdf
• Tasks: • De-identification: find all of • Patients’ and doctors’ first & last names • Id numbers • Phone, fax, pager numbers • Hospital names • Geographic locations • Dates WWW.BSSVE.IN• Try to resolve ambiguity: • E.g., “Mr. Huntington, who has Huntington’s Disease” • Extract semantic categories • Extract semantic relations
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 175 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Classifier for De-Id
• Features: – Target word to be classified – Words up to 2 words left/right of target – Words up to 2 syntactic links left/right of target (using Link Parser, vide infra) – Target part of speech – Target capitalization – Target length – MeSH ID of noun phrase containing the target – PresencesWWW.BSSVE.IN of target ± 1 word in name, location, hospital and month dictionaries – Heading of document section where target appears – Whether “-” or “/” characters are in target • Support Vector Machine (linear kernel)
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 176 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
“Secret Sauce”: Syntax
• Link Grammar Parser –Lexical database of constraint formulas for each word (many inherit by category) –Hundreds of feature pairs; e.g., “plural”
“John lives with his brother.” WWW.BSSVE.IN+------Xp------+ | +----Js----+ | +---Wd--+--Ss-+--MVp-+ +--Ds--+ | | | | | | | | LEFT-WALL John lives.v with his brother.n .
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 177 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Evaluation
• Precision = # instances of x correctly classified/total # classified as x (=PPV) • Recall = # instances of x correctly classified/total # of x in data (=sensitivity) • F-measure = harmonic average(precision, (1 + β2) P R recall):WWW.BSSVE.IN F = × × (β2 P )+R –Asymmetry can be× modeled by changing β
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 178 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Test on Four Corpora
1. Re-identified with randomly selected dictionary names and numbers, retaining original formats; e.g., “Szolovits, Peter” ==> “Smith, John” 2. Ambiguous: all names selected from disease, treatment & test dictionaries 3. Non-dictionary: synthesized names; e.g., “O. YmfgiWWW.BSSVE.IN was admitted …” 4. Authentic: genuine PHI
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 179 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
PHI in four corpora
Category Re-identified Ambiguous Non- Authentic Dictionary Non-PHI 17,874 19,275 17,875 112,669
Patient 1,048 1,047 1,037 294
Doctor 311 311 302 738
Location 24 24 24 88
HospitalWWW.BSSVE.IN 600 600 404 656
Date 735 736 735 1,953
ID 36 36 36 482
Phone 39 39 39 32
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 180 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in De-Id Results Authentic Corpus
Method Class Precision Recall F-measure Stat De-ID PHI 98.46% 95.24% 96.82% iFinder PHI 26.17% 61.98% 36.80% H + D PHI 82.67% 87.30% 84.92% Stat Non-PHI 99.84% 99.95% 99.90% iFinder Non-PHI 98.68% 94.19% 96.38% H+D Non-PHI 99.58% 99.39% 99.48% WWW.BSSVE.IN Method Class Precision Recall F-measure Stat De-ID PHI 98.40% 93.75% 96.02% SNoW PHI 96.36% 91.03% 93.62% Stat De-ID Non-PHI 99.90% 99.98% 99.94% SNoW Non-PHI 99.86% 99.95% 99.90%
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 181 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Most important features (considered independently) • Target word • Syntactic bigrams • Lexical bigrams • POS • Dictionary WWW.BSSVE.IN • MeSH • Orthography (punctuation)
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 182 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
i2b2 Workshop on Challenges in NLP for Clinical Data, 2006
Ozlem Uzuner, Peter Szolovits, and Isaac Kohane SUNY Albany, MIT, and Partners Healthcare WWW.BSSVE.IN
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 183 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Challenge Questions
• Automatic de-identification of clinical data (de-identification challenge)
• Automatic evaluation of smoking status of patients based on medical records (smokingWWW.BSSVE.IN challenge)
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 184 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Data
• ~1000 medical discharge summaries from Partners HealthCare • Scrubbed semi-automatically –One system pass –Three manual passes • TrainWWW.BSSVE.IN and test sets representing similar distributions of relevant classes
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 185 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in De-identification Challenge Data • Focus on the PHI present in discharge summaries – Patient: first and last names of patients, their health proxies, and family members. Exclude titles. – Doctors: medical doctors and other practitioners; for transcribed records, the transcribers, and their initials. Excludes titles, such as Dr. and MD. – Hospitals: hospital names, names of nursing homes where patients are treated and may also reside, room numbers of patients, and buildings and floors related to doctors’ affiliations. Some hospitals, morgues, or nursing homes are described with their street address. These are included in the hospital category. – IDs: Any combination of numbers and letters identifying medical records, patients, doctors, or hospitals. All reports start with an id number. WWW.BSSVE.IN– Dates: excludes years. – Location: Geographic locations such as cities, states, street names, zip codes, and building names and numbers. The professional affiliations of patients and their families are also considered locations. – Phone numbers: Telephone, pager, and fax numbers. – Ages: Ages over 90. – None: none of the above.
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 186 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in De-identification Challenge Data • PHI categories marked while scrubbing – Five annotators – Agreement 100% • Realistic surrogates substituted for each PHI type – Surrogates obtained by permuting the letters of existing names obtained from the US Census bureau • Followed the format of the authentic PHI • Most surrogate names are out-of-vocabulary • Can generate authentic-looking names which are kept WWW.BSSVE.IN• Dates in the same record are all offset by the same number – Ambiguity injected into PHI to make the task more challenging • Ambiguous PHI lexically overlap with medical terms such as diseases, treatments, and medical tests
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 187 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
De-identification Evaluation
• Metrics –Precision, recall, and f-measure (B=1) at token level –Micro- and macro-averaged metrics for system-level performance • Reports on –Overall performance 9-way and 2-way (PHI vs. non- PHI) –PerformanceWWW.BSSVE.IN on ambiguous PHI –Performance on out-of-vocabulary PHI
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 188 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
De-id (9-way) F-measure
System ID Micro-averaged F- System ID Macro-averaged measure F-measure Wellner,3, 0.997434578 Mitre Wellner,3 0.941419964 Szarvas,2, 0.997413881 Szarvas,1 0.940304335 Szeged Aramaki,1, U 0.996031341 Hara,3 0.922680373 Tokyo WWW.BSSVE.IN Aramaki,1 0.91541839 Hara,3, Nara 0.993694909 Remaining 0.5974-0.8940 Systems Remaining 0.9767-0.9931 Systems * Systems are identified by the last name of the first author and the submission number www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 189 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
De-id (2-way)
System ID Macro-averaged Micro-averaged F- F-measure measure Wellner,3, 0.989693751 0.997774522 Mitre Szarvas,2, 0.989634637 0.997767856 Szeged Aramaki,1, 0.983954094 0.996559061 UWWW.BSSVE.IN Tokyo Hara,3, 0.972942838 0.99417348 Nara Remaining 0.9518-0.9714 0.9786-0.9938 Systems
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 190 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
De-identification Systems
• Ranking remains almost the same on ambiguous and out-of-vocabulary PHI
WWW.BSSVE.IN
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 191 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
General Patterns • Diverse set of approaches – Systems varied in their use of rules and machine learning – Systems varied in the features they used for identifying PHI • Interesting ideas from one or more systems – Many made use of rules to recognize PHI with unique format – Some systems were adapted from other Natural Language Processing tasks to de-identification • Named Entity Recognition systems are easily adapted to this task (though dictionary dependencies cause problems) • Text segmentation (parts of a text), WWW.BSSVE.IN• Sentence classification, • clause chunking – Constraints--every mention of a phrase interpreted similarly
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 192 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
General Patterns • General observations – Clinical records vary from data traditionally used in Natural Language Processing – Despite the difference in the nature of data, systems used for well- studied NLP problems were successfully adapted to de- identification of clinical records – Many systems made use of structure of the documents, e.g., headers and footers • Szarvas et al. • Aramaki et al. • Guillen et al. WWW.BSSVE.IN– Regular expressions for the structured PHI – On this data, surface features and context help de-identification – Ambiguities and absence of names from dictionaries make this data more challenging than real data • Even on this deliberately more challenging data, performance of systems is impressive
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 193 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Quo Vadis?
• Anecdote: –Shawn was admitted to Brigham and Women’s on March 3, 2006. –Shawn was admitted to BWH on March 3, 2006. –Shawn was admitted to Mass General on March 3, 2006. –Mr. Smith was admitted to Massachusetts General HospitalWWW.BSSVE.IN on March 3, 2006. –Instance of overtraining • Much more data should help –But annotation is very costly
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 194 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Extracting Assertions
• Semantic Category Recognition: identify semantic category of each word in a discharge summary – Diseases – Treatments – Abusive (sic) substances – Dosages – Practitioners – Diagnostic tests – Results – Signs and symptoms – “none”WWW.BSSVE.IN • Assertion classification: – Patient definitely has this – Someone other than the patient has this – Patient may have this – Patient does not have this
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 195 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Semantic Category Recognizer • 8-way + none Support Vector Machine (linear) classifier • Features: – Target – Left/right lexical bigrams – Section heading – Left/right syntactic bigrams – Head of noun phrase + syntactic bigrams of head – PartsWWW.BSSVE.IN of Speech of target and words ± 2 left/right – UMLS semantic type of noun phrase containing target – Capitalized? – Contains numerals? – Contains punctuation?
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 196 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Comparison
Baseline UMLS lookup Statistical Classifier Class Precision Recall F-Measure Precision Recall F-Measure None 0.828 0.883 0.855 0.938 0.962 0.950 Disease 0.656 0.707 0.680 0.911 0.899 0.905 Treatment 0.548 0.726 0.625 0.924 0.901 0.912 Test 0.764 0.560 0.646 0.931 0.913 0.922 Result 0.404 0.358 0.380 0.857 0.809 0.832 Dosage 0.901 0.597 0.718 0.966 0.941 0.954 Symptom 0.653 0.334 0.442 0.901 0.815 0.856 Practitioner 0.486 0.733 0.584 0.978 0.934 0.956 Substance WWW.BSSVE.IN0.685 0.128 0.215 0.934 0.853 0.892
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 197 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Assertion Classifier
• Rule-based, using regular expressions on common phrases that precede or succeed a problem (± 4 words): –“Alter-association” phrases: imply that the problem is someone else’s –Negation phrases –Uncertainty phrases • GreedyWWW.BSSVE.IN algorithm, in above order • If none of the above match, then assert as present.
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 198 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Assertion Classification
Class Precision Recall F-Measure Present 0.929 0.967 0.947 Absent 0.947 0.900 0.923 Uncertain 0.723 0.556 0.629 Alter-Association 1.000 0.810 0.895 WWW.BSSVE.IN
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 199 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Semantic Relation Recognition
• Relations of interest: –Symptom <==> treatment –Uncertain symptom <==> treatment –Disease <==> test –Uncertain disease <==> test –Disease <==> treatment –UncertainWWW.BSSVE.IN disease <==> treatment • Mode of relation –Test reveals disease –Test conducted to investigate disease –none
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 200 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Semantic Relations
• For each relation, T. S. developed a k-way SVM classifier to get the mode. • E.g., disease <==> test features – # words between concepts – Whether disease precedes test – Whether other concepts occur in between – Verbs between disease and test – Two verbs before/after disease and test – Head words of disease and test phrases – Right/leftWWW.BSSVE.IN lexical bigrams of disease and test – Right/left syntactic bigrams of disease and test – Words between disease and test – Path of syntactic links between disease and test – Path of syntactically connected words between disease and test
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 201 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
MIT OpenCourseWare http://ocw.mit.edu
HST.950J / 6.872 Biomedical Computing Fall 2010
For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.
WWW.BSSVE.IN
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 202 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Introduction to Modeling 6.872/HST950
WWW.BSSVE.IN
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 203 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Why build Models?
• To predict (identify) something • Diagnosis • Best therapy • Prognosis • Cost • WWW.BSSVE.INTo understand something • Structure of model may correspond to structure of reality
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 204 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Where do models come from?
• Pure induction from data • Even so, need some “space” of models to explore • Maximum A-posteriori Probability (MAP) • Maximum Likelihood (ML) • Assumes uniform priors over all hypotheses in the space • WWW.BSSVE.INA-priori knowledge, expressed in • Structure of the space of models • • Adjustments to observed data
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 205 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in An Example (Russell & Norvig)
• Surprise Candy Corp. makes two flavors of candy: cherry and lime • Both flavors come in the same opaque wrapper • Candy is sold in large bags, which have one of the following distributions of flavors, but are visually indistinguishable: • h1: 100% cherry • h2: 75% cherry, 25% lime • h3: 50% cherry, 50% lime h : 25% cherry, 75% lime WWW.BSSVE.IN• 4 • h5: 100% lime • Relative prevalence of these types of bags is (.1, .2, .4, .2, .1) • As we eat our way through a bag of candy, predict the flavor of the next piece; actually a probability distribution.
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 206 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Bayesian Learning
• Calculate the probability of each hypothesis given the data • To predict the probability distribution over an unknown quantity, X, • If the observations d are independent, then • E.g., suppose the first 10 candies we taste are all lime WWW.BSSVE.IN
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 207 h1www.onlineeducation.bharatsevaksamaj.net: 100% cherry www.bssskillmission.in h2: 75% cherry, 25% lime Learning Hypotheses h3: 50% cherry, 50% lime h4: 25% cherry, 75% lime and Predicting from Them h5: 100% lime
• (a) probabilities of hi after k lime candies; (b) prob. of next lime
a b 1 1
0.9 0.8 0.8 0.6 0.7 0.4 0.6
0.2 0.5 Probability that next candy is lime Posterior probability of hypothesis 0 0.4 0 2 4 6 8 10 0 2 4 6 8 10 WWW.BSSVE.INNumber of samples in d Number of samples in d
P(h1 | d) P(h2 | d) P(h3 | d) P(h4 | d) P(h5 | d)
Image by MIT OpenCourseWare. • MAP prediction: predict just from most probable hypothesis • After 3 limes, h5 is most probable, hence we predict lime Even though, by (b), it’s only 80% probable • www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 208 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Observations
• Bayesian approach asks for prior probabilities on hypotheses! • Natural way to encode bias against complex hypotheses: make their prior probability very low • Choosing hMAP to maximize • is equivalent to minimizing • but as we know that entropy is a measure of information, these two terms are • # of bits needed to describe the data given hypothesis • # bits needed to specify the hypothesis WWW.BSSVE.IN• Thus, MAP learning chooses the hypothesis that maximizes compression of the data; Minimum Description Length principle • Regularization is similar to 2nd term—penalty for complexity • Assuming uniform priors on hypotheses makes MAP yield hML, the maximum likelihood hypothesis, which maximizes
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 209 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Learning More Complex Hypotheses
• Input: • Set of cases, each of which includes • numerous features: categorical labels, ordinals, continuous • these correspond to the independent variables • Output: • For each case, a result, prediction, classification, etc., corresponding to the dependent variable • In regression problems, a continuous output WWW.BSSVE.IN• a designated feature the model tries to predict • In classification problems, a discrete output • the category to which the case is assigned • Task: learn function f(input)=output • that minimizes some measure of error
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 210 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Linear Regression
• General form of the function • For each case: • Find to minimize some function of over all • e.g., mean squared error:
WWW.BSSVE.IN
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 211 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Logistic Regression
• Logistic function:
• E.g, how risk factors contribute to probability of death • are the log odds ratios
WWW.BSSVE.IN
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 212 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
More sophisticated models
• Nearest Neighbor Methods • Classification Trees • Artificial Neural Nets • Support Vector Machines • Bayes Networks (much on this, later) • WWW.BSSVE.INRough Sets, Fuzzy Sets, etc. (see 6.873/HST951 or other ML classes)
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 213 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
How?
• Given: pile of training data, all cases labeled with gold standard outcome • Learn “best” model • Gather new test data, also all labeled with outcomes • Test performance of model on new test data WWW.BSSVE.IN • Simple, no?
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 214 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Simplest Example
Test Test Positive Negative
Disease True False TP+FN • Relationship Present Positive Negative between a diagnostic Disease False True FP+TN conclusion and Absent Positive Negative a WWW.BSSVE.INdiagnostic test TP+FP FN+TN
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 215 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Test Test Positive Negative Definitions Disease True False TP+FN Present Positive Negative Disease False True FP+TN Absent Positive Negative
TP+FP FN+TN
Sensitivity (true positive rate): TP/(TP+FN) False negative rate: 1-Sensitivity = FN/(TP+FN) Specificity (true negative rate): TN/(FP+TN) WWW.BSSVE.IN False positive rate: 1-Specificity = FP/(FP+TN) Positive Predictive Value (PPV): TP/(TP+FP) Negative Predictive Value (NPV): TN/(FN+TN)
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 216 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Test Thresholds
-
+
WWW.BSSVE.IN FN FP
T
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 217 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Wonderful Test
-
+
WWW.BSSVE.IN FN FP
T
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 218 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Test Thresholds Change Trade-off between Sensitivity and Specificity
-
+
WWW.BSSVE.IN FN FP
T
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 219 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Receiver Operator Characteristic TPR (sensitivity) (ROC) Curve 1 T
WWW.BSSVE.IN
0 0 1 www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.inFPR (1-specificity) 220 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
TPR What makes a better test? (sensitivity) superb 1
OK
worthless WWW.BSSVE.IN
0 0 1 www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.inFPR (1-specificity) 221 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Need to explore many models
• Remember: • training set => model • model + test set => measure of performance • But • How do we choose the best family of models? • How do we choose the important features? • Models may have structural parameters WWW.BSSVE.IN• Number of hidden units in ANN • Max number of parents in Bayes Net • Parameters (like the betas in LR), and meta-parameters • Not legitimate to “try all” and report the best !!!!!!!!!!!!!!!!!!
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 222 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
The Lady Tasting Tea
• R.A. Fisher & the Lady • B. Muriel Bristol claimed she prefers tea added to milk rather than milk added to tea • Fisher was skeptical that she could distinguish • Possible resolutions • Reason about the chemistry of tea and milk • Milk first: a little tea interacts with a lot of milk • Tea first: vice versa • Perform a “clinical trial” • Ask her to determine order for a series of test cups WWW.BSSVE.IN• Calculate probability that her answers could have occurred by chance guessing; if small, she “wins” • ... Fisher’s Exact Test • Significance testing • Reject the null hypothesis (that it happened by chance) if its probability is < 0.1, 0.05, 0.01, 0.001, ..., 0.000001, ..., ????
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 223 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
How to deal with multiple testing
• Suppose Ms. Bristol had tried this test 100 times, and passed once.Would you be convinced of her ability to distinguish? • Bonferroni correction: for n trials, insist on a p-value that is 1/n of what you would demand for a single trial WWW.BSSVE.IN
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 224 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Cross-validation Training Data “Real” Training Data
Validation Data
Test Data
• WWW.BSSVE.INAny number of times • Train on some subset of the training data • Test on the remainder, called the validation set • Choose best meta-parameters • Train, with those meta-parameters, on all training data • Test on Test data, once! www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 225 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Aliferis lessons (part)
• Overfitting • bias, variance, noise • O = optimal possible model over all possible learners • L = best model learnable by this learner • A = actual model learned • Bias = O - L (limitation of learning method or target model) • Variance = L - A (error due to sampling of training cases) WWW.BSSVE.IN• Compare against learning from randomly permuted data • Curse of dimensionality • Feature selection • Dimensionality reduction
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 226 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Causality
• Suppes, 1950’s • Statistical association • Temporal succession No confounders (!) • U1 Un • hidden variables A node, X, is conditionally independent • X WWW.BSSVE.INof all other nodes in the network given Z1 Zn its Markov blanket: its parents, Ui, children,Yi, and children’s parents, Zi.
Y1 Yn
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 227 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Using MIMIC data to build predictive models
• Caleb Hug’s 2009 PhD thesis: • Mortality http://dspace.mit.edu/handle/1721.1/46690 • Comparison to SAPS II • Daily Acuity Scores • Real-time Acuity Scores • Other outcomes • Good • Weaning from Ventilator • Weaning from Intra-Aortic Balloon Pump • Weaning from Vasopressors • Bad WWW.BSSVE.IN • Septic shock • Hypotension • Acute kidney injury
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 228 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
3.2. EXTRACTING THE VARIABLES 31
Cleaning the data—half the researchSystolic Blood Pressuretime Glasgow Coma Scale Frequency Frequency 0e+00 8e+05 0e+00 2e+05 • Missing values 0 1 2 3 4 0 5 10 15 • Some values are not measured for someHours clinical Between SBP Measurements situations Hours Between GCS Observations • Failures in data capture process Hematocrit Arterial Base Excess
• Episodically measured variables 10000 4000 Frequency Frequency 0 0
• Unclear/undefined clinical states 0 10 20 30 40 0 5 10 15 20 25 30 • Imprecise timing of meds, ... Hours Between HCT Measurements Hours Between Art_BE Measurements • Partially measured i/o Blood Urea Nitrogen INR • Proxies: e.g., which ICU⇒what disease Frequency Frequency 0 10000 0 4000 8000 • Derived variables: integrals, slopes, ranges,0 frequencies, 5 10 15 20 25 30 etc. 0 5 10 15 20 25 30 Hours Between BUN Measurements Hours Between INR Observations • TransformedWWW.BSSVE.IN variables: square root, log, etc. WBC Counts FiO2 Settings • Select subset of data with enough data! Frequency Frequency 0 10000 0 60000
0 5 10 15 20 25 30 0 5 10 15 20 25 30
Hours Between WBC Counts Hours Between FiO2Set Recordings Figure by Hug, Caleb Wayne. "Detecting Hazardous Intensive Care Patient Episodes UsingFigure Real-time 3-1: Observation Mortality Models." frequency Massachusetts histograms Institute of Technology, 2009.
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 229 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Descriptive look 3.4. PRELIMINARY DATASET 453.4. PRELIMINARY DATASET 47
Patient Length of Stay Patient Age Histogram of Blood Urea Nitrogen (BUN) Histogram of White Blood Cell Counts
n 13923 n 13923 n 2448468 n 2448468 missing 0 missing 78 missing 186042 missing 224258 mean 5.02 mean 63.5 mean 30.8 mean 12.9 median 2.67 median 65 median 23 median 11.7 std dev 8.1 std dev 17 std dev 24.3 std dev 6.5 Frequency Frequency Frequency Frequency 0 50000 150000 0 50000 150000 0 1000 2000 3000
0 10 20 30 40 50 20 40 60 80 100 0 50 100 150 0 10 20 30 40 50
Length of Stay (days) Age (Years) BUN (mg/dL) WBC (1000/mm^3)
Patient Admit Weight Sex of Patients Histogram of Potassium Histogram of Sodium
n 13923 n 13923 n 2448468 n 2448468 missing 1085 missing 56 missing 105388 missing 139134 mean 81 mean 4.07 mean 139 median 78.5 median 4 median 139 std dev 21.8 std dev 0.534 std dev 4.76 Frequency Frequency Frequency Frequency 0 50000 150000 0 50000 150000 0 500 1000 1500
50 100 150 200 250 Female Male 2 3 4 5 6 7 8 120 130 140 150 160 WeightWWW.BSSVE.IN (kg) Sex Potassium (mEq/L) Sodium (mEq/L)
Service types for patients ICD9 Chronic Illnesses Histogram of Bicarbonate (CO2) Histogram of Bilirubin
n 13923 n 13923 n 2448468 n 2448468 missing 0 missing 0 missing 204650 missing 1842053 mean 24.6 mean 4.01 median 24 median 1.1 std dev 4.72 std dev 7.46 Frequency Frequency Frequency Frequency 50000 150000 0 0 20000 60000 0 1000 2000 3000 0 4000 8000 12000 0 2000 6000 10000 0 200 400 600
Other NSICU MSICU CSRU No Yes 10 20 30 40 50 0 5 10 15 20 25 30 ICD9 Chronic Illness? Service Type (metastatic carcinoma, hematologic malignancy, AIDS) Bicarbonate (mEq/L) Bilirubin (mg/dL)
Figure by Hug, Caleb Wayne. "Detecting Hazardous Intensive Care Patient www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in Figure 3-2: Histograms for demographic information Episodes UsingFigure Real-time 3-4: SAPS Mortality II Variables Models." (cont) Massachusetts Institute of Technology, 2009. 3.4. PRELIMINARY DATASET 49 a given patient. Some patients, for example, are recorded as having died over one year after their first ICU discharge. To limit such cases, one might define death as “died within the ICU or within 30 days of discharge”. This limit marks a patient as alive if he or she is still in the hospital 30 days after ICU discharge. If the patient is not in the hospital at this point, the hospital discharge status of the patient is used to indicate mortality: if the patient was discharged alive (censored) they are marked as survived; otherwise they are marked as expired. Figure 3-6 illustrates the change in mortality rate as patients stay in the ICU for longer periods of time. This figure includes both hospital mortality (i.e., the patient died at any point during any recorded visit) and the within-30-days-of-ICU-discharge mortality. As the figure shows, the two mortality rates track each other closely for the first several days. However, it is clear that patients who stay in the ICU longer are more apt to remain in the hospital for a significant period of time before dying. For the remainder of this work, references to “mortality” indicate death in the ICU or within the following 30 3.5. FINAL DATASET 51 230 days.www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Table 3.14: Final Dataset: Partial Patient Exclusions Drop Rule Number of Rows Outcomes In the ICU for longer than seven days 728739 Received limited treatment, including 198942 CMO (“comfort measures only”) DNR (“do not resuscitate”) DNI (“do not intubate”) “no CPR” or “other code” Total Number of Patients Received hemodialysis or hemofiltrationHospital Mortality 139561 14000 Number of Hospital Mortalities 14000 Number of ICU + 30 days Mortalities 25 The motivation for these rules generally follows the reasoning for excluding entire patients. For example, as Figure 3-6 indicates by plotting the mortality rate versus 10000 10000
the number of days20 spent in the ICU, most patients leaveICU + the30 days ICU Mortality within seven days of admission. For patients that do not leave in this 7-day window, the 30-day mortality rate starts to noticeably decrease as caregivers are able to successfully prolong the 6000 6000 patient’s life while15 the patient remains in a compromised state often dependent on
various interventions.Mortality Rate (%) Number of patients
10 Number of Patients 2000
2000 3.5.2 Dataset Summary 0 0 The final dataset — after applying all of the exclusions mentioned above — is sum- marized in Table 3.15. In addition, Appendix B lists the 438 individual variables 0 5 10 15 20 25 30 0 5 10 15 20 25 30 WWW.BSSVE.INwith brief summary statistics. Figure 3-7 provides an updated version of Figure 3-6 Days in ICU for the final dataset. Days in ICU
Table 3.15: Preprocessed Data Figure 3-6: Patient counts versus the number of daysNumber spent of Patients in the10,066 ICU (left) and mortality rate versus the number of days spent in theNumber ICU of Rows (right). 1,044,982 For each patient, Number of Features 438 only the first ICU stay of the first recorded hospital visit is considered. “ICU + 30 day mortality” excludes deaths that occur afterFigure long by Hug, post-ICU Caleb Wayne. "Detecting discharge Hazardous hospitalizations.Intensive Care Patient Episodes Using Real-time Mortality Models." Massachusetts Institute of Technology, 2009. If a patient leaves the hospitalwww.bsscommunitycollege.in alive withinwww.bssnewgeneration.in this 30-day www.bsslifeskillscollege.in period, they are assumed to have survived. 4.3. OTHER SEVERITY OF ILLNESS SCORES 63
231 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
SAPS II
Table 4.1: SAPS II Variables Variable Max Points Age 18 Heart rate 11 Systolic BP 13 Body temperature 3 PaO2:FiO2 (if ventilated or continuous 11 positive airway pressure) Urinary output 11 Serum urea nitrogen level 10 WBC count 12 Serum potassium 3 Serum sodium level 5 WWW.BSSVE.INSerum bicarbonate level 6 Bilirubin level 9 Glasgow Coma Scorea 26 Chronic diseases 17 Type of admission 8
aIf the patient is sedated, the estimated GCS prior to sedation Figure by Hug, Caleb Wayne. "Detecting Hazardous Intensive Care Patient Episodes Using Real-time Mortality Models." Massachusetts Institute of Technology, 2009. www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 68 CHAPTER 5. MORTALITY MODELS
Fold 1 Fold 2 ROC Area ROC Area GCS_max_sq GCS_max_sq 4/5 Train, n=18307 4/5 Train, n=18499 1/5 Val, n=4581 1/5 Val, n=4389 5.2. DAILY ACUITY SCORES0.70 0.80 0.90 0.70 0.80 0.90 69
0 5 10 15 20 25 30 35 0 5 10 15 20 25 30 probability of mortality (positive correlation)Number of covariates while a negative coefficientNumber means of covariates less risk of mortality (negative correlation). Model 5.1 includes a number of interestingFold 3 covariates. By examining the WaldFold 4 Z scores, however, it appears that the model can likely be improved.232 For example, the www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in contributions from the largest length of stay fluid balance (LOSBal max, Z =3.94) and the largest 24 hour fluid balance (Bal24 max, Z = −3.68) may largely counteract each other and the model may benefit from removing at least one of these variables ROC Area ROC Area and possibly replacing it with an input variable (a mean hourly outputGCS_max_sq for the day, 4/5 Train, n=18220 4/5 Train, n=18227 Training models—5-fold cross validation mechVent_mean_sq 1/5 Val, n=4668 1/5 Val, n=4661 OutputB 60 mean sqrt, is0.70 already 0.80 0.90 included). Similarly, the meaningfulness0.70 0.80 0.90 of the pressD01 sd sq variable (that is, the squared standard deviation of the points marked 68 CHAPTER 5. MORTALITY MODELS 0 5 10 15 20 25 30 35 0 5 10 15 20 25 30 1 following the first pressor infusion and marked 0 before any pressors) is questionable Number of covariates Number of covariates as it decreases risk for patients who have a pressor started in the middle of their first day, but increases risk for patients who receive pressors early or late on their first day. Fold 1 Fold 2 Fold 5 Examining all five folds from Figure 5-1, it is clear that there was negligible impact on the validation performance by reducing the number of covariates to about 25. Considering this, I took the top 25 covariates from the models for cross-validation folds 1, 2, 3, and 5 (excluding fold 4) and created a model using these covariates.
ROC Area ROC Area By performing backward elimination one last time on this model a final model was ROC Area GCS_max_sq GCS_max_sq 4/5 Train, n=18307 selected.4/5 Train, As n=18499 done with each cross-validation fold in Figure 5-1, a plot of performance orientation_max_i 4/5 Train, n=18299 1/5 Val, n=4581 1/5 Val, n=4389 0.70 0.80 0.90 0.70 0.80 0.90 1/5 Val, n=4589 versus the number of covariates0.70 0.80 in 0.90 a given model was created. This plot is shown in
0 5 10 15 20 25 30 35 0 5 10 Figure 15 20 5-2. 25 30 0 5 10 15 20 25 30 35
Number of covariates Number of covariates Number of covariates
SDAS Model Sensitivity Fold 3 Fold 4 Figure 5-1: SDAS model selection. Sensitivity to number of covariates on each cross- validation fold. The covariate(s) from the simplest model are marked on the training curves. WWW.BSSVE.IN 0.80 0.90 0.80 0.90 ROC Area ROC Area GCS_max_sq 0.80 0.90 4/5 Train, n=18220 4/5 Train, n=18227 mechVent_mean_sq 1/5 Val, n=4668 1/5 Val, n=4661 ROC Area 0.70 0.70 GCS_max_sq 0 5 10 15 20 25 30 35 0 5 10 15 20 25 30 Training Data: n=22888 Number of covariates Number of covariates 0.70
0 10 20 30 40 Fold 5 Figure by Hug, Caleb Wayne. "Detecting Hazardous Intensive Care Patient Number of Covariates Episodes Using Real-time Mortality Models." Massachusetts Institute of Technology, 2009.
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in Figure 5-2: SDAS model selection (all development data) ROC Area The models in Figure 5-2 indicate that most of the performance was captured orientation_max_i 4/5 Train, n=18299 1/5 Val, n=4589 0.70 0.80 0.90 with about 35 inputs. This model was chosen for additional refinement. Upon exam-
0 5 10 15 20 25 30 35 ination, it was found to contain a number of pressor-related
Number of covariates
Figure 5-1: SDAS model selection. Sensitivity to number of covariates on each cross- validation fold. The covariate(s) from the simplest model are marked on the training curves. 233 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
70 CHAPTER 5. MORTALITY MODELS
Many univariate analyses 72 CHAPTER 5. MORTALITY MODELS
Model 5.1 SDAS Model for Fold 2 with 30 Covariates Model 5.2 Final SDAS model Obs Max Deriv Model L.R. d.f. P C Dxy Obs Max Deriv Model L.R. d.f. P C Dxy 20172 1e-09 5415.11 30 0 0.893 0.785 20130 3e-10 5619.28 35 0 0.898 0.797 Gamma Tau-a R2 Brier Gamma Tau-a R2 Brier 0.798 0.177 0.456 0.074 0.787 0.176 0.439 0.076 Coef S.E. Wald Z P Coef S.E. Wald Z P GCS_max_sq -0.0064668 5.032e-04 -12.85 0.0000 INR_mean_i -1.795e+00 1.423e-01 -12.61 0.0000 INR_mean_i -1.8734049 1.458e-01 -12.85 0.0000 GCS_max_sq -7.485e-03 6.000e-04 -12.47 0.0000 pacemkr_max -0.9337190 1.179e-01 -7.92 0.0000 svCSRU_max -0.9137522 1.250e-01 -7.31 0.0000 OutputB_60_mean_sqrt -6.561e-02 6.885e-03 -9.53 0.0000 RikerSAS_mean -0.3430971 5.151e-02 -6.66 0.0000 pacemkr_max -1.084e+00 1.183e-01 -9.16 0.0000 Platelets_Slope_1680_min -5.8856843 8.839e-01 -6.66 0.0000 svCSRU_max -9.516e-01 1.208e-01 -7.88 0.0000 urineByHr_mean_sqrt -0.0584113 9.453e-03 -6.18 0.0000 GCSrdv_mean -1.138e-01 1.528e-02 -7.45 0.0000 GCSrdv_mean -0.0902717 1.552e-02 -5.82 0.0000 pressD01_mean_am -2.774e+00 3.893e-01 -7.13 0.0000 GCSrng_min_am -0.0812232 1.459e-02 -5.57 0.0000
Platelets_Slope_1680_min -5.493e+00 8.615e-01 -6.38 0.0000 pressD01_mean_am -1.6132643 3.005e-01 -5.37 0.0000 CV_HRrng_max -0.0061979 1.216e-03 -5.10 0.0000 pressD01_sd_sq -5.085e+00 8.678e-01 -5.86 0.0000 Insulin_sd_sq -2.1686950 4.372e-01 -4.96 0.0000 sedatives_mean_sq -4.375e-01 8.455e-02 -5.17 0.0000 alloutput_max_la -0.0890330 2.265e-02 -3.93 0.0001 Bal24_max -4.493e-05 1.222e-05 -3.68 0.0002 MetCarcinoma_min 0.4468763 1.567e-01 2.85 0.0043 CV_HRrng_max -3.267e-03 1.083e-03 -3.02 0.0026 WBC_mean_am 0.0147036 5.149e-03 2.86 0.0043 Intercept 4.292e-01 4.085e-01 1.05 0.2934 AIDS_min 0.5954305 1.991e-01 2.99 0.0028 Intercept 1.5314512 4.529e-01 3.38 0.0007
Milrinone_perKg_min_sq 3.523e+00 1.113e+00 3.17 0.0015 MBPm.pr_min_am 1.4601630 3.518e-01 4.15 0.0000 LOSBal_max 2.247e-05 5.703e-06 3.94 0.0001 HemMalig_min 0.6032027 1.212e-01 4.98 0.0000 hrmVA_max 3.410e-01 6.767e-02 5.04 0.0000 RESP_mean_sq 0.0006615 1.324e-04 5.00 0.0000 MBPm.pr_min_am 1.904e+00 3.711e-01 5.13 0.0000 hrmVA_max 0.3520834 6.823e-02 5.16 0.0000 Mg_min_sq WWW.BSSVE.IN1.067e-01 1.798e-02 5.93 0.0000 PaO2toFiO2_mean 0.2672376 4.336e-02 6.16 0.0000 beta.Blocking_agent_mean_lam 2.418e-01 3.955e-02 6.11 0.0000 Na_mean_am 0.0549066 8.506e-03 6.45 0.0000 Mg_min_sq 0.1173220 1.815e-02 6.46 0.0000 Na_mean_am 5.214e-02 8.415e-03 6.20 0.0000 ShockIdx_max 0.5742182 8.853e-02 6.49 0.0000 mechVent_mean_sq 7.183e-01 1.047e-01 6.86 0.0000 Platelets_mean_i 24.0719462 3.560e+00 6.76 0.0000 RESP_mean_sq 9.226e-04 1.293e-04 7.13 0.0000 hospTime_min_sqrt 0.0057514 8.158e-04 7.05 0.0000 Platelets_mean_i 2.512e+01 3.512e+00 7.15 0.0000 day_min_sq 0.0170075 2.372e-03 7.17 0.0000 Lasix_max_lam 2.550e-01 3.457e-02 7.38 0.0000 jaundiceSkin_mean_la 0.1469141 2.045e-02 7.18 0.0000 CO2_mean_i 19.3845272 2.682e+00 7.23 0.0000 CO2_mean_i 2.038e+01 2.741e+00 7.43 0.0000 Lasix_max_lam 0.2523702 3.444e-02 7.33 0.0000 jaundiceSkin_mean_la 1.523e-01 2.014e-02 7.56 0.0000 beta.Blocking_agent_mean_lam 0.2918077 3.923e-02 7.44 0.0000 hospTime_min_sqrt 6.860e-03 7.939e-04 8.64 0.0000 Sympathomimetic_agent_min 0.8576883 9.254e-02 9.27 0.0000 pressorSum.std_mean_sqrt 7.758e-01 7.225e-02 10.74 0.0000 SpO2.oor30.t_mean_sqrt 0.4059329 4.128e-02 9.83 0.0000 SpO2.oor30.t_mean_sqrt 4.929e-01 4.095e-02 12.04 0.0000 BUNtoCr_min_sqrt 0.2829088 2.348e-02 12.05 0.0000 BUNtoCr_min_sqrt 2.867e-01 2.323e-02 12.34 0.0000 Age_min_sq 0.0002601 1.495e-05 17.40 0.0000 Figure by Hug, Caleb Wayne. "Detecting Hazardous Intensive Care Patient Age_min_sq 2.258e-04www.bsscommunitycollege.in 1.450e-05 15.57 0.0000 www.bssnewgeneration.in www.bsslifeskillscollege.in Episodes Using Real-time Mortality Models." Massachusetts Institute of Technology, 2009. 234 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
5.2. DAILY ACUITY SCORES 73 88 CHAPTER 5. MORTALITY MODELS Evaluating the models
SDAS: All Days SDAS: Day 1 1.0
1.0 0.8
0.8
0.6
0.6
0.4 Sensitivity
0.4 AUC 0.898 Sensitivity Sensitivity
0.2 n 20130 AUC 0.876 AUC 0.87
0.2 Missing 2758 n 8428 n 2434
0.0 Missing 1164 Missing 584 0.0 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 1!Specificity 1!Specificity 1!Specificity
Figure 5-3: SDAS ROC curve (development data). AUC = the area under the curve; n = the total number of available predictions used for curve; Missing = number of 92 CHAPTER 5. MORTALITYSDAS: MODELS Day 2 SDAS: Day 3 missing predictions. WWW.BSSVE.INDevel Validation SDAS
Dxy 0.752 1.0 C (ROC) 0.876 R2 0.385 D 0.223
Table 5.4: SDAS Hosmer-Lemeshow H risk deciles0.8 (allU days) 0.006 Q 0.217 Brier 0.077 Died Survived Intercept !0.257
0.6 Slope 0.820 Decile Prob.Range Prob. Obs. Exp. Obs. Exp.Emax Total 0.094
1 [0.000203,0.00335) 0.002 2 4.2 20110.4 2008.8 2013 Sensitivity Sensitivity 2 [0.003353,0.00682) 0.005 3 9.9Actual Probability 2010 2003.1 2013 Ideal 3 [0.006825,0.01281) 0.010 11 19.2 20020.2 1993.8 2013 Logistic calibration Nonparametric AUC 0.877 AUC 0.883 4 [0.012812,0.02277) 0.017 24 34.8 1989 1978.2 2013 Grouped observations n 2108 n 1389 0.0 5 [0.022771,0.03971) 0.031 53 61.5 1960 1951.5 2013 Missing 216 Missing 137 6 [0.039706,0.06691) 0.052 104 104.8 19090.0 1908.2 20130.2 0.4 0.6 0.8 1.0 Predicted Probability 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 7 [0.066911,0.11297) 0.088 198 176.7 1815 1836.3 2013 SDAS Day 1 Figure by Hug, Caleb Wayne. "Detecting Hazardous Intensive Care Patient 8 [0.112972,0.20128) 0.152 324 305.3 1689www.bsscommunitycollege.in 1707.7 2013 www.bssnewgeneration.in0.0 www.bsslifeskillscollege.in 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 Dxy 0.741 Episodes Using Real-time Mortality Models." Massachusetts Institute of Technology, 2009. 9 [0.201280,0.40232) 0.285 610 574.7 1403 1438.3C (ROC) 2013 0.871 R2 0.374 10 [0.402321,0.99876] 0.634 1239 1276.9 774 736.1D 2013 0.209 U 0.003 1!Specificity 1!Specificity Q 0.205 Brier 0.074 2 Intercept !0.079 χ = 24.47, d.f. = 8; p =0.002 Slope 0.866 Emax 0.046
Actual Probability SDAS: Day 4 SDAS: Day 5 Ideal Logistic calibration Nonparametric Grouped observations 0.0 0.2 0.4 0.6 0.8 1.0
0.0 0.2 0.4 0.6 0.8 1.0 Predicted Probability DAS1
Dxy 0.755 C (ROC) 0.877 R2 0.376 D 0.205 U 0.005 Q 0.200 Brier 0.072 Intercept !0.132
Slope 0.829 Sensitivity Sensitivity Emax 0.066 AUC 0.882 AUC 0.863
Actual Probability n 928 n 662 Ideal Logistic calibration Missing 85 Missing 59 Nonparametric 0.0Grouped 0.2 observations 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
0.0 0.2 0.4 0.60.0 0.8 0.2 0.4 1.0 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 Predicted Probability
Figure 5-9: Calibration plots for SDAS, SDAS day 1, and DAS1 (validation data).1!Specificity The 1!Specificity relative frequencies for each predicted probability are indicated by the bars along the x-axis. Figure 5-7: SDAS ROC curves (validation data) 235 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Selected features for each day of ICU stay
w
s
a
F 82
10
c
r
ig
it
o
e
r
h
ur
r
e
a
s
“
e
n
+
f k
o BUNtoCr_mean_sqrt
5
”
e r
- Age_min_sq
,
d 5
e + Amiodarone_min_am !
a :
a Age_min_sq SpO2.oor30.t_mean_sqrt (
n + beta.Blocking_agent_mean_lam c
b BUNtoCr_min_sqrt Age_min_sq Age_min_sq R d jaundiceSkin_mean_la h BUNtoCr_min_sqrt u Age_min_sq
jaundiceSkin_mean_la SpO2.oor30.t_mean_sqrt hospTime_min_sqrt SpO2.oor30.t_mean_sqrt
a t v BUNtoCr_mean_sqrt m Na_max_am RESP_mean_sq ! n Na_max_am
a Sympathomimetic_agent_min SpO2.oor30.t_mean_sqrt CO2_mean_i e
5 ShockIdx_mean_sq k o + Sympathomimetic_agent_min r + Lasix_perKg_max_sqrt Sympathomimetic_agent_min !
qu !
e ia
d hospTime_min_sqrt hospTime_min_sqrt + HCTrdv_max ! + CO2_mean_i BUNtoCr_mean_sqrt d
e hrmVA_max ShockIdx_mean_sq ! Na_max_am ! Platelets_mean_i ! b
a + InputOtherBloodB_mean_lam
l. + MBPm.pr_min_am hospTime_min_sqrt !
le Platelets_max_am c PaO2toFiO2_mean ! lly beta.Blocking_agent_mean_lam ! MBPm.pr_min_am ShockIdx_max
o SpO2_mean_am hrmVA_max ! s O + SBPm.oor30.t_max_sq ! + ShockIdx_max
m Platelets_mean_i HemMalig_min ! + InputB_mean_sqrt ! jaundiceSkin_mean_la ! MBPm.pr_min_am
a n s + Mg_min_sq !
p + RESP_mean_sq temp_mean_am !
p CO2_mean_i !
b + Na_Slope_1680_mean ! Intercept + Mg_min_sq a
a Intercept a s Lasix_max_lam ! Intercept
r + HemMalig_min e
c HemMalig_min MetCarcinoma_min jaundiceSkin_mean_la
g
is n e + AIDS_min ! + temp_mean_am
iv dopLg_mean_la ! d t Platelets_max_am
o MetCarcinoma_min ! n
) MetCarcinoma_min
e
f r
n Intercept
f
o o
0
o
m
d
f
r
a
DAS
t
y
t
h
, h
e + urineByHr_mean_sqrt !
e
v n
a + Insulin_sd_sq ! p
f Intercept
r o
in Insulin_mean_la o ! pressD01_mean_am ! ia + Platelets_Slope_1680_min
llo GCSrng_mean_sq s
p INR_mean_i
it b
Wald z LactateM_sd_i ! INR_mean_i
w
u le
iv GCSrng_mean_sq
t C
in Platelets_Slope_1680_min
s + CV_HRrng_max ! + RikerSAS_mean s
e Insulin_mean_la + pressD01_mean_am
HAP g
a svCSRU_max RikerSAS_mean
o W
5 b v Platelets_Slope_1680_min
d svCSRU_max
! pacemkr_max
s
e
a a e + UrineOutB_max_sqrt ! r GCS_max_sq GCS_max_sq
ld + svCSRU_max
y
n T
d INR_mean_i
t
s
Z
a
E
u
f
ys r
ffi GCSrdv_mean alloutput_min_sqrt
R o
s + GCSrng_mean_sq
m
c
xe
n o
alloutput_max_sqrt 5
r
.
d t
∈ WWW.BSSVE.IN e
h GCSrdv_mean ! s
w INR_max_i
e GCS_max_sq
{
M
a
it
1
p
n
,
O
h
r 2
10 d
e pacemkr_max !
R ,
! vio
“ alloutput_max_sqrt !
3
t
-”
T
h
,
ALIT
4
u
. e
, INR_max_i
s n
5 GCS_max_sq
d
}
e
a
g
.
y a
GCS_max_sq Y
I
t
n
a
iv
r
p
M
e
e
u
O
t
p
W r
15
D
n
e a
!
a
fi
ld
E
m
xe
LS
e
Z
d s 1 2 3 4 5 DAS model (day n) www.bsscommunitycollege.in www.bssnewgeneration.inFigure by Hug, www.bsslifeskillscollege.in Caleb Wayne. "Detecting Hazardous Intensive Care Patient Episodes Using Real-time Mortality Models." Massachusetts Institute of Technology, 2009. 236 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
MIT OpenCourseWare http://ocw.mit.edu
HST.950J / 6.872 Biomedical Computing Fall 2010
For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.
WWW.BSSVE.IN
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 237 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Decision Analysis & Decision Support 6.872/HST.950
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 238 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Tasks?
Mechanics Record keeping Administration Scheduling … Diagnosis Prognosis Therapy
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 239 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Types of Decision Support
• “Doctor's Assistant” for clinicians at any level of training • Expert (specialist) consultation for non- specialists • Monitoring and error detection • Critiquing, what-if • Guiding patient-controlled care • Education and Training • Contribution to medical research • …
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 240 Twowww.onlineeducation.bharatsevaksamaj.net Historical www.bssskillmission.in Views on How to Build Expert Systems Great cleverness Powerful inference abilities Ab initio reasoning
Great stores of knowledge Possibly limited ability to infer, but Vast storehouse of relevant knowledge, indexed in an easy-to-apply form
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 241 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Change over 30 years
• 1970’s: human knowledge, not much data • 2000’s: vast amounts of data, traditional human knowledge (somewhat) in doubt
• Could we “re-discover” all of medicine from data? I think not! • Should we focus on methods for reasoning with uncertain data? Absolutely! • But: Feinstein, A. R. (1977). “Clinical Biostatistics XXXIX. The Haze of Bayes, the Aerial Palaces of Decision Analysis, and the Computerized Ouija Board.” Clinical Pharmacology and Therapeutics 21: 482-496. www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 242 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Cancer Test
• We discover a cheap, 95% accurate test for cancer. • Give it to “Mrs. Jones”, the next person who walks by 77 Mass Ave. • Result is positive. • What is the probability that Mrs. Jones has cancer?
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 243 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Figuring out Cancer Probability
Assume Ca in 1% of general population:
+ 950 95%
1,000 Ca - 50 100,000 + 4,950 99,000
95% - 94,050
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 244 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in At the Extremes
• If Ca probability in population is 0.1%, – Then post positive result, p(Ca)=1.87%
• If Ca probability in population is 50%, – Then post-positive result, p(Ca)=95%
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 245 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Bayes’ Rule
+
-
+
- www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 246 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Odds/Likelihood Form
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 247 www.onlineeducation.bharatsevaksamaj.netDeDombal, etwww.bssskillmission.in al. Experience 1970’s & 80’s • “Idiot Bayes” for appendicitis • 1. Based on expert estimates -- lousy • 2. Statistics -- better than docs • 3. Different hospital -- lousy again • 4. Retrained on local statistics -- good
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 248 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Rationality
• Behavior is a continued sequence of choices, interspersed by the world’s responses • Best action is to make the choice with the greatest expected value • … decision analysis
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 249 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Example: Gangrene
• From Pauker’s “Decision Analysis Service” at New England Medical Center Hospital, late 1970’s. • Man with gangrene of foot • Choose to amputate foot or treat medically • If medical treatment fails, patient may die or may have to amputate whole leg. • What to do? How to reason about it?
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 250 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Decision Tree for Gangrene
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 251 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Evaluating the Decision Tree
880
841.5
871.5
686 871.5
686
597 www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 252 www.onlineeducation.bharatsevaksamaj.netDecision www.bssskillmission.in Analysis: Evaluating Decision Trees
• Outcome: directly estimate value • Decision: value is that of the choice with the greatest expected value • Chance: expected value is sum of (probabilities x values of results) • “Fold back” from outcomes to current decision. • Sensitivity analyses often more important than result(!) www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 253 www.onlineeducation.bharatsevaksamaj.netHELP System www.bssskillmission.in uses D.A.
Intellectual Tools
MC = .014 4000
3000 MC = .010 M = .014
u) C
∆ 2000
+1000 MC = .014 MC = .008 0 MC = .008 -1000 Utility of operating minus Utility of not operating ( 2000 Age 27 Age 57
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 Probability that patient has appendicitis
Effect of age patient and MC (mortality for appendicitis without operation) on the probability threshold (point of crossing zero ∆ u line) for decision to operate.
Image by MIT OpenCourseWare. Adapted from Warner, Homer R. "Computer-assisted medical decision making." Academic Press, 1979. Warner HR, Computer-Assisted Medical www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in WWW.BSSVE.INDecision Making, Acad. Press 1979 254 www.onlineeducation.bharatsevaksamaj.netUtility Analysis www.bssskillmission.in of Appendectomy
4: Computer Representation of Medical Knowledge
5000
4000
3000 ) ∆µ 2000
+1000
0 Value of 1 day -1000 Salary of good health Utility of operating minus
Utility of not operating ( 35000 140 2000 16000 140 8000 140 3000 35000 70 16000 70 8000 70
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 Probability of appendicitis Effect of patient's salary and assumed value of one day of good health ($70 or $140) on decision to operate for appendicitis.
www.bsscommunitycollege.inImage bywww.bssnewgeneration.in MIT OpenCourseWare. Adapted www.bsslifeskillscollege.in from Warner, Homer R. WWW.BSSVE.IN"Computer-assisted medical decision making." Academic Press, 1979. 255 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in PROB OF APPENDICITIS
A APPENDICITIS BY HISTORY B REBOUND TENDERNESS IN RLQ C PRIOR APPENDECTOMY D IF C THEN EXIT E WHITE BLOOD COUNT (WBCx100) TH/M3, LAST F PROB B A 620 90 G PROB F 43 18 9, 74 23 7, 93 18 11, 108 10 11, 121 16 13, 134 6 16, 151 5 16, 176 4 14 FVAL G
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 256 UTILITYwww.onlineeducation.bharatsevaksamaj.net OF APPENDECTOMY www.bssskillmission.in IS ESTIMATED AS $----
A (A) AGE B SEX C (A) SALARY, GET A/365 D JOB, PERCENT ACTIVITY NEEDED E LE A,B F DLOS D 30 1, 65 2, 80 4, 90 1, 100 – 0 G DLOS D 40 1, 80 4, 95 5, 100 – 0 … I COND E, F, 7, 1800, 0, C J COND E, G, 1, 900, 0, C … M PROB OF APPENDICITIS N UTIL M, I, J, K, L O IF N LT 0, EXIT FVAL N
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 257 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in “Paint the Blackboards!”
DECISION PATIENT STATE UTILITY
Treat disease Disease (p)
treat No disease (1-p) Treat no disease
Disease (p) No treat disease No treat
No disease (1-p) No treat no disease www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 258 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Threshold
• Benefit B = U(treat dis) – U (no treat dis) • Cost C = U(no treat no dis) – U(treat no dis) • Threshold probability for treatment:
Pauker, Kassirer, NEJM 1975 www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 259 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Test/Treat Threshold
Figure removed due to copyright restrictions. See Kassirer, Jerome P., and Stephen G. Pauker. "Should Diagnostic Testing be Regulated?" New England Journal of Medicine (1978).
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 260 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Visualizing Thresholds
Figure removed due to copyright restrictions. See Kassirer, Jerome P., and Stephen G. Pauker. "Should Diagnostic Testing be Regulated?" New England Journal of Medicine (1978).
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 261 www.onlineeducation.bharatsevaksamaj.netMore Complex www.bssskillmission.in Decision Analysis Issues • Repeated decisions • Accumulating disutilities • Dependence on history • Cohorts & state transition models • Explicit models of time • Uncertainty in the uncertainties • Determining utilities – Lotteries, … • Qualitative models
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 262 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Example: Acute Renal Failure
• Based on Gorry, et al., AJM 55, 473-484, 1973. • Choice of a handful (8) of therapies (antibiotics, steroids, surgery, etc.) • Choice of a handful (3) of invasive tests (biopsies, IVP, etc.) • Choice of 27 diagnostic “questions” (patient characteristics, history, lab values, etc.) • Underlying cause is one of 14 diseases – We assume one and only one disease
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 263 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Decision Tree for ARF
• Choose: – Surgery for obstruction – Treat with antibiotics – Perform pyelogram – Perform arteriography – Measure patient’s temperature – Determine if there is proteinuria – …
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 264 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Decision Tree for ARF
Surgery for obstruction Value = ??? Treat with antibiotics Perform pyelogram Perform arteriography Measure patient’s temperature Determine if there is proteinuria
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 265 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in What happens when we act?
• Treatment: leads to few possible outcomes – different outcomes have different probabilities • probabilities depend on distribution of disease probabilities – value of outcome can be directly determined • value may depend on how we got there (see below) • therefore, value of a treatment can be determined by expectation • Test: lead to few results, revise probability distribution of diseases, and impose disutility • Questions: lead to few results, revise probability distribution
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 266 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Full decision tree
A1 A1 A1 A2 P´ A2 A2 A3 P A3 A3 A4 P´´ A4 A4
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 267 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Initial probability distribution
ATN Acute tubular necrosis 0.250 FARF Functional acute renal failure 0.400 OBSTR Urinary tract obstruction 0.100 AGN Acute glomerulonephritis 0.100 CN Renal cortical necrosis 0.020 HS Hepatorenal syndrome 0.005 PYE Pyelonephritis 0.010 AE Atheromatous Emboli 0.003 RI Renal infarction (bilateral) 0.002 RVT Renal vein thrombosis 0.002 VASC Renal vasculitis 0.050 SCL Scleroderma 0.002 CGAE Chronic glomerulonephritis, acute exacerbation 0.030 MH Malignant hypertension & nephrosclerosis 0.030
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 268 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in ARF’s Database: P(obs|D)
Conditional probabilities for Probabilities Proteinuria Trace 3+ to Diseases 0 to 2+ 4+
ATN 0.1 0.8 0.1 FARF 0.8 0.2 0.001 OBSTR 0.7 0.3 0.001 AGN 0.01 0.2 0.8 CN 0.01 0.8 0.2 HS 0.8 0.2 0.001 PYE 0.4 0.6 0.001 AE 0.1 0.8 0.1 RI 0.1 0.7 0.2 RVT 0.001 0.1 0.9 VASC 0.01 0.2 0.8 SCL 0.1 0.4 0.5 CGAE 0.001 0.2 0.8 MH 0.001 0.4 0.6
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 269 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Questions
• Blood pressure at onset • history of proteinuria • proteinuria • symptoms of bladder obstruction • casts in urine sediment • exposure to nephrotoxic drugs • disturbance in clotting mechanism • hematuria • pyuria • history of prolonged hypotension • bacteriuria • urine specific gravity • sex • large fluid loss preceding onset • transfusion within one day • kidney size • jaundice or ascites • ischemia of extremities or aortic • urine sodium aneurism • strep infection within three weeks • atrial fibrillation or recent MI • urine volume • recent surgery or trauma • age • papilledema • flank pain www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 270 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Invasive tests and treatments
• Tests • Treatments – biopsy – steroids – retrograde – conservative therapy pyelography – iv-fluids – transfemoral – surgery for urinary tract arteriography obstruction – antibiotics – surgery for clot in renal vessels – antihypertensive drugs – heparin www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 271 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Updating probability distribution
Bayes’ rule www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 272 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Value of treatment
• Three results: improved, unchanged, worsened – each has an innate value, modified by “tolls” paid on the way • Probabilities depend on underlying disease probability distribution I V(I)
Tx U V(U)
W V(W) www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 273 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Modeling treatment
Utilities: improved: 5000 unchanged: -2500 worse: -5000
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 274 www.onlineeducation.bharatsevaksamaj.netModeling www.bssskillmission.in test: transfemoral arteriography
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 275 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in How large is the tree?
• Infinite, or at least (27+3+8)^(27+3+8), ~10^60 • What can we do? – Assume any action is done only once – Order: • questions • tests • treatments • 27! x 4 x 3 x 2 x 8, ~10^30 • Search, with a myopic evaluation function – like game-tree search; what’s the static evaluator? – Measure of certainty in the probability distribution www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 276 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in How many questions needed?
• How many items can you distinguish by asking 20 (binary) questions? 2^20 • How many questions do you need to ask
to distinguish among n items? log2(n) • Entropy of a probability distribution is a measure of how certainly the distribution identifies a single answer; or how many more questions are needed to identify it
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 277 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Entropy of a distribution
For example: H(.5, .5) = 1.0 H(.1, .9) = 0.47 P H(.01, .99) = 0.08 H(.001, .999) = 0.01
H(.33, .33, .33) = 1.58 (!) H(.005, .455, .5) = 1.04 H(.005, .995, 0) = 0.045 j
(!) -- should use logn www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 278 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Interacting with ARF in 1973
Question 1: What is the patient's age? 1 0-10 2 11-30 3 31-50 4 51-70 5 Over 70 Reply: 5
The current distribution is: Disease Probability FARF 0.58 IBSTR 0.22 ATN 0.09
Question 2: What is the patient's sex? 1 Male 2 Pregnant Female 3 Non-pregnant Female Reply: 1 . . . www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 279 www.onlineeducation.bharatsevaksamaj.netARF www.bssskillmission.in in 1994
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 280 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Local Sensitivity Analysis
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 281 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Case-specific Likelihood Ratios
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 282 www.onlineeducation.bharatsevaksamaj.netTherapy Planning www.bssskillmission.in Based on Utilities
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 283 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Assumptions in ARF
• Exhaustive, mutually exclusive set of diseases • Conditional independence of all questions, tests, and treatments • Cumulative (additive) disutilities of tests and treatments • Questions have no modeled disutility, but we choose to minimize the number asked anyway www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 284 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in MIT OpenCourseWare http://ocw.mit.edu
HST.950J / 6.872 Biomedical Computing Fall 2010
For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 285 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Decision Support via Expert Systems
6.872/HST950 Peter Szolovits
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 286 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Components of an Expert System
• Knowledge – In various forms: associations, models, etc. • Strategy – Baconian, exhaustive enumeration, on-line, etc. • Implementation – Programs, pattern matching, rules, etc.
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 287 Flowchartwww.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
BI/Lincoln Labs Clinical Protocols 1978
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 288 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Codifying Human Knowledge
• Decomposition into “chunks” of knowledge, chaining of inferences • Matching of case data to prototypical situations • Using causal models (pathophysiology) to figure out cases
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 289 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Mycin—Rule-based Systems
• Task: Diagnosis and prescription for bacterial infections of the blood (and later meningitis) • Method: RULE037 – Collection of modular IF the organism rules 1) stains grampos – Backward chaining 2) has coccus shape – Certainty factors 3) grows in chains
THEN There is suggestive evidence (.7) that the identity of the organism is streptococcus. www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 290 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Mycin consult ------PATIENT-1------1) Patient's name: FRED SMITH 2) Sex: MALE 3) Age: 55 4) Have you been able to obtain positive cultures from a site at which Fred Smith has an infection? YES ------INFECTION-1------5) What is the infection? PRIMARY-BACTEREMIA 6) Please give the date when signs of INFECTION-1 appeared. 5/5/75 The most recent positive culture associated with the primary- bacteremia will be referred to as: ------CULTURE-1------7) From what site was the specimen for CULTURE-1 taken? BLOOD 8) Please give the date when this culture was obtained. 5/9/75 The first significant organism from this blood culture will be called: ------ORGANISM-1------9) Enter the identity of ORGANISM-1. UNKNOWN 10) Is ORGANISM-1 a rod or coccus (etc.)? ROD 11) The gram stain of ORGANISM-1: GRAMNEG . . . Davis, et al., Artificial Intelligence 8: 15-45 (1977) www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 291 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in How Mycin Works
• To find out a fact – If there are rules that can conclude it, try them – Ask the user • To “run” a rule – Try to find out if the facts in the premises are true – If they all are, then assert the conclusion(s), with a suitable certainty • Backward chaining from goal to given facts
Dynamically traces out behavior of (what might be) a flowchart Information used everywhere appropriate Single expression of any piece of knowledge
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 292 www.onlineeducation.bharatsevaksamaj.netExplore Mycin’s www.bssskillmission.in Use of Knowledge ** Did you use RULE 163 to find out anything about ORGANISM-1? RULE163 was tried in the context of ORGANISM-1, but it failed because it is not true that the patient has had a genito-urinary tract manipulative procedure (clause 3). ** Why didn't you consider streptococcus as a possibility? The following rule could have been used to determine that the identity of ORGANISM-1 was streptococcus: RULE033 But clause 2 (“the morphology of the organism is coccus”) was already known to be false for ORGANISM-1, so the rule was never tried. Davis, et al., Artificial Intelligence 8: 15-45 (1977)
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 293 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Even Simpler Representation
s1 s1 s2 s2 s3 s3 s4 s4 s5 s5 s6 s6 s7 s7 s8 s8 s9 Disease s9 Disease s10 s10 s... s...
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 294 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Diagnosis by Card Selection
s1 s1 s1 s2 s1 s2s3 s2s3 s2s3 s4 s3 s4s5 s4s5 s4s5 s6 s5 s6s7 s6s7 s6s7 s8 s7 s8s9 Disease s8s9 Disease s8s9 s10 Disease s9 s10s...Disease s10s... s10s... s...
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 295 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Diagnosis by Edge-Punched Cards
Dx is intersection of sets of diseases that may cause all the observed symptoms Difficulties: Uncertainty Multiple diseases ~ “Problem-Knowledge Coupler” of Weed
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 296 Takingwww.onlineeducation.bharatsevaksamaj.net the Present www.bssskillmission.in Illness—Diagnosis by Pattern Directed Matching
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 297 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in PIP's Theory of Diagnosis
• From initial complaints, guess suitable hypothesis. • Use current active hypotheses to guide questioning • Failure to satisfy expectations is the strongest clue to a better hypothesis; differential diagnosis • Hypotheses are activated, de-activated, confirmed or rejected based on (1) logical criteria (2) probabilities based on: findings local to hypothesis causal relations to other hypotheses
The Scientific Method www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 298 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Memory Structure in PIP
Triggers Logical Criteria
Probabilistic Manifestations Scoring Function Hypothesis
Differential Causally and Diagnosis Associationally Heuristics Related Hyp's
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 299 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in PIP's Model of Nephrotic Syndrome
NEPHROTIC SYNDROME, a clinical state FINDINGS: 1* Low serum albumin concentration 2. Heavy proteinuria 3* >5 gm/day proteinuria 4* Massive symmetrical edema 5* Facial or peri-orbital symmetric edema 6. High serum cholesterol 7. Urine lipids present IS-SUFFICIENT: Massive pedal edema & >5 gm/day proteinuria MUST-NOT-HAVE: Proteinuria absent SCORING . . . MAY-BE-CAUSED-BY: AGN, CGN, nephrotoxic drugs, insect bite, idiopathic nephrotic syndrome, lupus, diabetes mellitus MAY-BE-COMPLICATED-BY: hypovolemia, cellulitis MAY-BE-CAUSE-OF: sodium retention DIFFERENTIAL DIAGNOSIS: neck veins elevated ➠ constrictive pericarditis ascites present ➠ cirrhosis pulmonary emboli present ➠ renal vein thrombosis www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 300 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in QMR Partitioning
M1
M2
H1 M3 H2
M4
M5
M6 www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 301 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Competitors
M1
M2
H1 M3 H2
M4
M5
M6 www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 302 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Still Competitors
M1
M2
H1 M3 H2
M4
M5
M6 www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 303 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Probably Complementary
M1
M2
H1 M3 H2
M4
M5
M6 www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 304 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Multi-Hypothesis Diagnosis
Set aside complementary hypotheses … and manifestations predicted by them Solve diagnostic problem among competitors Eliminate confirmed hypotheses and manifestations explained by them Repeat as long as there are coherent problems among the remaining data
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 305 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Internist/QMR
Knowledge Base: 956 hypotheses 4090 manifestations (about 75/hypothesis) Evocation like P(H|M) Frequency like P(M|H) Importance of each M Causal relations between H’s Diagnostic Strategy: Scoring function Partitioning Several questioning strategies www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 306 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in QMR Database
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 307 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in QMR Scoring
Positive Factors Evoking strength of observed Manifestations Scaled Frequency of causal links from confirmed Hypotheses Negative Factors Frequency of predicted but absent Manifestations Importance of unexplained Manifestations Various scaling parameters (roughly exponential) www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 308 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Example Case
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 309 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Initial Solution
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 310 www.onlineeducation.bharatsevaksamaj.netSymptom www.bssskillmission.in Clustering for Multi-Disorder Diagnosis
— Tom Wu, Ph.D. 1991
Assume a bipartite graph representation of diseases/ symptoms Given a set of symptoms, how to proceed? If we could “guess” an appropriate clustering of the symptoms so that each cluster has a single cause …
… then the solution is (d5, d6) x (d3, d7, d8, d9) x (d1, d2, d4)
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 311 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Clustering Alternatives
Symptom Possible Causes Hep Asth Fever TB, Hepatitis, Malaria TB Bron Cough TB, Asthma, Bronchitis, Mal Emphysema Emph
H1 H2 Fever, Cough Fever Cough Hep Asth
TB Bron TB Hep Asth Mal Mal Bron Emph Emph www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 312 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Synopsis in Renal Disease • Diseases • Symptoms – Hypertension (HTN) – High urine osmolality (Osm↑) – Acute glomerulonephritis (AGN) – High urine specific gravity (Sg↑) – IgA nephropathy (IgA) – Low urine sodium (Na↓) – Prerenal azotemia (PRA) – Low urine pH (pH↓) – Hepatorenal syndrome (HRS) – Renal vasculitis (RV) – Congestive heart failure (CHF) – Aldosteronism (Aldo) – Constrictive pericarditis (Peri) – Diabetic ketoacidosis (DKA) – Analgesic nephropathy (AN) – Hypokalemic nephropathy (HKN) – Chronic renal failure (CRF) HTN AGN IgA PRA HRS RV CHF Aldo Peri DKA AN HKN CRF RTA Osm X X X X X X ↑ Sg↑ X X X X X X X Na↓ X X X X X pH↓ X X X X X X X www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 313 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in After Osm Osm↑ ↑
HTN AGN IgA PRA HRS RV
HTN AGN IgA PRA HRS RV CHF Aldo Peri DKA AN HKN CRF RTA Osm X X X X X X ↑ Sg↑ X X X X X X X Na↓ X X X X X pH↓ X X X X X X X www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 314 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Osm↑, Sg↑ Add Sg↑
HTN Cover AGN IgA PRA HRS RV
HTN AGN IgA PRA HRS RV CHF Aldo Peri DKA AN HKN CRF RTA Osm X X X X X X ↑ Sg↑ X X X X X X X Na↓ X X X X X pH↓ X X X X X X X www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 315 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Add Na↓ Restrict Append
Osm↑, Sg↑, Osm↑, Sg↑ Na↓ Na↓ HTN Aldo PRA AGN CHF HRS or IgA Peri RV
HTN AGN IgA PRA HRS RV CHF Aldo Peri DKA AN HKN CRF RTA Osm X X X X X X ↑ Sg↑ X X X X X X X Na↓ X X X X X pH↓ X X X X X X X www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 316 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Search Space C=cover (Osm↑) R=restrict C A=append (Osm↑, Sg↑) E=extract R A
(Osm↑, Sg↑, Na↓) (Osm↑, Sg↑) (Na↓)
R E R A A (Osm↑, Sg↑, Na↓, pH↓) (Na↓) (Osm↑, Sg↑, pH↓)
(Osm↑, Sg↑, Na↓) (pH↓) (Osm↑, Sg↑) (Na↓) (pH↓)
HTN AGN IgA PRA HRS RV CHF Aldo Peri DKA AN HKN CRF RTA Osm X X X X X X ↑ Sg↑ X X X X X X X Na↓ X X X X X pH↓ X www.bsscommunitycollege.inWWW.BSSVE.INX www.bssnewgeneration.in www.bsslifeskillscollege.inX X X X X 317 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Symptom Clustering is Efficient
• Like in any “planning island” approach, reducing an exponential problem to several smaller exponential problems vastly improves efficiency, if it captures some insight into the problem.
• Wu's algorithm (SYNOPSIS) will keep a compact encoding even if it overgenerates slightly.
• E.g., suppose that of the set of diseases represented by (d5, d6) x (d3, d7, d8, d9) x (d1, d2, d4), d6 x d8 x d1 is not a candidate. To represent this precisely would require enumerating the 23 valid candidates. Instead, the factored representation is kept.
In a diagnostic problem drawn from a small subset of the Internist database, it is a power of 3 faster and a power of 5 more compact than standard symptom clustering.
Guide search via probabilities, if we have a reasonable model(!)
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 318 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in More Expert Systems
• Causality? • What’s in a Link? • Temporal reasoning • Quantitative reasoning • Model-based reasoning • Workflow
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 319 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Meaning of Representation?
causes D S
• Always? probability • Magnitude? severity; bad cold worse fever? • Delay? temporality • Where? spatial dependency • Under what conditions? context • Interaction of multiple causes physical laws • Cross-terms high-dimensional descriptions
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 320 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Temporal Reasoning
Keeping track of multiple forms of temporal relations (Kahn '75) The time line “On Dec. 12 last year . . .” Special reference events “Three days after I was hospitalized in 1965 . . .” Temporal Ordering Chains “It must have been before I graduated from high school.”
Constraint propagation (Kohane '87) Primitive relation: e1, e2, lower, upper bounds Heuristics for propagation based on semantic grouping
3 ≤ T(E2)-T(E1) ≤ 5 E2 2 ≤ T(E3)-T(E2) ≤ 7 2, 7 3, 5 Therefore E1 E3 l=5 ≤ T(E3)-T(E1) ≤ 12=u l, u www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 321 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Exploiting Temporal Relations
abdominal ? pain blood transfusion ?
jaundice
transfusion precedes both abdominal pain and jaundice implies transfusion-borne acute hepatitis B as in 1, but only by one day jaundice occurred 20 years ago, transfusion and pain recent Can be very efficient at filtering out nonsense hypotheses. www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 322 www.onlineeducation.bharatsevaksamaj.netInterpreting www.bssskillmission.in the Past with a Causal/Temporal Model
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 323 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Postdiction
Long, Reasoning about State from Causation and Time in a Medical Domain, AAAI 83
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 324 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Temporal Representation can be Complex
Figure of EKG data from heart removed due to copyright restrictions.
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 325 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Time The usual: e1 e2 • point, intervals, constraints e2 min ≤ duration ≤ max e1 e3 • timelines, reference events, fuzz, … The unusual • cyclic edema • focal glomerulonephritis • patterns of fever Systems issues • flow of "now" • supporting the illusion of "instantaneous" decision-making within a temporal reasoner – correcting the past – reasoning by hindsight
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 326 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in The Surprisingly Normal pH
Diarrhea causes bicarbonate (alkali) loss Vomiting causes acid loss Therefore, normal pH is a manifestation of {diarrhea + vomiting}!
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 327 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Multi-Level Causal Model
Figure of three-level causal model removed due to copyright restrictions.
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 328 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Reasoning from Models
Model handles all possible interactions, without having explicitly to anticipate them all Reasoning: Fit parameters to a physiological model, then predict consequences to suggest other expected findings reasonable interventions Qualitative models Combining associational and model-based reasoning
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 329 www.onlineeducation.bharatsevaksamaj.netGuyton's www.bssskillmission.in Model of Cardiovascular Dynamics
Figure removed due to copyright restrictions.
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 330 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Long's Clinical Model of Heart Failure Predictions for Mitral Stenosis with Exercise
Figure removed due to copyright restrictions.
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 331 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Physiological
"All variations in myocardial contractile activity can be expressed as displacements of the force-velocity curve. However, there are two fundamental ways in which the force-velocity curve can be shifted. Figure {left} shows a family of force-velocity curves obtained from an isolated cardiac muscle; each curve was obtained at a different preload, i.e., with a different degree of stretch on the muscle. Note that changing the preload has altered the intercept of the force-velocity curve on the horizontal axis; i.e., it has increased the isometric force developed by the muscle. However, these alterations in preload have not altered the intrinsic velocity of shortening, since all the curves extrapolate to the same intercept on the vertical axis. Thus, a change in initial length of heart muscle shifts the force-velocity curve by altering the total force which can be developed by the muscle. This type of shift in the force-velocity curve may be contrasted with that obtained when a positive inotropic agent, such as norepinephrine or digitalis, is added to the muscle while the initial length is held constant (Fig. {right}). These agents not only increase the force which the muscle is capable of lifting, i.e., the intercept of the force-veolocity curve on the horizontal axis, but also increase the velocity of shortening of the unloaded muscle, i.e., the extrapolated intercept on the vertical axis." — Harrison's (6th ed.) www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 332 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Figures
Normal cat-muscle Inotropic Agent
Figures removed due to copyright restrictions.
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 333 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Clinical Knowledge "… from the clinical point of view, heart failure may be considered to be a disease state in which an abnormality of myocardial function is responsible for the inability of the heart to pump blood at a rate commensurate with the requirements of the metabolizing tissues. Though a defect in myocardial contraction always exists in heart failure, this disorder may result from a primary abnormality in the heart muscle or it may be secondary to a chronic excessive work load. It is important to distinguish heart failure from (1) states of circulatory insufficiency in which myocardial function is not primarily impaired, such as cardiac tamponade, hemorrhagic shock, or tricuspid stenosis, (2) conditions in which there is circulatory congestion because of abnormal salt and water retention but in which there is no serious disturbance of myocardial function, and (3) conditions in which the normal heart is suddenly presented with a load which exceeds its capacity, e.g., accelerated hypertension."
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 334 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in MIT OpenCourseWare http://ocw.mit.edu
HST.950J / 6.872 Biomedical Computing Fall 2010
For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 335 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Bayes Networks 6.872/HST.950
WWW.BSSVE.IN
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 336 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in What Probabilistic Models Should We Use?
• Full joint distribution • Completely expressive • Hugely data-hungry • Exponential computational complexity • Naive Bayes (full conditional independence) • Relatively concise • Need data ~ (#hypotheses) × (#features) × (#feature-vals) WWW.BSSVE.IN• Fast ~ (#features) • Cannot express dependencies among features or among hypotheses • Cannot consider possibility of multiple hypotheses co- occurring
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 337 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Bayesian Networks (aka Belief Networks)
• Graphical representation of dependencies among a set of random variables • Nodes: variables • Directed links to a node from its parents: direct probabilistic dependencies • Each Xi has a conditional probability distribution, P(Xi|Parents(Xi)), showing the effects of the parents on the node. WWW.BSSVE.IN• The graph is directed (DAG); hence, no cycles. • This is a language that can express dependencies between Naive Bayes and the full joint distribution, more concisely • Given some new evidence, how does this affect the probability of some other node(s)? P(X|E) —belief propagation/updating • Given some evidence, what are the most likely values of other variables? —MAP explanation
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 338 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Burglary Network (due to J. Pearl)
Burglary Earthquake
Alarm
JohnCalls MaryCalls WWW.BSSVE.IN
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 339 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Burglary Network (due to J. Pearl)
P(B) P(E) 0.001 0.002 Burglary Earthquake
B E P(A|B,E) t t 0.95 Alarm t f 0.94 f t 0.29 f f 0.001
JohnCalls MaryCalls
WWW.BSSVE.INA P(J|A) A P(M|A) t 0.90 t 0.70 f 0.05 f 0.01
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 340 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
If everything depends on everything
21 0 Burglary Earthquake 2
22 Alarm
JohnCalls MaryCalls WWW.BSSVE.IN23 24
• This model requires just as many parameters as the full joint distribution!
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 341 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Computing the Joint Distribution from a Bayes Network
• As usual, we abuse notation:
• • E.g., what’s the probability that an alarm has sounded, there was neither an earthquake nor a burglary, and both John and Mary called? WWW.BSSVE.IN
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 342 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Requirements for Constructing a BN • Recall that the definition of the conditional probability was • and thus we get the chain rule, • Generalizing to n variables, • and repeatedly applying this idea,
WWW.BSSVE.IN
• This “works” just in case we can define a partial order so that
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 343 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Topological Interpretations
U1 Un U1 Un
X X Z1 Zn Z1 Zn
Y1 Yn Y1 Yn
A node, XWWW.BSSVE.IN, is conditionally independent of A node, X, is conditionally independent of all other its non-descendants, Zi, given its parents, Ui. nodes in the network given its Markov blanket: its parents, Ui, children,Yi, and children’s parents, Zi.
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 344 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
BN’s can be Compact
• For a network of 40 binary variables, the full joint distribution has 240 entries (> 1,000,000,000,000) • If |Par(xi)| ≤ 5, however, then the 40 (conditional) probability tables each have ≤ 32 entries, so the total number of parameters ≤ 1,280 • WWW.BSSVE.INLargest medical BN I know (Pathfinder) had 109 variables! 2109 ≈ 1036
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 345 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Burglary Earthquake How Not to Build BN’s Alarm
JohnCalls MaryCalls • With the wrong ordering of nodes, the network becomes more complicated, and requires more (and more difficult) conditional probability assessments
MaryCalls JohnCalls MaryCalls JohnCalls
WWW.BSSVE.INAlarm Earthquake
Burglary Earthquake Burglary Alarm
Order: M, J,A, B, E Order: M, J, E, B,A www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 346 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Simplifying Conditional Probability Tables • Do we know any structure in the way that Par(x) “cause” x? • If each destroyer can sink the ship with probability P(s|di), what is the probability that the ship will sink if it’s attacked by both?
• For |Par(x)| = n, this requires O(n) parameters, not O(kn)
WWW.BSSVE.IN
Image by MIT OpenCourseWare. Image by MIT OpenCourseWare. Photo by Konabish on Flickr.
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 347 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Inference
• Recall the two basic inference problems: Belief propagation & MAP explanation • Trivially, we can enumerate all “matching” rows of the joint probability distribution • For poly-trees (not even undirected loops—i.e., only one connection between any pair of nodes; like our Burglary example), there are efficient linear algorithms, similar to WWW.BSSVE.INconstraint propagation • For arbitrary BN’s, all inference is NP-hard! • Exact solutions • Approximation
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 348 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Burglary Earthquake Exact Solution of BN’s Alarm (Burglary example)
JohnCalls MaryCalls
WWW.BSSVE.IN • Notes: • Sum over all “don’t care” variables • Factor common terms out of summation • Calculation becomes a sum of products of sums of products ...
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 349 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Poly-trees are easy
• Singly-connected structures Escape allow propagation of observations via single paths Burglary Earthquake “Down” is just use of • Drunk Alarm TVshow conditional probability
“Up” is just Bayes rule JohnCalls MaryCalls • Bottles • Formulated as message WWW.BSSVE.INpropagation rules • Linear time (network diameter) • Fails on general networks!
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 350 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Exact Solution of BN’s (non-poly-trees) A
• What is the probability of a specific state, say B C A=t, B=f, C=t, D=t, E=f?
D E • What is the probability that E=t given B=t? • Consider the term P(e,b) WWW.BSSVE.IN
Alas, optimal factoring is NP-hard • 12 instead of 32 multiplications (even in this www.bsscommunitycollege.insmall www.bssnewgeneration.inexample) www.bsslifeskillscollege.in 351 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Other Exact Methods
A A
B C B,C
D E D E • Join-tree: Merge variables into (small!) sets of variables to make graph into a poly-tree. Most commonly-used; aka Clustering, WWW.BSSVE.INJunction-tree, Potential) • Cutset-conditioning: Instantiate a (small!) set of variables, then solve each residual problem, and add solutions weighted by probabilities of the instantiated variables having those values • ... • All these methods are essentially equivalent; with some time- space tradeoffs. www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 352 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Approximate Inference in BN’s
• Direct Sampling—samples joint distribution • Rejection Sampling—computes P(X|e), uses ancestor evidence nodes in sampling • Likelihood Weighting—like Rejection Sampling, but weights by probability of descendant evidence nodes • WWW.BSSVE.INMarkov chain Monte Carlo • Gibbs and other similar sampling methods
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 353 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Direct Sampling
function Prior-Sample(bn) returns an event sampled from bn inputs: bn, a Bayes net specifying the joint distribution P(X1, ... Xn) x := an event with n elements for i = 1 to n do xi := a random sample from P(Xi|Par(Xi)) return x
• WWW.BSSVE.INFrom a large number of samples, we can estimate all joint probabilities • The probability of an event is the fraction of all complete events generated by PS that match the partially specified event • hence we can compute all conditionals, etc.
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 354 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Rejection Sampling
function Rejection-Sample(X, e, bn, N) returns an estimate of P(X|e) inputs: bn, a Bayes net X, the query variable e, evidence specified as an event N, the number of samples to be generated local: K, a vector of counts over values of X, initially 0
for j = 1 to N do y := PriorSample(bn) if y is consistent with e then K[v] := K[v]+1 where v is the value of X in y return Normalize(K[X]) WWW.BSSVE.IN • Uses PriorSample to estimate the proportion of times each value of X appears in samples that are consistent with e • But, most samples may be irrelevant to a specific query, so this is quite inefficient
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 355 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Likelihood Weighting
• In trying to compute P(X|e), where e is the evidence (variables with known, observed values), • Sample only the variables other than those in e Weight each sample by how well it predicts e WWW.BSSVE.IN•
l m S (z, e)w(z, e)= P (z Par(Z )) P (e Par(E )) WS i| i i| i i=1 i=1 ! ! = P (z, e)
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 356 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in l m Likelihood S (z, e)w(z, e)= P (z Par(Z )) P (e Par(E )) WS i|i i | i i=1 i=1 Weighting ! ! = P (z, e) function Likelihood-Weighting(X, e, bn, N) returns an estimate of P(X|e) inputs: bn, a Bayes net X, the query variable e, evidence specified as an event N, the number of samples to be generated local:W, a vector of weighted counts over values of X, initially 0 for j = 1 to N do y,w := WeightedSample(bn,e) if y is consistent with e then W[v] := W[v]+w where v is the value of X in y return Normalize(W[X]) functionWWW.BSSVE.IN Weighted-Sample(bn,e) returns an event and a weight x := an event with n elements; w := 1 for i = 1 to n do if Xi has a value xi in e then w := w * P(Xi = xi | Par(Xi)) else xi := a random sample from P(Xi | Par(Xi)) return x,w
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 357 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Markov chain Monte Carlo U1 Un
function MCMC(X, e, bn, N) returns an estimate of P(X|e) X Z1 Zn local: K[X], a vector of counts over values of X, initially 0 Z, the non-evidence variables in bn (includes X) x, the current state of the network, initially a copy of e Y1 Yn initialize x with random values for the vars in Z for j = 1 to N do for each Zi in Z do sample the value of Zi in x from P(Zi|mb(Zi)), given the values of mb(Zi) in x K[v] := K[v]+1 where v is the value of X in x return Normalize(K[X]) • Wander incrementally from the last state sampled, instead of re- WWW.BSSVE.INgenerating a completely new sample • For every unobserved variable, choose a new value according to its probability given the values of vars in it Markov blanket (remember, it’s independent of all other vars) • After each change, tally the sample for its value of X; this will only change sometimes • Problem: “narrow passages” www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 358 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Most Probable Explanation
• So far, we have been solving for P(X|e), which yields a distribution over all possible values of the x’s • What it we want the best explanation of a set of evidence, i.e., the highest-probability set of values for the x’s, given e? • Just maximize over the “don’t care” variables rather than summing • WWW.BSSVE.INThis is not necessarily the same as just choosing the value of each x with the highest probability
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 359 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Rules and Probabilities
• Many have wanted to put a probability on assertions and on rules, and compute with likelihoods • E.g., Mycin’s certainty factor framework • A (p=.3) & B (p=.7) ==p=.8==> C (p=?) • Problems: • How to combine uncertainties of preconditions and of rule How to combine evidence from multiple rules WWW.BSSVE.IN• • Theorem:There is NO such algebra that works when rules are considered independently. • Need BN for a consistent model of probabilistic inference
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 360 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
MIT OpenCourseWare http://ocw.mit.edu
HST.950J / 6.872 Biomedical Computing Fall 2010
For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.
WWW.BSSVE.IN
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 361 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Exploring MIMIC to learn from practice variation Leo Anthony Celi MD, MS, MPH Beth Israel Deaconess Medical Center Harvard-MIT Health Sciences & Technology Division
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 362 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Collaborative Ecosystem
• Beth Israel Deaconess Medical Center – Department of Medicine – Surgical ICU – Division of Cardiothoracic Anesthesia – Division of Dermatology – Department of Pharmacy – Division of Infectious Disease
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 363 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Collaborative Ecosystem
Various logos have been removed due to copyright restrictions, including Mount Sinai School of Medicine, Escuela de Ingeniera de Antioquia, Mount Auburn Hospital, University of Oxford, NHS, MIT Portugal, among others.
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 364 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Goals
• Present an overview of clinical research in progress • Provide a unifying theme as regards the motivation behind the projects • Introduce a vision of an empiric data-driven day-to-day practice
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 365 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Evidence-Based Medicine
• Multi-center PRCTs and systematic reviews are gold standard • PRCTs provide aggregated outcomes – difficult to apply to individual patients • Benefits may not translate into the real world – efficacy vs. effectiveness • Errors and biases abound: 41% of the most cited original clinical research later refuted (Ioannidis, JAMA 2005)
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 366 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Evidence-Based Medicine
• 2007 analysis of >1000 Cochrane systematic reviews – 49%: current evidence does not support either benefit or harm – 96%: additional research is recommended • Most of what clinicians do has never been formally put to the test
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 367 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Evidence-Based Medicine
• Large-scale evidence impossible to obtain for the millions of questions posed in day-to-day practice • Is there a role for highly granular clinical databases such as MIMIC?
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 368 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Collective Experience
• Aggregation of knowledge extractable from actual patient care of numerous clinicians • Capture clinician heuristics mathematically : predicting fluid requirement (Celi et al., Crit Care 2008) • Build patient subset-specific models: mortality prediction (Celi et al., J Healthcare Eng 2011) • Examine areas with significant care variability
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 369 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Practice Variation
• Variability in care not explained by patient or contextual factors • Up to 85% variation in care (Millenson, Health Aff 1997) – Provider training – Provider knowledge base and experience – Local culture • Treatment variation: Does it translate to variation in clinical outcomes?
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 370 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
What Matters During a Hypotensive Event? Fluids, Vasopressors, or Both?
Kothari R, Lee J, Ladapo J, Celi LA
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 371 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Practice Variation
• Hypotension in the ICU: assess fluid responsiveness and optimize cardiac preload , + vasopressors • Variable opinion among clinicians as regards harm from excess fluid and risk of vasopressor use
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 372 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Methods
• Definition of hypotensive episode • Interventions: fluid rate, use of vasopressors • Primary outcomes: Mortality • Secondary outcomes – Duration of hypotensive episode – ICU length-of –stay – Rise in creatinine within 3 days after the hypotensive event
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 373 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Methods
• Control variables or confounders: – SAPS – Average MAP 3 hours prior to the hypotensive event – Minimum MAP during the hypotensive event – Average MAP during the hypotensive event • Multivariate regression analysis • Propensity score analysis: pressors vs. mortality
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 374 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Results
Table 1. Interventions given during HE according to ICU type
Interventions Given During HE According to ICU Type
MICU SICU CCU Total
Fluids only 69 (26%) 115 (31%) 25 (18%) 209 (27%)
Pressors only 147 (54%) 171 (46%) 82 (61%) 400 (51%)
Fluids & Pressors 54 (20%) 87 (23%) 28 (21%) 169 (22%)
Total 270 373 135 778 www.bsscommunitycollege.inWWW.BSSVE.INImage by MIT OpenCourseWare. www.bssnewgeneration.in Adapted from www.bsslifeskillscollege.in upcoming publication by Leo Anthony Celi. 375 www.onlineeducation.bharatsevaksamaj.net Results www.bssskillmission.in Table 2. Type of vasopressor used according to ICU type
Type of Vasopressor Used According to ICU Type
MICU SICU CCU Total
Dobutamine 5 (2%) 4 (2%) 8 (7%) 17 (3%)
Dopamine 50 (25%) 31 (12%) 52 (47%) 133 (23%)
Epinephrine 2 (1%) 2 (1%) 4 (4%) 8 (1%)
Norepinephrine (56%) 133 (52%) 47 (43%) 293 (51%)113
Phenylephrine (34%) 120 (47%) 30 (27%) 219 (38%)69
Vasopressin (6%) 9 (3%) 10 (9%) 31 (5%)12
Total patients 201 258 110 569
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in WWW.BSSVE.INImage by MIT OpenCourseWare. Adapted from upcoming publication by Leo Anthony Celi. 376 www.onlineeducation.bharatsevaksamaj.net Results www.bssskillmission.in Figure 1. Fluid rate during hypotensive event
Fluid Rate During Hypotensive Event 60
50
40
30
Number of patients 20
10
0 0 1000 2000 3000 4000 5000 Fluid rate (ml/h) www.bsscommunitycollege.inWWW.BSSVE.INImage by MIT OpenCourseWare. www.bssnewgeneration.in Adapted from upcoming www.bsslifeskillscollege.in publication by Leo Anthony Celi. 377 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Results
Table 3. Multivariate analysis for HE duration (N=730, Hosmer-Lemeshow p=0.906)
Odds Ratio 95% CI P Value Fluid rate < 500 ml/hr but > 250 ml/hr 1.261 0.803-1.981 0.314 Fluid rate > 500 ml/hr 0.876 0.562-1.366 0.560 Vasopressor use 0.444 0.818-2.532 < 10-5 Average MAP prior to HE 0.978 0.310-0.635 0.002 SAPS 1.018 0.965-0.992 0.214 SICU (vs. MICU) 0.600 0.428-0.842 0.003 CCU (vs. MICU) 0.686 0.442-1.065 0.093
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 378 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Results
Table 4. Multivariate analysis for hospital mortality (N=730, Hosmer-Lemeshow p=0.678) Odds Ratio 95% CI P Value Fluid rate < 500 ml/hr but > 250 ml/hr 1.057 0.666-1.679 0.813 Fluid rate > 500 ml/hr 0.647 0.408-1.028 0.065 Vasopressor use 1.934 1.340-2.791 < 10-3 Average MAP prior to HE 0.985 0.971-0.999 0.03 Average MAP during HE 1.005 0.973-1.038 0.768 Minimum MAP during HE 0.997 0.970-1.024 0.821 SAPS 1.121 1.086-1.158 < 10-11 SICU (vs. MICU) 0.670 0.473-0.949 0.024 CCU (vs. MICU) 0.636 0.403-1.005 0.052
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 379 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Results
Table 5. Propensity score model (N=730, Hosmer-Lemeshow p=0.845)
Odds Ratio 95% CI P Value Fluid rate < 500 ml/hr but > 250 ml/hr 0.217 0.139-0.338 < 10-10 Fluid rate > 500 ml/hr 0.333 0.211-0.526 < 10-5 Average MAP prior to HE 1.011 0.995-1.027 0.166 SAPS 1.050 1.015-1.086 <0.005 SICU (vs. MICU) 0.750 0.511-1.100 0.141 CCU (vs. MICU) 1.375 0.789-2.394 0.261
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 380 www.onlineeducation.bharatsevaksamaj.net Results www.bssskillmission.in Figure 3. Calibration of the propensity score model
Calibration of the Propensity Score Model 1
0.9
0.8
0.7
0.6
0.5 Proportion with pressor administration
0.4
12345678910 Pressor propensity deciles
Expected Observed www.bsscommunitycollege.inWWW.BSSVE.INImage by MIT OpenCourseWare. www.bssnewgeneration.in Adapted from upcoming publication www.bsslifeskillscollege.in by Leo Anthony Celi. 381 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Results
Table 6. Vasopressor use vs. hospital mortality after adjustment for propensity score (N=730, Hosmer-Lemeshow p=0.345)
Odds Ratio 95% CI P Value Vasopressor use 1.820 1.282-2.584 0.001 Propensity score 4.858 1.670-14.131 0.004
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 382 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Results Table 4. Multivariate analysis for ICU length-of-stay among survivors (N=347, Hosmer-Lemeshow p=0.291) Odds Ratio 95% CI P Value Fluid rate < 500 ml/hr but > 250 ml/hr 1.000 0.432-2.314 1.000 Fluid rate > 500 ml/hr 2.957 0.836-10.453 0.092 Vasopressor use 1.490 0.743-2.987 0.262 Average MAP prior to HE 1.013 0.982-1.044 0.424 Average MAP during HE 0.953 0.888-1.023 0.185 Minimum MAP during HE 0.988 0.923-1.058 0.726 SAPS 1.125 1.043-1.213 0.002 SICU (vs. MICU) 1.082 0.517-2.263 0.835 CCU (vs. MICU) 1.95 0.673-5.651 0.218
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 383 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Results
Table 4. Multivariate analysis for creatinine rise (N=618, Hosmer-Lemeshow p=0.745) Odds Ratio 95% CI P Value Fluid rate < 500 ml/hr but > 250 ml/hr 0.734 0.455-1.185 0.206 Fluid rate > 500 ml/hr 0.744 0.457-1.210 0.233 Vasopressor use 1.060 0.725-1.550 0.763 Average MAP prior to HE 0.992 0.997-1.007 0.281 Average MAP during HE 0.984 0.951-1.019 0.365 Minimum MAP during HE 0.974 0.945-1.003 0.077 SAPS 1.030 0.998-1.064 0.068 SICU (vs. MICU) 0.870 0.606-1.251 0.453 CCU (vs. MICU) 1.072 0.667-1.724 0.773
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 384 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Discussion
• Vasopressor use during a hypotensive event is an independent predictor of mortality – Multivariate logistic regression – Propensity score analysis • Mean vasopressor load associated with increased risk of 28-day mortality (Dunser, Crit Care 2009) • Side effects – impaired microcirculation – increased metabolic demands – altered immune response
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 385 www.onlineeducation.bharatsevaksamaj.netIncorporating Dynamic www.bssskillmission.in Information during a Hypotensive Episode to Improve Mortality Prediction Mayaud L, Celi LA, Kothari R, Clifford G, Tarrasenko L, Annane D
Observation Pre hypo Post ABP (mmHg) sys-mean-dia 60
Pseudo-continous variables
Discrete variables
Ton-24h Ton-2h Ton Toff Toff+2h Toff+24h Time (min) Lab values Heamodynamics Heamodynamics Patient's data 24H window 2H window fluids - pressors
Image by MIT OpenCourseWare. Adapted from Mayaud, et al.
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 386 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Hypotensive Episode
Treatments: Fluids Vasopressors
Physiologic Response to Treatments
Images by MIT OpenCourseWare. Images by MIT OpenCourseWare. Initial Outcome Event -> Treatment -> Response Presentation www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in Prediction 387 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Transfusing the Non-Bleeding Patient Samani S, Samani Z, Malley B, Celi LA • Compare survival curves of transfused and non-transfused non-bleeding patients with hemoglobin between 7 and 10 g/dL • Control variables: age, severity score, co- morbidities, hemoglobin • Cox regression model to calculate hazards ratio • Propensity score analysis and instrumental variable analysis to confirm findings
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 388 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Impact of 24/7 Intensivist on Clinical Outcomes Celi LA, Stevens J, Lee J, Osorio J, Howell M
• Nocturnal intensivist program initiated in MICU in 2002, SICU in 2010 • Control for potential confounding by other ICU quality improvement projects by comparing adjusted clinical outcomes of MICU and SICU patients • Perform analysis on patients admitted at night as day admissions may dilute treatment effect
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 389 www.onlineeducation.bharatsevaksamaj.netQuantifying the www.bssskillmission.in Risk of Unnecessary Broad-Spectrum Antibiotics Snyder G, Pho M, Golik M, Celi LA • Antibiotic use is the main driver of antimicrobial resistance in the hospital • Vancomycin/Cefepime for every healthcare facility-associated fever & leukocytosis • Streamlining rarely happens despite negative cultures • Difficult to distinguish infectious vs. non- infectious SIRS
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 390 www.onlineeducation.bharatsevaksamaj.netPredicting Whether www.bssskillmission.in a Laboratory Test will be Significantly Changed from the Previous Determination Cismondi F , Celi LA • Frequency of laboratory testing very ad hoc – Hematocrits for GI bleed – Chem 7 for Hyperglycemic Hyperosmolar State, DKA – ABG for status asthmaticus • Can we predict whether a test will give us additional information? • Reduce iatrogenic anemia, false positives
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 391 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Other Works in Progress • Developing mortality prediction models for elderly patients undergoing open heart surgery • Cost effectiveness of CABG vs. PCI among elderly patients • Looking at coupling/uncoupling of physiologic variables using information transfer among different patient subsets • Influence of MELD scores on Kaplan-Meier curves among patients with cirrhosis admitted to the ICU • Impact of troponin leaks during critical illness on long- term survival • Epidemiology of rash in the ICU • Are there racial disparities in resource utilization at the end-of-life at BIDMC?
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 392 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Conclusions
• Clinical databases such as MIMIC present an opportunity to study areas where practice variation exists • Large-scale evidence impossible to obtain for the millions of questions posed in day-to-day practice - impractical, expensive, “unethical” • Data mining might allow us to catch-up with a century of non-evidence-based medicine
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 393 www.onlineeducation.bharatsevaksamaj.netThe MIMIC www.bssskillmission.in Vision
ICU
Medicine for Dummies Database
Images by MIT OpenCourseWare. Select patients similar in important features as Build regards a specific question, model e.g. Will my patient benefit from blood transfusion?
“Our vision is the creation of a learning system that aggregates and analyzes day-to-day experimentations, where new knowledge is constantly extracted and propagated, and where practice is driven by outcomes, and less so by heuristics and www.bsscommunitycollege.ingutWWW.BSSVE.IN instinct.” www.bssnewgeneration.in www.bsslifeskillscollege.in 394 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in MIT OpenCourseWare http://ocw.mit.edu
HST.950J / 6.872 Biomedical Computing Fall 2010
For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 395 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Global Health Informatics
Hamish SF Fraser
Director of informatics, Partners In Health Assistant professor, Harvard medical School & Division of Global Health Equity, BWH
Logos of Brigham and Women's Hospital and Harvard Medical School have been removedwww.bsscommunitycollege.inWWW.BSSVE.IN from all slides due to copyright www.bssnewgeneration.in restrictions. www.bsslifeskillscollege.in 396 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Overview
• Challenges and opportunities for global health informatics (eHealth) • The PIH-EMR system in Peru • The background for OpenMRS • The OpenMRS platform • Evaluation of medical information systems • Systematic review of evaluations
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 397 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Typical problem:
• Can HIV and MDR-TB care be delivered 1. In settings with limited or absent infrastructure? 2. To thousands or tens of thousands of patients? 3. Over long periods of time? 4. With outcomes equivalent to ARV treatment in the US? 5. At a “manageable” cost?
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 398 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Status of Global Health Informatics
• Rapid development over the last 2 years – Bellagio meeting on e-Health in July 2008
• Driven by the coincidence of: – need for better Global Health Delivery – increased resources for health system strengthening such as the Global Fund – more effective, robust, low-cost technologies
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 399 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Partners In Health Model of Care
• Access to health care for all people • Creation of long-term development by partnering with local people and communities • Use of community health workers to grow a local and sustainable work force • Addressing the effects of poverty including poor nutrition, water, and housing • Drawing on the resources of the world’s elite medical and academic institutions and on the lived experience of the world’s poorest and sickest communities www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 400 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Directly observed therapy in Haiti www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 401 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Chronic disease management
• Identifying patients requiring treatment • Starting patients on the correct medication • Ensuring stable and economical supply of medication • Ensuring compliance with treatment • Monitoring treatment progress and outcomes and addressing adverse events promptly
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 402 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Core Functions of e-Health Systems
• Clinical care and quality improvement
• Monitoring and reporting
• Drug supply management
• Research
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 403 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Example: MDR-TB in Lima, Peru
• Highest incidence of TB in South America • 40,000 patients treated with DOTS per year • > 3% have MDR-TB • Require up to 9 drugs to treat MDR-TB
DOTS = directly observed therapy short course www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 404 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in PIH-EMR System in Peru
• Secure web-based EMR • Operational since 2001 • Usable with low-speed dialup connections • Bilingual (Spanish/English) • 50,000+ patients tracked • 13,000 patients treated for MDR-TB
Fraser HSF, et al. Evaluating the impact and costs of deploying an electronic medical record system www.bsscommunitycollege.intoWWW.BSSVE.IN support TB treatment inwww.bssnewgeneration.in Peru. Proc AMIA Symp www.bsslifeskillscollege.in 2006: 264-268 405 www.onlineeducation.bharatsevaksamaj.netPIH-EMR www.bssskillmission.in Data
Registration form History/exam Smears Biochem. Previous Rx Drug regimens Cultures Hematology Previous Dx Follow up Pharmacy Drug sensitivity Contacts Chest X-ray (DST) www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 406 www.onlineeducation.bharatsevaksamaj.netRequirements forwww.bssskillmission.in general purpose medical record system
• Simple to setup • Multiple computing platforms • Local users can create EMR forms and reports • Web based (but can also be run locally) • Open standards - HL7, LOINC, SNOMED, ICD10 • Fully open source – supported by a community of programmers – using best ideas and software from many projects
• Able to be setup, modified and owned by the countries where we work, not just a “present from the US” but a full transfer of technology, skills and ownership
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 407 www.onlineeducation.bharatsevaksamaj.netOpenMRS: www.bssskillmission.in a modular, open source, EMR platform
• Developed as a collaboration of PIH, the Regenstrief Institute and South African MRC • Uses concept dictionary for data storage • Modular design simplifies adding new functions and linking to other systems • Supports multiple languages • Released with open source license (April 2007) • Core of paid programmers with growing community support • www.openmrs.org
Partners In Health Regenstrief Institute Medical reseach council SA www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 408 www.onlineeducation.bharatsevaksamaj.netThe concept www.bssskillmission.in dictionary
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 409 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in OpenMRS sites - fall 2010
Image of Google Maps showing locations of OpenMRS sites in Uganda, Congo, Kenya, Tanzania, and Malawi, has been removed due to copyright restrictions.
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 410 www.onlineeducation.bharatsevaksamaj.netRwanda www.bssskillmission.inhealth indicators
• A small central African country: – Population 9 M people – Highest population density in Africa, 85% rural
• Achieved rapid economic growth since genocide in 1994, but still has very poor health outcomes: – Life expectancy 38-44 years – Infant mortality 152/1000 – Maternal mortality 1071/100K – Medium income $230 – HIV prevalence 3% – Malaria prevalence 46% www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 411 www.onlineeducation.bharatsevaksamaj.netOpenMRS at www.bssskillmission.inPIH sites in Rwanda
• Currently used for 12 PIH – supported health centers • Data for patients with HIV, TB and now heart failure • Over 10,000 patients tracked (Sept. 2009) • Team of Rwandan data officers trained to enter data, ensure quality & produce reports • Clinicians lookup of electronic patient summaries • 8 sites have their own server, 6 remote sites maintain a synchronized copy of the entire database • Many new research and clinical applications • Primary care version is under development www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 412 www.onlineeducation.bharatsevaksamaj.netPhysician looking www.bssskillmission.in up ARV patient
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 413 www.onlineeducation.bharatsevaksamaj.netPotential components www.bssskillmission.in of integrated national eHealth architecture in Rwanda
National Supply chain reporting system systems TRACNet Camerwa
SDMX-HD
Pharmacy HL7 HL7 Laboratory EMR System system System OpenMRS PIH PIH-Lab-system
HL7 Dicom HL7? HL7 Registration and Mobile health Radiology / insurance systems telemedicine Mutuelle OpenROSA system www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 414 www.onlineeducation.bharatsevaksamaj.netRelationship between www.bssskillmission.in facilities for reporting in Rwanda
Village District Clinic MOH CHW
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 415 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Government of Rwanda EMR roll out
• The Government of Rwanda is committed to having a strong national EMR program • MoH has announced that OpenMRS will be used for the national roll out to health centers and small hospitals • MoH wants a non-disease specific system which: – Can assist in the management of all outpatients – Will also continue to be used for HIV management • Detailed rollout plan being developed at present
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 416 www.onlineeducation.bharatsevaksamaj.netDeveloper training, www.bssskillmission.in Rwanda
• We are running a training program in Kigali for computer science graduates • One year, mentored training course – Web development – Java programming – OpenMRS programming – Medical informatics • Ten students graduated last week • They will support OpenMRS rollout as well as building software development capacity in Rwanda
International Development Research Center www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 417 www.onlineeducation.bharatsevaksamaj.netCommunity: www.bssskillmission.in OpenMRS Wiki
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 418 www.onlineeducation.bharatsevaksamaj.netDisease-specific www.bssskillmission.in EMR (MDR-TB)
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 419 www.onlineeducation.bharatsevaksamaj.netPrevious drug www.bssskillmission.in prescriptions and decision support tools
Alternativewww.bsscommunitycollege.inWWW.BSSVE.IN alerts and warningswww.bssnewgeneration.in view www.bsslifeskillscollege.in 420 www.onlineeducation.bharatsevaksamaj.netMDR-TB treatment www.bssskillmission.in history flowsheet
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 421 www.onlineeducation.bharatsevaksamaj.netOpenMRS-Google www.bssskillmission.in Maps–SMS-Integration, Karachi
Image of Google Maps mash-up with OpenMRS has been removed due to copyright restrictions.
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in WWW.BSSVE.INCredit: Owais Ahmed, Aamir Khan 422 www.onlineeducation.bharatsevaksamaj.netTB in homeless patients www.bssskillmission.in in Los Angeles
Credit: Monica Waggoner
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 423 www.onlineeducation.bharatsevaksamaj.netResearch Data www.bssskillmission.in Coordination
Courtesy of Dave Thomas. Used with permission.
Credit: Dave Thomas www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 424 www.onlineeducation.bharatsevaksamaj.netAdaptive Turnaround www.bssskillmission.in Documents
Credit: Child Health Informatics Research and Development Lab Credit: Vibha Anand, Paul Biondich (CHIRDL) and Children's Health Services Research (CHSR) Program. (Regenstrief)www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.inCourtesy www.bsslifeskillscollege.in of Vibha Anand, Paul Biondich. Used with permission. 425 www.onlineeducation.bharatsevaksamaj.netTesting touch screen www.bssskillmission.in patient registration in Rwinkwavu, Rwanda
© The Rockefeller Foundation. All rights reserved. This content is excluded from our Creative Commons license. For more information, see http://ocw.mit.edu/fairuse.
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 426 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Synchronization
• We created a new component to allow bi- directional synchronization between OpenMRS instances • Uses limited internet capability (soon to be usable with USB memory stick) • 6 sites in Rwanda are now synchronizing • Working on a general version, requires modification to the data model
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 427 www.onlineeducation.bharatsevaksamaj.netSecurity and www.bssskillmission.in confidentiality of medical data • Patient data is highly sensitive in all countries – HIV in Africa a key example • We encrypt data transfers with SSL • Staff receive training in patient data and security management • All logins and page views can be audited • Government policy on health data ownership and control are required
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 428 www.onlineeducation.bharatsevaksamaj.netChallenges www.bssskillmission.in for OpenMRS deployments
• Reliability and support for equipment, power supplies and software
• Training
• Data management and quality control
• Evaluation
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 429 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Technical challenges • Online-offline data use and synchronization • Building complex applications with modular architectures • Rapid data entry from clinical staff • Simple drug order entry • Reporting from EAV data models
We welcome opportunities to share the work of building open and interoperable systems and expanding collaboration. www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 430 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Evaluation of Global Health Informatics Projects
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 431 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Why Evaluate?
• Quality of care
• Efficiency and economics
• Evidence based medicine
• Advance the science of Medical Informatics
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 432 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in 5 Levels of Evaluation (Stead et al)
1. Problem definition 2. Bench testing 3. Field trials: observational 4. Field trials: interventional 5. Long term follow-up
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 433 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Evaluation Types
• Formative Evaluation 1. Determine important functionality of and improve system 2. More qualitative methods 3. Usually performed by implementers
• Summative Evaluation 1. Determine benefits and sustainability of system 2. More quantitative methods 3. Usually performed by outside researchers
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 434 www.onlineeducation.bharatsevaksamaj.netPDA Data www.bssskillmission.in Management Collecting lab data in sites without internet
Processing Section Sync through local PC Processing & Verification
clinical Bacteriology Section Palm Pilot PIH-EMR www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 435 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Palm Project: Study Design
Controlled study
• (A) Prospective
• (B) Historical
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 436 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Palm Project: Study Results
Median processing time Frequency of Errors
Intervention Control * Districts Districts days (n) days (n) ‡ Pre-Palm 30.5 (4876)* 30.8 (5954) *‡
Post-Palm 7.7 (2890)*† 22.7 (3263)†
* p<0.001 * p < 0.001 † p<0.001 ‡ p = 0.055
Joaquinwww.bsscommunitycollege.inWWW.BSSVE.IN Blaya, PhD, www.bssnewgeneration.in Harvard-MIT www.bsslifeskillscollege.in HST program 437 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Palm Project: Study Results
• Work Efficiency – 66% reduction in collection and processing time
• Users’ Preference – All users wanted to end study and expand use of system – All users felt system was perceived positively by health center personnel – Cost of moving system to new sites
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 438 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in PDA system, cost analysis
• The total cost and time to develop and implement the intervention - US$26,092 and 22 weeks (add on to EMR). • The cost to extend the system to cover nine more districts - $1125 • Cost to implement collecting patient weights - $4107.
Blaya etwww.bsscommunitycollege.inWWW.BSSVE.IN al, INT J TUBERC www.bssnewgeneration.in LUNG DIS 12(8):921 www.bsslifeskillscollege.in–927 439 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Drug Sensitivity Lab Data Flow
Baseline problems with DST data
- 10% of results took > 60 days to arrive at clinic
- 16% of patients waited > 100 days to start treatment
- (17%) of DSTs were duplicates
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.inYagu www.bsslifeskillscollege.ini et al. Int J Lung Dis 2006 440 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in E-CHASQUI logo removed due to copyright restrictions. Laboratory Reporting System 1. Connects laboratories to health centers – Email notifications to health center personnel 2. Tools to improve data quality 3. Reporting functions for laboratory personnel cluster randomized controlled trial of 1846 patients recently completed
Joaquinwww.bsscommunitycollege.in WWW.BSSVE.INBlaya, PhD student, www.bssnewgeneration.in Harvard-MIT www.bsslifeskillscollege.in HST program 441 www.onlineeducation.bharatsevaksamaj.neteChasqui study www.bssskillmission.in results: error rates • Intervention HCs showed: – 82% less errors compared to controls in reporting for drug susceptibility tests (2.1 vs. 11.9%, p<0.001) – 87% fewer errors compared to controls for cultures (2.0 vs. 15.1%, p<0.001) • eChasqui allowed missing results to be viewed online: – these accounted for at least 72% of all errors • 66% of control and 55% of intervention HC users responded they were missing at least 10% of paper results Blaya J etwww.bsscommunitycollege.inWWW.BSSVE.IN al, Int J Tuberc Lung www.bssnewgeneration.in Dis. 2010 Aug;14(8):1009- www.bsslifeskillscollege.in15 442 www.onlineeducation.bharatsevaksamaj.netMDR-TB Drug www.bssskillmission.in Regimen Design
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 443 www.onlineeducation.bharatsevaksamaj.netEvaluation of impact www.bssskillmission.in of order entry system on drug data accuracy
• Prospective and historical controlled study • Drug regimen quality and timeliness were surveyed in two districts in Lima, Peru • Drug errors per patient
Callao (EMR) Lima Este (control) Before 17.4%* 8.6%** After 3.1%* 6.9%** *P= 0.0075 **P= 0.66,
Choi S, www.bsscommunitycollege.inetWWW.BSSVE.IN al. Proc. Medinfo2004, www.bssnewgeneration.in 11: www.bsslifeskillscollege.in 202-206 444 www.onlineeducation.bharatsevaksamaj.netStock www.bssskillmission.in Card
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 445 www.onlineeducation.bharatsevaksamaj.netPharmacy andwww.bssskillmission.in Warehouse Stock Tracking
Reduction in product-days of stocked out medication (daily report – a method of triangulation)
System was set up in 2005 but scaled in 2006.
Q1 2006 Q4 2006 Prod. Days stocked out 1569 634 (P<0.001)
Prod. Days 60,608 58,576 2.6% 1.1%
Berger L, www.bsscommunitycollege.in WWW.BSSVE.INet al, Proc. AMIA SYMP. www.bssnewgeneration.in 2007:46-50 www.bsslifeskillscollege.in 446 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in “Stop the Stock-outs”, Kenya
• Led by Health Action International, Oxfam and local civil society organizations • “Stop the Stock-outs” used a system developed by Frontline SMS • Patients to send text messages to a server if the drug they had been prescribed was stocked out at the clinic’s pharmacy • Data is linked to mapping software
http://www.scidev.net/en/news/software-allows-public-to-map-medicine-shortages-.html www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 447 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in “Stop the Stock-outs”
• The group was able to map the levels of stockouts of essential medications in more than 100 clinics in Kenya • Stockouts rates of 50-60% were documented for essential medications • This data was publicized and led to the Kenya parliament voting for increased funding for drug supply – The system is also being used in Malawi, Zimbabwe and Uganda www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 448 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Supporting HIV treatment
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 449 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Consequences of Inadequate Patient Tracking for PMTCT and ARV programs
“Thus, 12 months after delivery, only a fraction (19% in one study in Malawi) of HIV positive mothers who received antiretroviral drugs will attend health services to have their infant tested for HIV.” “Clearly, this may have lethal consequences for those children who become HIV positive.” (Reithinger et al, BMJ June 1st 2007)
A review in 2007 of adult HIV treatment programs in Africa estimated that only 61% of patients were still in care 2 years after starting treatment. (Other studies suggest ~85%) (Rosen S, PLoS Med 2007 Oct 16;4(10):e298) www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 450 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Clinical Alerts (Rwinkwavu, Rwanda)
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 451 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in CD4 Access, Rwinkwavu, Rwanda
• We evaluated whether the ID physicians had access to the latest CD4 count for their patients in Rwinkwavu, Rwanda • The physicians record the result they have on the follow-up form based on paper lab result forms • We checked if they were up to date before and after a new lab component was added to the EMR to generate results forms
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 452 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Results – Access to CD4 counts
• The proportion of CD4 counts conducted within the past 60 days but unknown to the clinician at the time of consultation was: • 24.7% in the pre-intervention period • 16.7% in the post intervention period • This is a 32.4% reduction in CD4 loss (p=.002) • We are now extending direct clinician access to the EMR
Amorosowww.bsscommunitycollege.inWWW.BSSVE.IN et al, Stud Health www.bssnewgeneration.in Technol Inform. www.bsslifeskillscollege.in 2010; : 453 www.onlineeducation.bharatsevaksamaj.netEvaluation www.bssskillmission.in 4: Physician looking up ARV patients
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 454 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 455 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 456 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 457 www.onlineeducation.bharatsevaksamaj.netEvaluation of PDA www.bssskillmission.in system for Home Based Care at AMPATH in Kenya
• Developed a Palm Pilot PDA application, very similar technology to Peru TB study • Data collected: – patient registration, HIV testing, TB screening, maternal care, vaccinations • Reported data on 14,648 households, 40,111 patients, mean of 12 new patient records per day • 899 (45%) pregnant women not receiving AN care • 693/1131 (61.3%) HIV+ patients never been tested • User satisfaction was high, technical issues rare • Cost to cover 2 million patients, $0.15/patient
Were etwww.bsscommunitycollege.inWWW.BSSVE.IN al, Stud Health Technol www.bssnewgeneration.in Inform. 2010;160:525- www.bsslifeskillscollege.in9. 458 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Broader evaluation perspectives
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 459 www.onlineeducation.bharatsevaksamaj.netAn evaluation www.bssskillmission.inof the District Health Information System in rural South Africa
• Outcomes: assessed data quality, the utilisation for facility management, perceptions of work burden, and usefulness of the system to clinic staff.
• Results. A high perceived work burden associated with data collection and collation • Some data collation tools were not used as intended. • There was good understanding of the data collection and collation process but little analysis, interpretation or utilisation of data. • Feedback to clinics occurred rarely
A Garrib,www.bsscommunitycollege.in WWW.BSSVE.INet al , SAMJ , Vol. www.bssnewgeneration.in 98, No. 7, p 549- www.bsslifeskillscollege.in552 460 www.onlineeducation.bharatsevaksamaj.net DHIS www.bssskillmission.in
• In the 10 clinics, 2.5% of data values were missing, and 25% of data were outside expected ranges without an explanation provided. • There was no computerisation of data collection and no facility for electronic submission of data in any clinic. • Clinic staff and supervisors reported that even if the data did not look correct, checking it was rarely done due to lack of time. • Little analysis of data occurred at the clinic or by clinic supervisors. • Data were not discussed in staff meetings nor analysed by them.
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 461 www.onlineeducation.bharatsevaksamaj.netMalawi Patient Management www.bssskillmission.in System (Baobab)
• Touch screen data entry system • Low cost, robust flat screen terminals • Large numbers of patients registered (>300,000) • May be best example of direct data entry system in a developing country
Reportwww.bsscommunitycollege.inWWW.BSSVE.IN CDC Malawi, www.bssnewgeneration.in presented www.bsslifeskillscollege.in at PHIN2009 462 www.onlineeducation.bharatsevaksamaj.net“Mateme” Touchscreen www.bssskillmission.in Registration
Credit: Jeff Rafter (Baobab), Evan Waters (PIH)
Courtesy Jeff Rafter and Evan Waters. Used with permission. www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 463 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in CDC Pilot Study: Objectives
• The pilot EDS will be evaluated using a set of criteria: – Usability – Sustainability – Reliability – Availability – Accessibility – Maintainability – Deployability
• Impact of the introduction of the EDS being assessed at multiple user levels – Clinician – Health facility – MOH www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 464 www.onlineeducation.bharatsevaksamaj.netSystematic www.bssskillmission.in review of evaluation studies
Blaya, Fraser, Holt, Health Affairs 2010, 29;2: 244-251 Surveyed 2043 articles and reports Used 45 in final analysis
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 465 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Summary of the Key Studies
eHealth Category Qualitative Quantitative Descriptive Controlled Studies Studies Electronic Health Record (EHR) 5 1 5 Laboratory Information Management Systems (LIMS) 0 1 2 Pharmacy Information Systems 4 2 3 Patient Registration or Scheduling Systems 1 0 2 Monitoring, Evaluation and Patient Tracking Systems 0 2 4 Clinical Decision Support Systems (CDSS) 1 0 3 Patient Reminder Systems 0 1 3 Research or Data Collection Systems 5 1 11 TOTAL 15 8 32
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 466 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Findings of the Review
Key functions supported by “initial” evidence:
• Tracking patients through treatment initiation, monitoring adherence, and detecting those at risk for loss to follow-up
• Decreasing time to create administrative reports
• Tools to label or register samples and patients
• Collection of clinical or research data using PDAs
• Reduction in errors in laboratory and medication data
• Reminding patients of health care actions
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 467 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in What has been invested in eHealth?
• Recent world bank study showed that over $480M has been awarded to ehealth projects by World Bank for current projects • 3 other major development agencies also funding at high levels: – USAID – PEPFAR – GFATM • Little if any evaluation has been carried out on those projects
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 468 www.onlineeducation.bharatsevaksamaj.netCollaborators www.bssskillmission.in and Funders • Partners In Health • Regenstrief institute • Medical Research Council, South Africa • World Health Organization • US Centers for Disease Control • Brigham and Women hospital • Harvard Medical School • University of KwaZulu-Natal • Millennium Villages Project • International Development Research Centre, Ottawa • Rockefeller Foundation • Fogarty International Center, NIH • Boston Consulting Group • Google Inc www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 469 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Patient Tracking Patient Set
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 470 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in PIH-EMR history • 2001 web based EMR system to support the scale up of MDR-TB treatment in Peru • 2003 created a version of PIH-EMR to support HIV treatment in rural Haiti • 2004 made the decision to create a new, general and flexible platform to build EMR systems for developing countries • OpenMRS first used in early 2006 in Kenya and then Rwanda and South Africa
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 471 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Methods used in Malawi
• Surveys, semi-structured interviews with system users, including facility level health care workers and central level staff involved in M&E/supervision. • Time-flow analyses (pre- and post-introduction of system) • Analysis of information entered onto patient master cards and into the electronic system to assess the accuracy of information entered. • Technical review of system • System logs of problems (e.g. power or system outages, etc.)
Found that 70% of clinicians preferred the touch screen system to the paper system
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 472 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in South Africa (HUPA) Study HIV Counselors ask a series of questions leading to a patient assessment.
Courtesywww.bsscommunitycollege.inWWW.BSSVE.IN Neal Lesh www.bssnewgeneration.in www.bsslifeskillscollege.inCourtesy of Neal Lesh. Used with permission. 473 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in MIT OpenCourseWare http://ocw.mit.edu
HST.950J / 6.872 Biomedical Computing Fall 2010
For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.
www.bsscommunitycollege.inWWW.BSSVE.IN www.bssnewgeneration.in www.bsslifeskillscollege.in 474 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Learning Bayes Networks 6.034 Based on Russell & Norvig, Artificial Intelligence:A Modern Approach, 2nd ed., 2003 and D. Heckerman. A Tutorial on Learning with Bayesian Networks. In Learning in Graphical Models, M. Jordan, ed.. MIT Press, Cambridge, MA, 1999. WWW.BSSVE.IN
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 475 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Statistical Learning Task
• Given a set of observations (evidence), • find {any/good/best} hypothesis that describes the domain • and can predict the data • and, we hope, data not yet seen • ML section of course introduced various learning methods • nearest neighbors, decision (classification) trees, naive Bayes WWW.BSSVE.INclassifiers, perceptrons, ... • Here we introduce methods that learn (non-naive) Bayes networks, which can exhibit more systematic structure
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 476 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Characteristics of Learning BN Models
• Benefits • Handle incomplete data • Can model causal chains of relationships • Combine domain knowledge and data • Can avoid overfitting • Two main uses: WWW.BSSVE.IN• Find (best) hypothesis that accounts for a body of data • Find a probability distribution over hypotheses that permits us to predict/interpret future data
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 477 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
An Example
• Surprise Candy Corp. makes two flavors of candy: cherry and lime • Both flavors come in the same opaque wrapper • Candy is sold in large bags, which have one of the following distributions of flavors, but are visually indistinguishable: • h1: 100% cherry • h2: 75% cherry, 25% lime • h3: 50% cherry, 50% lime h : 25% cherry, 75% lime WWW.BSSVE.IN• 4 • h5: 100% lime • Relative prevalence of these types of bags is (.1, .2, .4, .2, .1) • As we eat our way through a bag of candy, predict the flavor of the next piece; actually a probability distribution.
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 478 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Bayesian Learning
• Calculate the probability of each hypothesis given the data • To predict the probability distribution over an unknown quantity, X, • If the observations d are independent, then • E.g., suppose the first 10 candies we taste are all lime WWW.BSSVE.IN
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 479 www.onlineeducation.bharatsevaksamaj.net Learning www.bssskillmission.in Hypotheses and Predicting from Them
• (a) probabilities of hi after k lime candies; (b) prob. of next lime
a b 1 1
0.9 0.8 0.8 0.6 0.7 0.4 0.6
0.2 0.5 Probability that next candy is lime Posterior probability of hypothesis 0 0.4 0 2 4 6 8 10 0 2 4 6 8 10 WWW.BSSVE.INNumber of samples in d Number of samples in d P(h1 | d) P(h2 | d) P(h3 | d) P(h4 | d) P(h5 | d)
Images by MIT OpenCourseWare. • MAP prediction: predict just from most probable hypothesis • After 3 limes, h5 is most probable, hence we predict lime Even though, by (b), it’s only 80% probable • www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 480 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Observations
• Bayesian approach asks for prior probabilities on hypotheses! • Natural way to encode bias against complex hypotheses: make their prior probability very low • Choosing hMAP to maximize • is equivalent to minimizing • but from our earlier discussion of entropy as a measure of information, these two terms are • # of bits needed to describe the data given hypothesis WWW.BSSVE.IN• # bits needed to specify the hypothesis • Thus, MAP learning chooses the hypothesis that maximizes compression of the data; Minimum Description Length principle • Assuming uniform priors on hypotheses makes MAP yield hML, the maximum likelihood hypothesis, which maximizes
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 481 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
ML Learning (Simplest)
• Surprise Candy Corp. is taken over by new management, who abandon their former bagging policies, but do continue to mix together θ cherry and (1-θ) lime candies in large bags • Their policy is now represented by a parameter θ ∈ [0,1], and we have a continuous set of hypotheses, hθ • Assuming we taste N candies, of which c are cherry and l=N–c lime • For convenience, we maximize the log likelihood WWW.BSSVE.IN Setting the derivative = 0, • P(F=cherry) θ • Surprise! Flavor • But need Laplace correction for small data sets
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 482 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
ML Parameter Learning • Suppose the new SCC management decides to give a hint of the candy flavor by (probabilistically) choosing wrapper colors
• Now we unwrap N candies of which c are cherries, with rc in red wrappers and gc in green, P(F=cherry) and l are limes, with rl in red wrappers and gl in green θ
WWW.BSSVE.IN Flavor
F P(W=red|F)
cherry θ1
lime θ2 • With complete data, ML learning decomposes into n learning problems, one for each parameter Wrapper
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 483 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Use BN to learn Parameters •If we extend BN to continuous variables (essentially, replace by ) •Then a BN showing the Parameter Independence dependence of the
observations on the θ θ1 θ2 parameters lets us compute (the distributions over) the parameters using Sample 1 F W P(F=cherry) just the “normal” rules of θ Bayesian inference. Sample 2 F W WWW.BSSVE.IN Flavor •This is efficient if all observations are known Sample 3 F W F P(W=red|F) •Need sampling ... cherry θ1 methods if not lime θ2 Sample N F W Wrapper
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 484 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Learning Structure
• In general, we are trying to determine not only parameters for a known structure but in fact which structure is best • (or the probability of each structure, so we can average over them to make a prediction) WWW.BSSVE.IN
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 485 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Structure Learning
• Recall that a Bayes Network is fully specified by • a DAG G that gives the (in)dependencies among variables • the collection of parameters θ that define the conditional probability tables for each of the • Then • We define the Bayesian score as • But • First term: usual marginal likelihood calculation WWW.BSSVE.IN• Second term: parameter priors • Third term: “penalty” for complexity of graph • Define a search problem over all possible graphs & parameters
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 486 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
X Y
Searching for Models X Y • How many possible DAGs are there for n variables? • = all possible directed graphs on n vars X Y • Not all are DAGs • To get a closer estimate, imagine that we order the variables so that the parents of each var come before it in the ordering.Then • there are n! possible ordering, and • the j-th var can have any of the previous vars as a parent
• If we can choose a particular ordering, say based on prior WWW.BSSVE.INknowledge, then we need consider “merely” models • If we restrict |Par(X)| to no more than k, consider models; this is actually practical • Search actions: add, delete, reverse an arc • Hill-climb on P(D|G) or on P(G|D) All “usual” tricks in search: simulated annealing, random restart, ... • www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 487 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Caution about Hidden Variables
• Suppose you are given a dataset containing data on patients’ smoking, diet, exercise, chest pain, fatigue, and shortness of breath • You would probably learn a model like the one below left • If you can hypothesize a “hidden” variable (not in the data set), e.g., heart disease, the learned network might be much simpler, such as the one below right • But, there are potentially infinitely many such variables
S D E S WWW.BSSVE.IND E
H C F B
C F B www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 488 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in 22 10 21 13 16 19 20 31 15 23 Re-Learning the ALARM 6 5 4 27 11 32 34 35 36 37 17 Network from 10,000 Samples 28 29 12 24 25 18 26 7 8 9 33 14 1 2 3 30
a) Original Network
22 10 21 13 case # x1 x2 x3 . . . x37 16 19 20 31 15 23 3 3 2 41 . . . 6 5 4 27 11 32 34 35 36 37 2 2 2 2 3 3 1 3 3 3 17 28 29 12 24 4 2 33 1
25 18 26 . . . . 7 8 9 33 14 1 2 3 10,000 22 2 3 30 c) Sampled Data b) Starting Network Complete independence WWW.BSSVE.IN 22 10 21 13 16 19 20 31 15 23
6 5 4 27 11 32 34 35 36 37 deleted 17 28 29 12 24 25 18 26 7 8 9 33 14 1 2 3 30
www.bsscommunitycollege.ind) www.bssnewgeneration.inLearned Network www.bsslifeskillscollege.in Images by MIT OpenCourseWare. 489 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
MIT OpenCourseWare http://ocw.mit.edu
HST.950J / 6.872 Biomedical Computing Fall 2010
For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.
WWW.BSSVE.IN
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 490 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Ethics, Privacy, etc. Peter Szolovits 6.872/HST.950
WWW.BSSVE.IN
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 491 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Treatment of Human Subjects:The Belmont Report 1979 Ethical Principles and Guidelines for the Protection of Human Subjects of Research
• Balancing (societal) benefits vs. (individual) risks • History of abuses • Nazi “experiments” ⇒ Nuremberg code WWW.BSSVE.IN• Tuskegee syphilis study
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 492 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Nazi Medical Experiments
• Freezing / Hypothermia • Genetics • Infectious Diseases • Interrogation and Torture • Killing / Genocide • High Altitude • WWW.BSSVE.INPharmacological • Sterilization • Surgery A cold water immersion experiment at Dachau concentration camp presided Traumatic Injuries over by Professor Ernst Holzlöhner (left) and Dr. Sigmund Rascher (right). The • subject is wearing an experimental Luftwaffe garment http://en.wikipedia.org/wiki/Nazi_human_experimentation
© unknown. All rights reserved. This content is excluded from our Creative www.bsscommunitycollege.in www.bssnewgeneration.inCommons www.bsslifeskillscollege.in license. For more information, see http://ocw.mit.edu/fairuse. 493 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Tuskegee Syphilis Experiment
• 1932-1972 experiment to study natural progression of disease • 399 African-American sharecroppers w/ syphilis • failed to treat even WWW.BSSVE.INafter penicillin was shown to be an effective treatment in 1940’s
Public domain image from Wikimedia Commons. http://en.wikipedia.org/wiki/Tuskegee_syphilis_experiment
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 494 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Practice & Research
• The term “practice” refers to interventions that are designed solely to enhance the well-being of an individual patient or client and that have a reasonable expectation of success. • The term “research” designates an activity designed to test an hypothesis, permit conclusions to be drawn, and thereby to develop or contribute to generalizable knowledge. • Research and practice may be carried on together when WWW.BSSVE.INresearch is designed to evaluate the safety and efficacy of a therapy. ... if there is any element of research in an activity, that activity should undergo review for the protection of human subjects. From the Belmont Report
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 495 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Basic Ethical Principles
• Respect for Persons • Beneficence Justice •WWW.BSSVE.IN
From the Belmont Report
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 496 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Respect for Persons
• Each person is an autonomous agent, capable of deliberation about personal goals and of acting under the direction of such deliberation • Persons with diminished autonomy are entitled to protection: e.g., children, physically or mentally disabled, prisoners. •WWW.BSSVE.IN Requires Informed Consent • Adequate information • Voluntary participation
From the Belmont Report
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 497 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
(Informed Consent)
• Study involves research, purpose of research, duration, procedures, what is experimental? • Foreseeable risks and discomforts • Possible benefits to participants or others • Alternative procedures that might be beneficial • How confidentiality will be maintained • WWW.BSSVE.INFor research involving more than minimal risk, what compensations and treatments may be available, and where to get further information • Participation is voluntary; no penalty for refusal
http://www.hhs.gov/ohrp/policy/consent/index.html
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 498 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Beneficence
• Do no harm • one should not injure one person regardless of the benefits that might come to others • minimize risk to participants • Maximize possible benefits WWW.BSSVE.IN• to society • but, research subjects may not benefit directly • Some tradeoffs are unavoidable From the Belmont Report
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 499 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Justice
• Varied views of equal treatment • equal share • individual need • individual effort • societal contribution WWW.BSSVE.IN• merit • Select participants fairly • Distribute benefits fairly From the Belmont Report
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 500 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Enforcement: The Common Rule
• Applies to all US Government funded projects involving human subjects • Institutional Review Boards (IRB) review and must approve all such proposed research; responsible to protect subjects • yearly review of research protocols, informed consent, training of researchers, etc. Criteria of Belmont Report. • expedited review for research involving “no more than minimal risk”; consent may be waived • exemptions for educational research, food quality research, and WWW.BSSVE.INretrospective research on public or de-identified data • IRB’s also responsible for protection of confidentiality • MIT’s IRB is the Committee on Use of Humans as Experimental Subjects (COUHES)
http://www.hhs.gov/ohrp/policy/consent/index.html
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 501 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Privacy vs. privacy
WWW.BSSVE.IN
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 502 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Protecting…
• What? • Privacy • Individual’s desire to limit disclosure of personal information • Confidentiality • Information sharing in a controlled manner • Security • Protecting information against accident, disaster, theft, alteration, sabotage, denial of service, … • Against what? WWW.BSSVE.IN• “Evil hackers” • Malicious insiders • Stupidity • Information Warfare
www.bsscommunitycollege.in www.bssnewgeneration.in13 www.bsslifeskillscollege.in 503 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Privacy
• Right to be let alone; e.g.: • … applies mostly to • snooping on Dan Quayle known individuals by J. Rothfeder • “outing” of Arthur Ashe (HIV), Henry Hyde (adultery) • celebrity medical problems (Tammy WWW.BSSVE.INWynette, Nicole Simpson)
www.bsscommunitycollege.in www.bssnewgeneration.in14 www.bsslifeskillscollege.in 504 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Privacy in obscurity
• Right to remain unknown
• WWW.BSSVE.INCorrelation among Images by MIT OpenCourseWare. pervasive databases: • census • marketing • health
www.bsscommunitycollege.in www.bssnewgeneration.in15 www.bsslifeskillscollege.in 505 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Confidentiality
• Use and sharing of information by multiple users at many institutions • Should be controlled by coherent policy • Enforced by appropriate technology
• E.g., who may use results of your life insurance physical exam, for what WWW.BSSVE.INpurposes?
www.bsscommunitycollege.in www.bssnewgeneration.in16 www.bsslifeskillscollege.in 506 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Legitimate Concerns (some may be ameliorated by ACA)
• Difficulty getting insurance • “Individual insurers may deny you coverage based on your medical history if it includes: • Use of prescription drugs to treat anxiety, depression or a physical condition, including Ativan, Klonipin, Paxil, Prozac, Serzone, Zoloft, Xanax and Wellbutrin. • Counseling for anxiety, depression, grief or an eating or sleep disorder. Even if you briefly sought counseling as a way to cope with the Sept. 11 terrorist attacks, you could be denied individual health WWW.BSSVE.INinsurance, according to researchers with Georgetown's Health Privacy Project.” (MSN, March 9, 2004) • Medical Information Bureau • Data on all applicants for private life insurance in past 7 years
www.bsscommunitycollege.in www.bssnewgeneration.in17 www.bsslifeskillscollege.in 507 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Additional Legitimate Concerns
• When employer pays insurance premiums, you may lose your job • Self-insured companies • Small employers facing “experience rated” policies • Non-employment discrimination based on health • Adoption WWW.BSSVE.IN• Politics • Social stigma
www.bsscommunitycollege.in www.bssnewgeneration.in18 www.bsslifeskillscollege.in 508 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Patient’s Clinical Consulting Employer Laboratory Physician Employer’s Clinic & Wellness program State Bureau of Vital Statistics Care Provider (physician, hospital
Medical Managed Care Researcher Patient Organization
Accrediting Life Insurance Organization Company
Retail Pharmacy Medical Spouse’s Information Lawyer in Health Employer Bureau Insurance malpractice case WWW.BSSVE.INCompany
Pharmacy Benefits Manager long term repository, patient-identified data short term repository, patient-identified data
flow of patient-identified medical information temporary access, patient-identified data flow of non-identifiable medical information long term repository, non-patient-identified data temporary access, non-patient-identified data www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 509 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Security
• Integrity of data • No unauthorized modifications • No “dropped bits” • Availability • Natural disaster • Adversary attack WWW.BSSVE.IN• Inadequacy of backup, fail-over • Enforcement of confidentiality policies
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 510 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
De-Identification
WWW.BSSVE.IN
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 511 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Identifiable
• HIPAA: Name, address, phone number, fax number, email address, URL, IP address, social security number, medical record n., health plan n., account n., certificate/license n., vehicle id, device id, biometric id, full-face photo, date of birth, zip code, gender, race, profession • “any other unique identifying number, characteristic, or code” • “actual knowledge that the information could be used … to identify” • Patterns of doctor visits, immunizations, etc. • WWW.BSSVE.INidentifiable by inference • depends on knowledge and abilities of data user • Small bin sizes lead to identifiability • Aggregate data into larger bins • dob => age • 3 digits of zip code
www.bsscommunitycollege.in www.bssnewgeneration.in22 www.bsslifeskillscollege.in 512 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Sweeney’s Cambridge
• 1997 Cambridge, MA voting list on 54,805 voters • Name, address, ZIP, birth date, gender, … • Combinations that uniquely identify: • Birth date (mm/dd/yy) 12% • BD + gender 29% • BD + 5-digit ZIP 69% • BD + 9-digit ZIP 97% • WWW.BSSVE.INUnique individuals • Kid in a retirement community • Black woman resident in Provincetown
www.bsscommunitycollege.in www.bssnewgeneration.in23 www.bsslifeskillscollege.in 513 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Problem of “other information”
• Governor Weld’s data found in Mass “de-identified dataset” • Dates you visited a health care provider (over a lifetime) are probably unique • Can be used to re-identify you if someone has both de-identified data and other data that link to identifiers • WWW.BSSVE.INGenetics makes this immensely more problematic • Think Gattaca
www.bsscommunitycollege.in www.bssnewgeneration.in24 www.bsslifeskillscollege.in 514 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Danger of Re-identification
WWW.BSSVE.IN
Figure by Sweeney, Latanya. "Computational disclosure control: A primer on data privacy protection." Massachusetts Institute of Technology, 2001.
www.bsscommunitycollege.in www.bssnewgeneration.in25 www.bsslifeskillscollege.in 515 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Protection via generalization
WWW.BSSVE.IN
Figure by Sweeney, Latanya. "Computational disclosure control: A primer on data privacy protection." Massachusetts Institute of Technology, 2001.
www.bsscommunitycollege.in www.bssnewgeneration.in26 www.bsslifeskillscollege.in 516 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Computational Disclosure Control
• Make sure data cannot be traced back to a set of size < n • Generalization • Suppression of unique combinations • Account for leakage from what has been suppressed; e.g., back- calculating from aggregate statistics • WWW.BSSVE.INHow to estimate “external information”? • Every release becomes more external info.
www.bsscommunitycollege.in www.bssnewgeneration.in27 www.bsslifeskillscollege.in 517 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Methods of Generalization/ Suppression
• Underlying problem (find minimal generalization/suppression to achieve a level of anonymity) is NP-hard (Vinterbo) • Mainly heuristic search over space of possible generalizations/suppressions • Scrub, Datafly, µ-Argus (Netherlands), k-Similar • Lasko: spectral anonymization • Build a model of data that captures the n-th order statistics of the WWW.BSSVE.INdistribution • Synthesize “fake” patients from that distribution
www.bsscommunitycollege.in www.bssnewgeneration.in28 www.bsslifeskillscollege.in 518 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
MIT OpenCourseWare http://ocw.mit.edu
HST.950J / 6.872 Biomedical Computing Fall 2010
For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.
WWW.BSSVE.IN
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 519 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
In the Year 2000... --Conan O’Brien 6.872/HST950
WWW.BSSVE.IN
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 520 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Prognostication
• “There is nothing so hard to predict as the future.” • --Yogi Berra • Sources of insight: • Technology • Policy • Economics
WWW.BSSVE.IN
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 521 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Evidence-Based Prognostication --Slides from Bill Stead,Vanderbilt University
• Review NAS committee charged with finding ways in which computer science can bear on improving healthcare. • By charge, clearly oriented toward technology • We found that other components of the triad are perhaps even more important
WWW.BSSVE.IN
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 522 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Committee on Engaging the Computer Science Research Community in Healthcare Informatics
Computer Science and Telecommunications Board
National Research Council of the National Academies Chartered by the National Library of Medicine WWW.BSSVE.IN
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 523 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Project Scope
CSTB would conduct a 2-phase study to examine information technology (IT) challenges faced by the health care system in realizing the emerging vision of patient-centered, evidence-based, efficient health care using electronic heath records (EHR) and other IT. The study would focus on the foundation issue of the EHR.
WWW.BSSVE.IN
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 524 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Phase 1
Conduct a series of site visits to a variety of health care delivery sites. Provide a phase 1 report, based on the site visits . Match between today's health information systems and current plans for using EHR nationwide . Problems that could be solved relatively easily and inexpensively by today's technologies . Illustrate how today’s CS knowledge could be used to gain short term improvements . Important questions that future reports should address
WWW.BSSVE.IN
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 525 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Phase 2 Provide a phase 2 report . Technical areas where additional H/BMI or CS research is needed to advance health care IT . Priorities for research that would yield significantly increased medical effectiveness or reduced cost . Information management problems whose solutions require new practices and policies . Public policy questions that need to be resolved to allow such research to proceed
WWW.BSSVE.IN
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 526 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Committee Membership
William W. Stead, Chair, Vanderbilt University G. Octo Barnett, Massachusetts General Hospital Susan B. Davidson, University of Pennsylvania Eric Dishman, Intel Corporation Deborah L. Estrin, University of California, Los Angeles Alon Halevy, Google, Inc. Donald A. Norman, Northwestern University Ida Sim, University of California, San Francisco Alfred Spector, Independent Consultant Peter Szolovits, Massachusetts Institute of Technology Andries Van Dam, Brown University Gio Wiederhold, Stanford University WWW.BSSVE.IN
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 527 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Site Visits
University of Pittsburgh Partners Healthcare Medical Center Boston, MA Pittsburgh, PA Intermountain Health Care Veterans Administration Salt Lake City, UT Washington, DC University of California, HCA TriStar San Francisco Nashville, TN San Francisco, CA Vanderbilt University Palo Alto Medical Foundation Medical Center Palo Alto, CA Nashville, TN
WWW.BSSVE.IN
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 528 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in (Stead’s) Personal Observations Patient records are fragmented Clinical user interfaces mimic paper without human factors & safety design Biomedical devices are poorly integrated Systems are used often to document what has been done, after the fact Support for evidence-based medicine and computer-based advice is rare Clinical research activities are not well integrated into clinical care Legacy systems are predominant Centralization is the predominant method of standardization Implementations timelines are long and course changes are expensive Response times are variable and long down times occur WWW.BSSVE.IN
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 529 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
(Szolovits’) Personal Observations
• Absence of “systems” view of healthcare • Local optimization: e.g., documentation whose main purpose is to avoid losing lawsuits • Focus on “things”, not processes: e.g., do we really need to capture all data about every patient encounter? • Surprising heterogeneity of computer systems ==> tower of Babel • Awful system designs, poor HCI, no principles; all legacy • “Mainframe medicine”: conventional views of where leverage might come from • Poor support for communications WWW.BSSVE.IN
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 530 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Report: Computational Technology for Effective Health Care: Immediate Steps and Strategic Directions h p://www.nlm.nih.gov/pubs/reports/comptech_2009.pdf Released 1/9/2009, to good press
• Today’s healthcare is broken: high costs, low accomplishments • Poor compliance with evidence-based guidelines: inadequate care and inappropriate care • Estimated 40 mins/year to apply guidelines to “average” patient, yet patient sees doctor about 60 mins/year. • Causes: • complex tasks and workflows • Institutional structure and economics • Deficient healthcare information technology WWW.BSSVE.IN
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 531 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Complex Tasks and Workflows
• Uncertainty • Interrupted workflows, poor information flow • Time pressure, aging population, more knowledge
WWW.BSSVE.IN
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 532 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Economics and Institutions
• Large number of payers, each with different rules • Survey of medical centers: 578-20K different plans • Typical doc spends 50 mins/day with health plan hassles • Perverse incentives: $$$ for procedures, little for prevention • Greater pay for patients who develop complications! • Siloed institutions • Shortage of nurses, primary care docs, etc.
WWW.BSSVE.IN
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 533 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
(IOM) Goals of Healthcare
• Safe • Effective • Patient-centered • Timely • Efficient • Equitable
WWW.BSSVE.IN
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 534 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
How can HCIT support these?
• Comprehensive data on a patient’s conditions, treatments, and outcomes. • Cognitive support for HC professionals to help integrate patient- specific data • Cognitive support to apply evidence-based guidelines • Instruments for tracking a panel of patients, highlighting developing problems • “Learning” health care system--integrate new biology, instrumentation, treatments; use experience as basis for new knowledge • Provide care in many locales: home, drug store, clinic, hospital, ... • Empowerment of patients and families WWW.BSSVE.IN
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 535 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
But, with rare exceptions, we saw...
• IT not integrated into clinical practice • Little support for feedback or evidence-based practice • Process improvements are all external to practice • Research is external to practice • No integrated overview of patient data • Software much harder to use than Quicken • No cognitive support for data interpretation, planning, collaboration • Systems oriented around transactions, not “state of patient” WWW.BSSVE.IN
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 536 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Four Domains of HCIT --Marc Probst, Intermountain Health Care
• Automation: IT to do repetitive tasks. E.g, medication administration, lab results, invoice generation • Connectivity: From physical to logical to people • Decision Support: Provide information at a high conceptual level, related to decisions about care. • Data Mining:Analyze all collected data
WWW.BSSVE.IN
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 537 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Strategies
• Evolutionary change • Focus on improvements in care; technology is secondary • Incremental gain from incremental effort • Record available data for care, process improvement, research • Design for human and organizational factors • Support cognitive functions of HC professionals, patients and organizations
WWW.BSSVE.IN
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 538 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Strategies
• Radical change: • Architect to accommodate disruptive change • Archive data for subsequent re-interpretation • Develop technologies to eliminate ineffective work processes • Develop technologies to clarify the context of data
WWW.BSSVE.IN
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 539 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Framework for HCIT Challenges
General Healthcare Data & process integration Standardization High-quality graphics, better UI Codification of best practices design Open-source models for sharing Human language translation of information and knowledge capabilities Applied Use & develop medical Business process integration ontologies Ontology management Data prioritization Search paradigms Scalability
Reasoning Advanced models of differential Machine learning diagnosis Explanation Outcome-based population level Multi-modal interfaces learning Advanced Meta-modeling Super-“Archimedes” Adaptive models Privacy management WWW.BSSVE.INAddition of semantics Better uncertainty management Models of accuracy & precision
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 540 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Grand Challenge 1 Patient-Centered Cognitive Support
Figure removed due to copyright restrictions. See Figure 5.1 from Willam W. Stead and Herbert S. Lin, eds. "Computational Technology for Effective Health Care: Immediate Steps and Strategic Directions." National Academies Press, 2009.
WWW.BSSVE.IN
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 541 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Other Grand Challenges
• Comprehensive modeling, from molecular biology to public health; stops along the way for pathophysiology, organs, patients, populations • Automation, from pill-counting to autonomous closed-loop control of post-surgical patients • Data integration, sharing and collaboration • Data management at scale, including text, annotations, metadata, ontologies, privacy, HCI, and performance • Full capture of physician-patient interactions WWW.BSSVE.IN
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 542 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Committee Recommendations
• Comprehensive data on patients’ conditions, treatments, and outcomes; • Cognitive support for health care professionals and patients to help integrate patient- specific data where possible and account for any uncertainties that remain; • Cognitive support for health care professionals to help integrate evidence-based practice guidelines and research results into daily practice; • Instruments and tools that allow clinicians to manage a portfolio of patients and to highlight problems as they arise both for an individual patient and within populations; • Rapid integration of new instrumentation, biological knowledge, treatment modalities, and so on into a “learning” health care system that encourages early adoption of promising methods but also analyzes all patient experience as experimental data; • Accommodation of growing heterogeneity of locales for provision of care, including home instrumentation for monitoring and treatment, lifestyle integration, and remote assistance; • Empowerment of patients and their families in effective management of health care decisions and their implementation, including personal health records, education about the individual’s conditions and options, and support of timely and focused communicationWWW.BSSVE.IN with professional health care providers.
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 543 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Strategic Health IT Advanced Research Projects (SHARP)
• Four $15M projects to focus on outstanding technical issues: • Security of Health Information Technology research to address the challenges of developing security and risk mitigation policies and the technologies necessary to build and preserve the public trust as health IT systems become ubiquitous. (University of Illinois, Urbana-Champaign) • Patient-Centered Cognitive Support research to address the need to harness the power of health IT in a patient-focused manner and align the technology with the day-to-day practice of medicine to support clinicians as they care for patients. (University of Texas Health Sciences Center, Houston) • Health care Application and Network Platform Architectures research to focus on the development of new and improved architectures that are necessary to achieve electronic exchange and use of health information in a secure, private, and accurate manner. (Harvard University) • Secondary Use of Electronic Health Record Data research to identify strategies to enhance the use of health IT in improving the overall quality of health care, population health and clinical research while protecting patient privacy.WWW.BSSVE.IN (Mayo Clinic) http://www.hhs.gov/news/press/2009pres/12/20091218c.html
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 544 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Cognitive Patient-Centered Cognitive Support Support
• Taking care of the whole patient, not individual facts • Integration of individual patient information across modalities, time, doctors • Visualization of anatomical, functional and pathological conditions • Illustration of changes over time • “Drill down” to details • Semi-automated application of evidence-based guidelines • Background tracking of alternative actions consistent with guideline and patient state • Monitoring and alerting on deviations from guideline • Disease management “dashboard” • Standard order sets • Continuous instrumentation tracks real-time patient state: heart, temperature, movement, respiration, urine, eating, transdermal serum components (e.g., glucose) • Ubiquitous cell phones • Empower theWWW.BSSVE.IN patient • based on Computational Technology for Effective Health Care, 2009 http://www.nlm.nih.gov/pubs/reports/comptech_2009.pdf www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 545 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Architecture The iPhone(-like) Alternative
Mandl & Kohane prescription: “a flexible information infrastructure that facilitates innovation in wellness, health care, and public health.” •Liquidity of data •Substitutability of applications •Open standards, supporting both free and commercial software •Natural selection in a health information economy, based on value and cost • permits disruptive innovation • avoids “design by committee” WWW.BSSVE.IN Mandl KD, Kohane IS. No small change for the health information economy. NEJM 2009:360:1278-81 Images from http://www.technologyreview.com/biomedicine/22360/
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 546 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Categories of Substitutable Applications
prescribing, clinician order entry, medication reconciliation, drug-safety Medication management alerts
Documentation structured text entry, dictation
disease management, appointment and testing reminders, care Panel management instructions, results notification, behavior modification
Quality improvement HEDIS measurement, management of patient transfer and transition
Administrative tools billing, referral management, risk stratification
doctor-patient communication, multispecialty or team communication, Communication patient support, social networking
Public health reporting notifiable disease reporting, biosurveillance, pharmacosurveillance
clinical trial eligibility, cohort study tools, electronic data capture for Research trials
Decision support lab test interpretation, genomics, guideline management
WWW.BSSVE.INlab data feed, dispensed meds feed, PCHR data feed, public health Data acquisition data feed
Mandl KD, Kohane IS. No small change for the health information economy. NEJM 2009:360:1278-81 www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 547 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in Secondary Use Secondary Use of Clinical Data
• Goals: • Phenotype = f(Genotype, Environment) • Public Health reporting, modeling • High Throughput Genotyping demands high-throughput sources of Phenotype and Environment • Clinical record is our best proxy for these • Approaches • Share data • Ontologies to make meaning shareable • Natural language processing to turn narrative text into data
WWW.BSSVE.IN
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 548 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Bulk of Valuable Data are in Narrative Text
Mr. Blind is a 79-year-old white white male with a history of diabetes mellitus, inferior myocardial infarction, who underwent open repair of his increased diverticulum November 13th at Sephsandpot Center. The patient developed hematemesis November 15th and was intubated for respiratory distress. He was transferred to the Valtawnprinceel Community Memorial Hospital for endoscopy and esophagoscopy on the 16th of November which showed a 2 cm linear tear of the esophagus at 30 to 32 cm. The patient’s hematocrit was stable and he was given no further intervention. The patient attempted a gastrografin swallow on the 21st, but was unable to cooperate with probable aspiration. The patient also had been receiving generous intravenous hydration during the period for which he was NPO for his esophageal tear and intravenous Lasix for a question of pulmonary congestion. On the morning of the 22nd the patient developed tachypnea with a chest X-ray showing a question of congestive heart failure. A medical consult was obtained at the Valtawnprinceel Community Memorial Hospital. The patient was given intravenous Lasix. A arterial blood gases on 100 percent face mask showed an oxygen of 205, CO2 57 and PH 7.3. An electrocardiogram showed ST depressions in V2 through V4 which improved with sublingual and intravenousWWW.BSSVE.IN nitroglycerin. The patient was transferred to the Coronary Care Unit for management of his congestive heart failure , ischemia and probable aspiration pneumonia. www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 549 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Secondary Use SHARP Project
• Clinical Data Normalization Services and Pipelines • Natural Language Processing • High-Throughput Phenotyping • Scaling to enable near-real-time throughput • Data Quality • Real-world evaluation framework
Portraits of Chris Chute and Guergana Savova have been removed due to copyright restrictions.
WWW.BSSVE.IN
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 550 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
Theme
Static/Passive ⇒ Dynamic/Active Data ⇒ Interpretation Choices ⇒ Suggestions Signals ⇒ Alerts Repositories ⇒ Models Documentation ⇒ Decision Support Retrospective ⇒ Real-time ...
• Each such move requires advances in Artificial Intelligence, Data Mining, Natural Language Processing, Human-Computer Interaction, Computer Performance Engineering,WWW.BSSVE.IN ...
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 551 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in “Oh, the future’s so bright, we’ll have to wear sunglasses!” -- Barbara Kooyman, Timbuk 3 -- with thanks to Phil Greenspun
WWW.BSSVE.INImage by Benjamin.C on Flickr.
33 www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 552 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
MIT OpenCourseWare http://ocw.mit.edu
HST.950J / 6.872 Biomedical Computing Fall 2010
For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.
WWW.BSSVE.IN
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in 553 www.onlineeducation.bharatsevaksamaj.net www.bssskillmission.in
The Sana strategy is to form partnerships between provider organizations and academic centers (with computer science and public health capacity) in developing countries. It has been successful in developing such partnerships in the US, India, Great Britain, Mexico, the Philippines, Greece and Brazil. Furthermore with generous network coverage in even the poorest regions in the world, the falling price of cell phone handsets and the development of low cost tablet devices, Sana as a medical informatics solution is becoming ever more viable even in the poorest regions of the world. The platform is distributed for free via a BSD Open Source Software license. The team has developed an integrated package of hardware, software, training, maintenance, evaluation and quality improvement measures to support our partners in customizing the software to meet their needs.
In low- and middle-income countries, where health systems are centralized (care Typically, the initial contract with a partner organization has included a bundle of provision is focused in urban areas) and fragmented (providers of health care largely services for adapting clinical forms to a local workflow, including designing the entire work independently with weak quality assurance and improvement standards), much of operating system (software, hardware, and human resources) if one does not currently the population lacks access to high quality, affordable health care. Development exist. Sana also helps train clinicians and healthcare workers on proper use of the cell programs have tried to bridge the health gap, often with the use of technology, but have phone application. The team and implementing partners collaborate to raise grants to been criticized for their focus on short term vertical programs rather than longer term cover the costs of implementing a project. Partners, described below, are supported to capacity building. In order to create sustainable change in developing countries, develop sustainable business models to scale their programs. technology innovation by itself is not enough. The problem is not only one of a lack of technical solutions but also insufficient support for local leadership, and Karnataka, India with Narayana Hrudayalaya hospital and the Rajiv Gandhi University underdevelopment of networks and business model innovations necessary to propagate of Health Sciences and sustain value creation. The hospital was connected to rural health care providers in the state of Karnataka Sana is a volunteer group of engineers, doctors, social entrepreneurs and public using a customized version of the Sana platform with locally designed procedures for oral health experts focused on improving the quality of health care in developing countries cancer screening. Over 4000 patients were screened this past summer with about 300 through leveraging innovations in wireless technology. Its approach is interdisciplinary high risk cases identified at rural clinics and referred to Narayana Hruduyalaya hospital and collaborative: encouraging local leaders to develop solutions enabled by the Sana for definitive diagnosis and treatment. Current initiatives plan to scale up to cover the platform, an open source customizable medical records system for Android mobile entire state and develop further applications for cardiovascular disease screening and phones. The goal of Sana is to strengthen health care systems by using innovation to management. The intention is to launch a for-profit entity building a primary care maximize the effectiveness of existing resources and to foster sustainable change through network for the Narayana Hruduyalaya hospital enabled by the Sana platform. a focus on capacity building and cooperation. The approach enables technical innovation (based on the open source Sana platform), business model innovation (based on models Punjab, India with Health Point Service and the Public Health Foundation of India being developed and tested with our partner organizations) and the development of value creating networks (building coalitions of local and international academic and provider Sana worked with Health Point Service to extend the reach of their six primary organizations to identify and share examples of best practice and pool resources). care centers in rural Punjab, North India. In this pilot, one community healthcare worker has been trained for each clinic and equipped with a customized protocol for assessing The Sana open-source software platform enables smart phones or tablet devices the risk of cardiovascular disease and referring high risk patients for treatment. Outcomes running Google’s Android operating system to be used to data from this three month pilot will be available in Jan 2011. The aim is to integrate the Sana enabled outreach activities into Health Point Service’s for-profit model of low cost • Input patient histories (based on best practice algorithms), rural health care delivery. • Input multimedia information (including pictures and video), • Integrate with Point of Care diagnostic devices (EKG, Ultrasound, electronic Philippines with Negros Women for Tomorrow Foundation, the Center for Community stethoscopes and clinical chemistry) Transformation, University of the Philippines, Ateneo de Manila University, and Asia • Integrate this data to any electronic medical record database. Pacific College.
Sana is working with local partners to implement a screening program for hypertension, diabetes and chronic lung disease to the Rizal province, and Negros Occidental. The intention is to develop a suite of applications on the Sana platform to facilitate the expansion of existing social insurance offerings to include primary care.
Brazil with MIT Netra Lab, Instituto Nacional de Telecomunicações, and Universidade Federal de Sao Paulo
Sana has partnered with a Brazilian NGO, Anjos de Saude, formed by a group of currently Boston-based professionals - ophthalmologists, computer scientists and public health experts - to provide eye care to underserved populations in Brazil. A procedure has been developed that assesses the risk of eye disease based on questions, acuity testing on the phone (using Professor Ramesh Raskar’s work at the MIT Media Lab) and fundoscopy (using a low cost integrated device developed by the Brazilian team). The plan is for pilot implementation in Jan 2011 in the slums of Sao Paolo with scale up over 12 months to cover contiguous underserved areas. The goal is to secure recurrent funding from the government of BrazilWWW.BSSVE.IN as a low cost high quality vehicle for their outreach activities.
The Sana mission is intentionally aligned to Millennium Development Goal 8: Develop a global partnership for development. Specifically the organization came into being to “make available the benefits of new technologies, especially information and communications” through partnership. Pilot projects have shown that the Sana approach works. The organization believes that current initiatives can be scaled to achieve its vision of strengthening health systems through a multidisciplinary collaborative use of wireless technology. As more people adopt and develop the open source Sana platform and share their experiences with each other, a virtuous cycle of collaboration and growth can be established that will positively impact health outcomes even within existing resource constraints.
Visit www.sanamobile.org.
www.bsscommunitycollege.in www.bssnewgeneration.in www.bsslifeskillscollege.in