The Value of Semantic Interoperability for Healthcare

Orion Health White Paper Michael Hosking, CHIA, B, OccTh Clinical Product Specialist 022018 The Unambiguous Exchange and Reuse of Clinical Knowledge

How do we, as humans, convey meaning? Through speech, the written word, body language, other subtle cues or most importantly a combination of these. Around the world we have different ways or words to represent the same meaning.

What do you take when you go to the beach? Flip flops? Jandals? Maybe a pair of thongs? These are just some variations in the English language, not to mention international languages. Ultimately, they are all describing the same thing.

In healthcare, conveying and interpreting clinical meaning is complex, especially when it comes to a clinical information system (CIS). Healthcare professionals already spend a significant amount of their time understanding and interpreting complex intricacies and relationships between clinical data. So the need to use data to link disorders, medications, diagnostic imaging (and much more) along with specific individual patient attributes, is vital in all professional roles.

Does your health service understand the benefits of a CIS that supports clinical terminology systems and services?

Orion Health White Paper The Value of Semantic Interoperability for Healthcare 2 Problem

Information within a CIS cannot be easily systems and locally derived codes. This is found, appropriately reused or analyzed. As a a barrier to achieving meaningful shared result, due to the limitation of timely access to information and knowledge. crucial clinical information, clinical workflow and health outcomes such as accurate Understanding the Clinical System Knowledge resource allocation are inhibited. Cycle A clinical system knowledge cycle refers to the consumption and production of knowledge Research has found that 80%1 of healthcare across a CIS that informs appropriate clinical, data is unstructured and currently unable business and population health decisions. to be effectively reused or analyzed. This is This is supported by semantic interoperability because free text is often one of the primary through effective data re-use. sources of data along with classification

FIGURE 1: Ideal Clinical System Knowledge Above diagram modified from SNOMED International educational presentation “Fitting SNOMED CT into an EHR system” 2016

1 Wang, Y., Kung, L., & Byrd, T. A. (2018). Big data analytics: Understanding its capabil- ities and potential benefits for healthcare organizations. Technological Forecasting and Social Change, 126, 3-13.

Orion Health White Paper The Value of Semantic Interoperability for Healthcare 3 Many CIS do not effectively manage their approach is not conducive to developing a clinical system knowledge cycle2. Currently, it strong knowledge cycle with re-use of clinical is common for such systems to have a system information for improved workflows and data cycle which is considered a limited models of care. sub-function of a knowledge cycle. This

FIGURE 2: Typical Clinical System Data Cycle Above diagram modified from SNOMED International educational presentation “Fitting SNOMED CT into an EHR system” 2016

2 Yusof, M. M., Kuljis, J., Papazafeiropoulou, A., & Stergioulas, L. K. (2008). An evaluation framework for Health Information Systems: human, organization and technology-fit factors (HOT- fit). International journal of medical informatics, 77(6), 386-398.

Orion Health White Paper The Value of Semantic Interoperability for Healthcare 4 Solution Semantic interoperability is the ability to exchange health related data unambiguously Part of the solution to improving clinical and meaningfully between stakeholders4. knowledge sharing and development for Preserving contextual meaning within a improved decision making is interoperability. consistent format is needed to support This is a pretty vague term so let’s break it down. knowledge that is controlled, shareable and reusable. According to HIMSS, Interoperability can be defined at three levels within healthcare This meaning can be supported and information systems 3. represented by terminology systems with ontological characteristics. An ontology • Foundational Interoperability is a formal naming and definition of things • Basic level of interoperability to the level of known truths. These truths • Exchange of basic, raw data from one can include types, characteristics and system to another relationships in a specific domain. An example • Does not require system to interpret data of this is: we can confidently say that H1N1 • Structural Interoperability is a type of influenza virus. The statement is • Intermediate level of interoperability always true, regardless of context and is just • Exchange of structural format of data one of the types of knowledge representation from one system to another discussed in • Requires receiving system to interpret this paper. data field • Example: message format standards - Systematic adoption of controlled, HL7 FHIR standardized, appropriately flexible • Semantic Interoperability terminology systems that support the above • Highest level of interoperability are required for any CIS. Effective adoption • Enables exchange, use and re-use of data of these systems can improve primary and • Requires receiving system to interpret secondary data use. This results in improved meaning of data patient safety and health outcomes which • Example: SNOMED CT leads to significantly reduced care delivery costs. Such costs can be measured in various ways depending on the business problem needing to be solved.

To support semantic interoperability, foundational and structural interoperability Semantic must already exist. Often these exist at the level of technical interoperability including services and messages, transport protocols, Structural data integrity and accessibility. All of which attempt to convey the structure and context of

Foundational the clinical data being shared/used. Following this solid, more technical interoperability

FIGURE 3: Interoperability Levels

3 http://www.himss.org/sites/ 4 Assessing SNOMED CT himssorg/files/FileDownloads/ for Large Scale eHealth HIMSS%20Interoperabili- Deployments in the EU, http:// ty%20Definition%20FINAL.pdf assess-ct.eu

Orion Health White Paper The Value of Semantic Interoperability for Healthcare 5 base, semantic interoperability (using context of health, would be multipurpose for standards and terminologies) can support all users. As an example, a clinician providing the enriching, exchanging and reusing of this clinical care to an individual client does not data as clinically meaningful information. need to be aware of the service planning This supports healthcare professionals to for the purposes of funding requirements. represent and reuse clinical meaning across Likewise, a Government body or health all clinical care settings. This effective reuse organization coordinating funding allocations of clinical information can further enhance does not require complex detail of an effective healthcare insights and resource individual patient’s clinical data findings. allocation, which is especially needed considering our ever increasing population What is a Clinical Terminology? demands. Healthcare professionals can represent Quality clinical data is dependent on complex clinical information in various capturing information in a computer process- ways. Clinical terminologies can support a able format at the point of care. This aids consistent representation of meaning whilst decision support systems in correlation with also allow clinicians to have some flexibility clinical guidelines. Evidence based care and with representing information in familiar ways. service provisioning requires detailed data They enable: analysis. At scale, this can support novel technologies including natural language • Consistent naming and identification of processing with deep learning to accurately clinical knowledge relevant to healthcare predict diagnosis and complete personalized • Unambiguous understanding of prescriptions along with other decision relationships between complex clinical support mechanisms. concepts and associated attributes to provide richer contextual detail To overcome this issue of inconsistent • Semantic interoperability without loss of nomenclature within a CIS or Electronic detail or change in meaning Health Record (EHR), clinical terminology systems such as Systematized Nomenclature Purpose of Clinical Terminology of Medicine – Clinical Terms (SNOMED- CT) and LOINC (Logical Observations Primarily, clinical terminologies have Identifiers Names and Codes) can be used been developed to support both a CIS and as a structured terminology that represents clinician’s providing clinical care. Secondary accurate descriptions of clinical information benefits support the re-use of this information used in clinical practice. This paper will focus which can support other meaningful on the former clinical terminology system, representations for other users, organizations SNOMED-CT. and governments. It also supports a mechanism of clinical data governance for Context and purpose of use improved data collection. To support this, and re-use commitment is required from various health organizations to support and educate staff Depending on the intended purpose, on how to implement effective clinical data audience and outcomes varying levels governance. of information detail and complexity are required. Ultimately, an ideal ontology, in the

Orion Health White Paper The Value of Semantic Interoperability for Healthcare 6 FIGURE 45: Principles of Terminologies, Classification and grouping systems

This diagram shows the level of detail at which This filter-like mechanism is supported by clinical information should be collected and various standardized health coding systems. reused for secondary purposes and users. As Between each system level, there is mapping well as some common examples. If we imagine (linking) which supports the abstraction this as a series of sources and filters: (generalized representation) of information. This enables information to be displayed in 1. Primary clinical data source appropriate detail for the intended audience. a. Detailed data collected and entered by a Both business and user needs should determine healthcare professional which code system and respective mapping/s 2. Secondary filters are most appropriate. There are a range of b. Detailed data can be ‘filtered out’ use cases for mapping between local codes, (abstracted) which leaves the more classification systems and terminologies. generalized categories that are more useful for statistical and reporting needs

5 Bowman, S. E. (2005). systems. Coordination of MA, American Health Informa- Coordination of SNOMED-CT SNOMED-CT and ICD-10: Get- tion Management Association. and ICD-10: getting the most ting the Most out of Electronic 37(4-5), 394. out of electronic health record Health Record Systems/AHI-

Orion Health White Paper The Value of Semantic Interoperability for Healthcare 7 The filtering system appropriately disseminates assessments for service planning, resource information to the level of detail required from management and funding requirements. data source to data re-use based on the use case. This minimizes the need for unnecessary For optimal information accuracy, it is manual data entry and increases the potential best to support clinical users to codify for more accurate data for re-use. the information at the point of information capture. As for the best user experience, International Classification of Diseases (ICD), clinical users should not need to know or currently in its 10th Revision (ICD-10), is see that they are coding this information defined by WHO as a classification system via any user interface. Supporting coding designed for “Use for identification of health clinical detail, at the point of information trends and statistics, and the international capture maintains precision and accuracy of standard for reporting diseases and health information. This should be done to the level conditions”. It is a mutually exclusive of detail that is known by a user supporting system which means that “lung cancer” is familiar terms, abbreviations, shorthand and categorized under the category of cancer, spelling variations. and cancer alone. It cannot, however, also fit under the category of a “respiratory system A common CIS user interface limitation is use disorder”. Use of this system within a CIS of vague terms for clinical data entry. This is a results in limitations for advanced clinical limitation because generalized classifications analytics and complex decision support. often do not support the level of clinical depth required for accurate knowledge ICD classifications correspond to Diagnosis- representation for both clinical users and for Related Groups (DRGs). These are often used a CIS. Terms used from classification systems to classify specific inpatient classifications of like ICD-10 do not assist with clinical workflow care provision. These DRG codes relate to needs or clinical data quality at the point of care. This is demonstrated in the diagram below.

FIGURE 5: Systems and Equivalence

Orion Health White Paper The Value of Semantic Interoperability for Healthcare 8 As stated in the well-known “Desiderata and research. It is also a tool that can support for Controlled Medical Vocabularies in the clinical decision support and various clinical Twenty-First Century”7 there are some core administration process automation. characteristics of an ideal clinical terminology. It has been identified as the preferred national Fundamentally, full value of an EHR or CIS clinical terminology by many countries will not be realized without effective capture across the world. This includes; the United and use of quality, valid data with effective States Healthcare Information Technology clinical and data governance protocols. There Standards Panel, the National Health Service are many clinical terminology systems and in the UK, the Health Information Standards classifications in use today but some are more Organization for its use across the health and mature in their development and adoption disability sector in New Zealand, the ADHA than others. (Australian Digital Health Agency)9 along with other international government and industry What is SNOMED-CT? bodies. SNOMED International currently has over 30 member countries with 8 National SNOMED-CT currently is the most Extensions. comprehensive, multilingual clinical healthcare terminology standard in the Many international governments endorse world8. It contains a large, regularly (six the use of SNOMED-CT as the clinical monthly) revised database of clinical terminology to be used across health and concepts which represents clinical disability sectors. The reason for this is that knowledge. These concepts are assigned the application and use of SNOMED-CT within both human readable terms and each of these an EHR, CIS and knowledge management unambiguous concepts are represented systems support a range of tangible benefits. by a unique identifier. From here, clinical terms can be represented in a user interface Benefits with the unique identifier being captured and interpreted in the background by an SNOMED CT has the potential to benefit appropriately configured clinical information individuals, populations and evidence based system or application. There are many decisions for cost effective care delivery10,11. other applications of this including clinical Such benefits include: knowledge management, decision support mechanisms and population health analytics Enhancing individual care by: and prediction. • Enabling automated clinical decision support systems SNOMED-CT enables a consistent way to • Reducing data collection burdens index, store, retrieve, and aggregate clinical • Supporting data with consistent data across a range of clinical specialities unambiguous semantic information and sites. It also assists in organizing medical sharing record content. This reduces inconsistencies • Enabling clinical information capture using with recording/capturing data, which is encoded and used for clinical care of patients

7 Cimino, J. J. (1998). Desider- 8 SNOMED International 9 https://www.healthtermi- 11 http://www.snomed.org/ ata for controlled medical vo- SNOMED CT Starter Guide nologies.gov.au/ncts/ snomed-ct/why-should-i-get- cabularies in the twenty-first https://confluence.ihts- snomed-ct/business-case 10 century. Methods of informa- dotools.org/display/DOC- http://www.snomed.org/ tion in medicine, 37(4-5), 394. STAR snomed-ct/why-should-i-get- snomed-ct

Orion Health White Paper The Value of Semantic Interoperability for Healthcare 9 consistent, commonly used phrases nursing documentation and communication • Supporting accurate and comprehensive alone. The predicted efficiency gains after semantic based searches three months and an overall accumulated • Supporting multiple languages potential gain was measured to be US$3.7 million after three years and US$65 million over five years. There are a limited number Enhancing population health by: of documented examples of cost savings due • Facilitating population health monitoring to the limited number of nationally scaled • Reflecting evolutionary growth of implementations. These examples are limited changing clinical practices largely to European Union nations. • Enabling accurate access to relevant information • Enabling data to support clinical research and improved treatment • Enhancing audits of care delivery and detailed analysis of clinical records

Enhancing evidence-based healthcare decisions by: • Supporting links between clinical records and changing clinical guidelines • Evaluating and enhancing the care delivery experienced by individuals • Reducing need and cost of inappropriate and duplicate testing and treatment • Reducing the likelihood, frequency and impact of adverse healthcare events • Improving cost-effective care delivery and quality of care delivered

Tangible Cost benefits

Quantifiable cost benefits are difficult to measure due the fact that the economic value of eHealth standards are financially intangible12. It is understood though that the cost benefit of implementations of clinical terminologies in Sweden resulted in total rationalization potential of approximately US$34.7 million per year based on improving

12 Dennis Lee, Nicolette de Keizer, Francis Lau, Ronald Cornet; Literature review of SNOMED CT use, Journal of the American Medical Informatics Association, Volume 21, Issue e1, 1 February 2014, Pages e11–e19, https:// doi.org/10.1136/amia- jnl-2013-001636

Orion Health White Paper The Value of Semantic Interoperability for Healthcare 10 How does SNOMED CT work?

High level example of a SNOMED-CT Concept:

FIGURE 6: SNOMED-CT Concept Example

Formal SNOMED-CT Structure of “Myocardial Infarction”

FIGURE 7: SNOMED-CT Structure

Orion Health White Paper The Value of Semantic Interoperability for Healthcare 11 SNOMED International partners human language. In a health based context, this includes complex clinical language and There is active collaboration with many jargon, because let’s face it, it can be treated other internationals standards organizations as another language in itself. The capability and professional clinical bodies to support of NLP holds significant potential to assist adoption and use of SNOMED CT with clinicians with understanding, analyzing and other international standards14. This aims interpreting healthcare problems. to improve and enhance meaningful, consistent data capture and exchange as How can Natural Language well as improving safety, functionality and Processing (NLP) applications semantic interoperability on a global scale. assist? This includes partnerships with HL7, LOINC, genomic, medical device organization, GS1, Ideally we would want to codify all information the World Health Organization among other capture at the point of care. Defined data international organization15. fields cannot capture every complex detail that is required by a clinical user. This is Clinical Terminology Services where NLP assists in the clinical analysis and reuse of information within free text Health Level 7 (HL7) Fast Healthcare narratives. This can be further supported by Interoperability Resource (FHIR)® defines clinical the clinical system knowledge cycle on page terminology services as services that support 2. There are many clinical service, population healthcare applications use of codes and value health and financial use-cases which can be sets without implementers or users having to supported by Machine Learning (ML) and become experts in the explicit details and the NLP techniques, here are just a few: included code systems and terminological 16 principles . HL7 FHIR® supports terminology 1. Predictive modeling of populations17 APIs which contain various clinical and medicine One of the greatest potential applications based terminology systems (including SNOMED of NLP is population health analysis CT). Such services can improve user interfaces, through cohort identification for similarity referencing clinical records, implementation analytics. As systems mature, populations efforts, scalability, content management and with similar clinical features could be reduce costs through improved management identified through ML. This application of mechanisms and enhancing many features and ML can apply a forecasted risk analysis benefits mentioned in this paper. and clinical decision support for individual patients to prevent onset of chronic What is Natural Language conditions such as diabetes or rheumatoid Processing? arthritis18. In the same way, studies have been able to determine if a tumor was stable Natural Language Processing (NLP) is a or showed regression or progression with field of Artificial Intelligence which enables high sensitivity and specificity. computers to analyze and understand the

14 https://www.snomed.org/ 17 & 18 Sheng Yu, Katherine P automated feature extraction about/partnerships Liao, Stanley Y Shaw, Vivian S and selection from knowledge Gainer, Susanne E Churchill, sources. J Am Med Inform 15 https://www.snomed.org/ Peter Szolovits, Shawn N Mur- Assoc 2015; 22 (5): 993-1000. about/partnerships phy, Isaac S. Kohane, Tianxi doi: 10.1093/jamia/ocv034 16 https://www.hl7.org/fhir/ Cai; Toward high-through- terminology-service.html put phenotyping: unbiased

Orion Health White Paper The Value of Semantic Interoperability for Healthcare 12 2. Automated clinical trial cohort 4. Automated/semi-automated clinical matching19,20 coding25 De-identified populations that meet Extracting content from unstructured clinical trial criteria can be identified as clinical documents (discharge summaries, potential candidates. Upon obtaining reports, letters) for semi-automated appropriate consent, approval and coding and population health indexing. recruitment, this could result in more rapid This would be the basis of a real- developments of drug development and world health outcomes solution since treatment advancements. Faster matching prospective problem lists and messaging of patients to potential treatments which attributes have well recognized limitations relieve symptoms and prolong life, when used in isolation. especially in the case of cancer trials, has potential to assist researchers, clinicians, 5. Clinical decision support26 regulators and patients. Automated report classification and analysis can provide clinicians with guidance of 3. Automated risk/adverse event which clinical reports may recommend monitoring/detection21,22 further analysis or treatment. For Extraction of data from unstructured example, analysis of clinical documents of clinical documents with support from individuals identified to have a diagnosis machine learning algorithms which can of tuberculosis who are indicated to assist in monitoring hospital acquired require further screening or treatment. infections (HAI). This can be achieved through event detection within diagnostic 6. Automated de-identification of reports, discharge summaries and clinic clinical documents for research27 or to letters which are monitored for temporal information access processes28 details relating to specific infections, Privacy and sensitivity are significant diagnoses, physical features and considerations when analyzing protected treatments. One example is urinary tract health information (PHI). Data used can be infections (UTI)23, NLP based monitoring de-identified so that personal identifiers could assist in the identification of will be replaced with non-identifying variable patients/cohorts meeting certain risk names29 supported by Named Entity criteria for UTIs and other conditions to Recognition and co-reference detection30. improve patient safety24.

19 Patel, C., Cimino, J., Dolby, Analysing risk factors for A., & Hersh, W. R. (2010). A Automated de-identification J., Fokoue, A., Kalyanpur, A., urinary tract infection based systematic literature review of of free-text medical records. Kershenbaum, A. & Srinivas, on automated monitoring of automated clinical coding and BMC medical informatics and K. (2007). Matching patient hospital-acquired infection. classification systems. Journal decision making, 8(1), 32. records to clinical trials using Journal of Hospital Infection , of the American Medical 28 ontologies. The Semantic Volume 92 , Issue 4 , 397 - 400 Informatics Association, 17(6), https://github.com/NICTA/ Web, 816-829. 646-651 t3as-redact 22 & 24 Proux D, Hag.ge C, 29 20 Ni, Y., Wright, J., Perentesis, Gicquel Q, Pereira S, Darmoni 26 Demner-Fushman D, Stubbs, A., Kotfila, C., & J., Lingren, T., Deleger, L., S, Segond F, Metzger MH. Chapman WW, Mc- Donald Uzuner, Ö. (2015). Automated Kaiser, M., ... & Solti, I. (2015). Architecture and Systems for CJ. What can natural language systems for the de-identifi- Increasing the efficiency of Monitoring Hospital Acquired processing do for clinical cation of longitudinal clinical trial-patient matching: auto- Infections inside a Hospital decision support? J Biomed narratives: Overview of 2014 mated clinical trial eligibility Information Workflow -Pro- Inform 2009;42(5):760– i2b2/UTHealth shared task pre-screening for pediatric ceedings of the Workshop on 772. Track 1. Journal of Biomedical oncology patients. BMC med- Biomedical Natural Language Informatics, 58. doi:10.1016/j. 27 ical informatics and decision Processing. Pg 43-48 Neamatullah, I., Douglass, jbi.2015.06.007 M. M., Li-wei, H. L., Reisner, making, 15(1), 28. 30 25 Stanfill, M. H., Williams, A., Villarroel, M., Long, W. https://github.com/NICTA/ 21 & 23 Redder, J.D. et al. M., Fenton, S. H., Jenders, R. J., ... & Clifford, G. D. (2008). t3as-redact

Orion Health White Paper The Value of Semantic Interoperability for Healthcare 13 The benefit of using Deep Learning (DL)

Highly accurate predictions of diagnosis and patient needs through unsupervised deep feature learning31 has already been modeled with significant results. This has not, however, made full use of clinical terminologies to further support deep semantic description logic for more highly predictive modeling for personalized treatment options, diagnostic support and accurate precision medicine.

Patterns and relationships found through deep learning (DL) across clinical information has the opportunity for uncovering unexpected or previously unknown enhancements and knowledge of clinical best practice guidelines, treatments, diseases and medicine effectiveness. Results from a recent study was able to determine the regression or progression of disease, with a sensitivity of 81% and specificity of 92%32.

FIGURE 8: Model to Predict the Future of Patients Technology such as DL, ML and NLP support Above diagram from Deep patient: An unsupervised representation to healthcare professionals and clinicians to predict the future of patients from the electronic health records.33 improve their focus on supporting patients with complex diagnoses, treatments and interventions. This is in combination with the ever-important aspects of interpersonal relationships, empathy, informed decision making, trust and much-needed human attention and care.

31 Miotto, R., Li, L., Kidd, B. A., 32 Pons, E., Braun, L. M., Hun- 33 (FIGURE 8) Miotto, R., Li, L., & Dudley, J. T. (2016). Deep ink, M. M., & Kors, J. A. (2016). Kidd, B. A., & Dudley, J. T. patient: An unsupervised Natural language processing (2016). Deep patient: An representation to predict the in radiology: a systematic unsupervised representation future of patients from the review. Radiology, 279(2), to predict the future of patients electronic health records. 329-343. from the electronic health Scientific reports,6. records. Scientific reports,6.

Orion Health White Paper The Value of Semantic Interoperability for Healthcare 14 Conclusion Does your health service understand the benefits of a CIS To provide safe, quality healthcare and that supports clinical terminology achieve semantic interoperability in the systems and services? future, organizations need to employ consistent terminology systems that SNOMED®, SNOMED CT® and IHTSDO® match the wider healthcare market. While are registered trademarks of International cost benefits are difficult to measure, they Health Terminology Standards Development will likely be realized after appropriate Organisation. SNOMED CT® licensing assessments of the system are completed33. information is available at http://snomed. org/licensing. For more information about Clinical terminology services simplify the SNOMED International and SNOMED implementation of such systems, reducing International Membership, please refer to information management efforts and http://www.snomed.org or contact info@ enhancing user interfaces, clinical decision snomed.org. support and content management.

33 Thiel, R., Birov, S., Piesche, More advanced technology is required to K., Højen, A. R., Gøeg, K. R., better enable clinicians and healthcare Dewenter, H., ... & Stroet- mann, V. N. (2016, Septem- professionals to more effectively focus on ber). The Costs and Benefits of complex cases and the more human aspects SNOMED CT Implementation: An Economic Assessment of healthcare. Model. In MIE (pp. 441-445).

When considering any implementation of a CIS, it is critical to understand whether these systems can support the needs of all users of clinical and business information, not only through interoperability of data, but interoperability of meaning.

Orion HealthTM is a trademark of Orion Health group of companies. All other trademarks displayed in this document are the property of Orion Health or their respective owners, and may not be used without written permission of the owner. All patient information shown in any imagery is for representation and demonstration purposes only and is not related to a real patient. Orion Health makes no warranties and the functionality described within may change without notice.