Reviewed articles

Clinical coders and decision making Andrew Wooding Abstract Clinical coders operate at six identifiable levels, which can be described as beginner, trainee coder, entry, competent, accredited and advanced levels. In this article these levels are elaborated within the theoretical perspectives of Simon’s four-stage and Wilson and Walsh’s six-stage models of systematic decision making. The article then examines briefly the importance of understanding how clinical coders make decisions, because of both the coder’s crucial role in determining funding, and the reliance of research upon access to accurate data. Finally, future avenues for study in this area are suggested. Keywords: Decision making; clinical coding; health classification. Disclaimer: This work is the author's own opinion and does not necessarily represent the views of the Auckland District Health Board.

Introduction port systems in Decision Support Systems and Intelligent Systems, Turban and Aronson (2001) imply How do clinical coders decide what to code and which that they view this phase to be of minor importance. code or codes to use? Is there a systematic decision However, omitting the review phase would leave one making process that clinical coders use? How can without the knowledge that the process resulted in a managers ensure that the decision making processes ‘correct’ decision (or not). are accurate and effective, thereby resulting in clini- Wilson and Walsh (1996) synthesise several au- cally meaningful data? thors’ models and add two more phases to Simon’s: To answer these questions, this article will first • intelligence consider the classic four-phase model of systematic • decision making derived from management science intention • (Simon 1977) and compare it with a six-phase model design proposed by Wilson and Walsh (1996) from a litera- • choice ture review of information-seeking behaviour. Sec- • information extraction and integration ondly, it will look at the informational and knowledge • review needs of the six levels through which a Wilson and Walsh suggest that Simon’s focus on progresses (Wissmann 1999) and how each level uses machine operations is the reason for his omission of the six-phase model. Thirdly, it will explore the issues the information extraction and integration phase. associated with accurate decision making by clinical Definitions of the six phases will assist understand- coders; and finally, it will propose further avenues of ing of the process of decision making. By ‘intelligence’ research in this area. Simon (1977) intends the military meaning of ‘search- There has been very little work undertaken to in- ing the environment for conditions calling for decision’ vestigate how clinical coders think and make deci- (p. 40); or, in Turban and Aronson’s words, ‘reality is sions. Consequently the second part of this inquiry, examined and the problem is identified and defined’ the six levels of clinical coding, is of necessity descrip- (p. 41). tive of the author’s experience in training clinical cod- Norman (1984) defines intention as the ‘mental ers throughout New Zealand (NZ) and Singapore. characterisation of the desired goal’ (cited in Chapter However, there are obvious parallels here with the 5 of Wilson & Walsh 1996). In other words, having various studies on medical decision making in the lit- identified and defined the problem a desired goal is erature (eg, Patel et al. 1984; Friedman et al. 2001; defined. Allen et al. 1998; Eddy 1990). ‘Design’ is defined by Simon (1977) as ‘inventing, developing and analysing possible courses of action’ (p. 41). Wilson and Walsh see this phase as overlap- Models of systematic decision making ping with the intention phase but it is treated here as Simon’s model of systematic decision making in man- a distinct phase. agement has become a de facto standard throughout Simon defines ‘choice’ as ‘selecting a particular the literature on decision making and decision support. course of action from those available’ (Simon 1977: It is made up of four phases through which every de- 41). cision must move: ‘Information extraction and integration’ comes from • intelligence a synthesis by Armbruster and Armstrong (1993) of • design models of the knowledge acquisition process by Guthrie • choice and Mosenthal (1987) and Dreher (1992). The concept • review. captured here is the need to integrate extracted infor- In other words data are presented, formatted as in- mation with prior knowledge. This would enable future formation, a choice is made and the results are exam- similar decisions to be made more easily. ined (Simon 1977). The ‘review’ phase Simon defines as ‘the [assess- By omitting the review phase from their list of ma- ment] of past choices as part of a repeating cycle that jor phases in their chapter on modelling decision sup- leads again to new decisions’ (Simon 1977: 41). Arm-

Health Information Management 2004 ISSN 1322-4913 Vol 33 No 3 Page 79 Reviewed articles bruster and Armstrong characterise this phase as ‘re- interprets their abstraction in the light of the ICD-10- cycling with monitoring and evaluation of progress to- AM classification and assigns codes. On completion of wards the goal’ (cited in Chapter 5 of Wilson & Walsh code assignment the code set is then grouped and as- 1996). Norman regards it as ‘a review of the executed signed a DRG (Diagnosis Related Group). The clinical action to direct further activity’ (cited in Chapter 5 of coder then reviews the DRG to see if it matches what Wilson & Walsh 1996). they read. The coder should also review their code set To summarise, a systematised decision is made to check for inconsistencies and completion. through finding the problem, resolving to solve it, ex- Thus two of the six phases fall under one of the amining possible solutions, choosing the best solution three tasks. The task of abstraction covers the intelli- according to one’s knowledge of the situation, taking gence and intention phases (what are the diagnoses the information learnt and integrating it with one’s associated with this patient and do they need to be knowledge, and then looking to see if the solution cho- coded?). The assignment task takes the design and sen has resolved the problem. choice phases (what codes could be used and which in Turban and Aronson regard Simon’s model as ‘the particular is chosen?). Review picks up the final two most concise and yet complete characterisation of phases: information extraction and integration and rational decision making’ (Turban & Aronson 2001: review (what is there to be learnt here and does this 41). However, the six-phase synthesis of Wilson and code fit?). Walsh (1996) more adequately explains the reason- ing and decision making process undertaken by clini- cal coders, both novice and expert, than does the Clinical coders simplified three-phase model. As shall be seen, the The levels of proficiency in clinical coding that are con- information extraction and integration phase is very sidered here are based upon the several tiers in the important to clinical coders’ decision making. The Clinical Coding Structure as implemented at Auckland review phase will also be seen to be important as District Health Board (Wissmann 1999). People can clinical coders develop their knowledge. enter the profession at any of the first three levels, depending on their previous qualifications. There are aspects other than those covered in the Decision types following dicussion that are taken into consideration in Simon’s classification of decisions into programmed advancing a clinical coder through the levels, such as and non-programmed (Simon 1977: 45ff.) is a very communication with clinicians and data analytical cogent one for the clinical coding profession. Some of skills. These do not, however, impinge directly upon the decisions that must be made are highly pro- the decision making that clinical coders use in their grammed; that is, the decision is a matter of follow- day-to-day work. ing a particular process. In clinical coding, the situation has occurred so often that there is a specific Beginner level coding standard telling the clinical coder exactly what to do (National Centre for Classification in Health The beginner has neither knowledge of medical termi- [NCCH] 2000). For example, Standard 0026 states: nology nor of clinical coding, and cannot really be con- ‘Where the reason for admission is stated as being sidered to be a clinical coder. The first step towards for a clinical trial or drug monitoring, the condition of clinical coding proficiency is to learn medical terminol- the patient should be assigned as the principal diag- ogy (including basic and physiology). The nosis. Z-codes are not necessary in these cases’ (p. decisions at this level are curtailed because of the di- 12). Other sources of programmed decisions are rul- rective nature of the courses and texts; however, the ings by the New Zealand Coding Authority and pub- six-phase model is relevant as a learning process. lished answers to Frequently Asked Questions These six phases are very similar to those in Acceler- (FAQs), both in NZ and Australia. ated Learning techniques (Rose & Nicholl 1997). (The Some decisions are non-programmed in that, to six steps in Accelerated Learning are Motivating Your quote Simon, ‘the system has no specific procedures Mind; Acquiring the Information; Searching out the to deal with a situation . . . but must fall back on what- Meaning; Triggering the Memory; Exhibiting What You ever general capacity it has for intelligent, adaptive, Know; and Reflecting on How You’ve Learnt). problem-oriented action’ (Simon 1977: 47). For clini- cal coders this occurs when there is no coding stan- dard or other source as above to cover a particular Trainee coder level coding situation. The Clinical Coder’s Creed (NCCH Having successfully completed a recognised course in 1998) states the general principles under which a medical terminology, the trainee clinical coder then clinical coder operates. commences a course of training in using the clinical classification and the standards surrounding it. In some , this is an informal on-the-job process. The clinical coding process In others, such as Auckland District Hospital Board There are three primary tasks in clinical coding: ab- (DHB), this process is formalised. People may enter straction, assignment and review. ‘Abstraction’ is here the profession at this level if they have already com- defined as reading the entire record and analysing the pleted some recognised course in medical terminology information therein to find out what was wrong with or can demonstrate a sufficient knowledge based on a the patient, how it happened and what was done clinical background in , nursing or one of the about it. ‘Assignment’ occurs when the clinical coder allied health professions.

Health Information Management 2004 ISSN 1322-4913 Vol 33 No 3 Page 80 Reviewed articles

The informational needs at this level are sourced there is a tendency to blindly code whatever is found, from senior coders with whom the trainee is working rather than determining a coherent whole. or from the tutors of the coding courses. The coding The design and choice phases are sometimes the exercises and discharge summaries that are used as most challenging at the entry level. Every code seems teaching resources are chosen to highlight aspects of possible and picking one can be a daunting task. At the classification. This means that the intelligence Auckland District Health Board, we help coders at this phase of decision making concentrates on interpreting level in two ways. The first is that entry-level coders the question in the light of the aspect of the classifica- focus on one specialty at a time and then slowly add tion being studied. specialties as they increase their knowledge. The sec- One would hope that the intention phase is to find ond is that an experienced coder checks all work done the right code. The design phase varies at this level by an entry-level coder. We ask them to always depending on the background of the trainee. If the choose a code and then the feedback given passes to trainee has a clinical background, information is often both the information integration and review phases. gathered as if it was going to be applied in a clinical After receiving the same issue back several times, the context. However, if the trainee comes from a non- entry-level coder learns techniques of checking their clinical background, there will be a more limited range own work before passing it over. A further effect is the of options through which to negotiate. Assuming that initial formation of the coder’s knowledge scripts. the course is well designed, the design phase should have resulted in only one final option for the choice Competent level phase and information integration should flow auto- matically as the trainee is led in a structured fashion Many clinical coders reach this level and choose not to to build up their knowledge. (or are not able to) progress further. They continue to Review occurs when the trainee either checks their develop their coding skills, but never progress to the answers with the sample answers or receives an as- more integrated stages of decision making that the sessment. ‘What did I do right?’ ‘What did I do advanced coder reaches. The reasons for this are mul- wrong?’ ‘Why was it right or wrong?’ tifactorial, but appear to be based on confidence and the learning style of the individual. (For more informa- tion on learning styles refer to Chapter 2 of Rose & Entry level Nicoll, 1997). Successful completion of a clinical coding course gives The competent coder has demonstrated an ability the new clinical coder a framework on which to build to code accurately in a timely fashion and thus makes real-life decisions. The time a clinical coder spends at many decisions without the need to refer them to an- this level depends on their ability to provide accurate other coder for agreement. This can lead to a degree and timely data. Like nurse graduates, coding gradu- of overconfidence, which is balanced by regular audits. ates ‘who enter the workforce after graduation lack At this level, the clinical coder has learnt which docu- skills in solving unprecedented clinical problems with ments do not need to be read as part of the abstrac- multiple foci’ (Oliver et al. 1999: 2). tion or intelligence phase, and the notes the clinical In the coding course, the trainee saw neatly typed, coder makes are considerably shorter than those of well-constructed coding exercises. Now, however, the the entry-level coder. Time spent in the intention hand-written, badly organised documents that make phase is also shorter, as the knowledge scripts are of up clinical documentation have to be analysed for di- greater depth. agnoses and procedures. This analysis, or ‘intelli- The design phase for a competent coder consists of gence’, phase remains a constant throughout all the working through the various options in the classifica- ensuing levels, but it is performed differently at the tion and often coming up with several possibilities. various levels. At this level the clinical coder reads Through discussion with colleagues or working through everything in the , taking numerous the classification manuals a choice is made. The pref- notes as they do so. erence, as with medical professionals, is discussion Then the ‘intention’ has to be resolved. Which of with colleagues (Rosenberg & Sackett 1996). these notes relate to each other? Which can be ig- Competent coders are constantly on the lookout for nored? Which are important? The major fault of clinical new information that may be integrated into their coders at this level is ‘over-coding’ (ie, coding more knowledge scripts. This is particularly so in those who diagnoses or procedures than is necessary to repre- come from a non-clinical background. sent the episode of care). This comes about because Review at this level is in two forms. Firstly, the the clinical coder has not yet integrated the classifica- competent coder is learning to develop the instinct to tion and the Clinical Coders’ Creed with the ‘real world’ detect when a code set looks ‘wrong’. Secondly, regu- as presented through clinical documentation. lar audits confirm accuracy. Once the intention phase is entered the clinical or non-clinical background of the clinical coder comes to the fore. The danger for coders with a clinical back- Accredited level ground is that their coding could be based on their An accredited clinical coder has sat a further examina- clinical knowledge instead of only that which is docu- tion that demonstrates a thorough knowledge of clini- mented. This problem arises because of the knowledge cal coding using a particular classification across a scripts (Edwards 1995, 1997) that are carried across wide range of patient types and clinical specialties. from their original clinical context. It is very difficult to Whether this level represents an intermediate step ignore prior knowledge. For the coder with a non- between competent and advanced is as yet undeter- clinical background, there is no prior knowledge and mined, as the national accreditation program is cur-

Health Information Management 2004 ISSN 1322-4913 Vol 33 No 3 Page 81 Reviewed articles rently in abeyance due to the changes in the classifica- In addition, the hospital may be fined for under- tion over the previous 6 years. performing on its contract. Research is also dependent upon useful data. The Decision Support Unit at Auckland DHB receives sev- Advanced level eral requests every day for data for research pur- Advanced clinical coders have demonstrated a very poses, most of it dependent upon the codes assigned high degree of consistent accuracy and have sub- by clinical coders. Without accurate data from the sumed the Clinical Coders’ Creed to the point where all Clinical Coding Service much valuable time on the part the aspects of knowledge involved in clinical coding of the researcher will be wasted and, at worst, the are fully integrated. They will still consult other clinical wrong conclusions will be reached, which may then coders on individual cases, but this is often to support impact on patient treatment. decisions already made. As they analyse the documen- tation they are performing the intelligence, intention and design phases simultaneously. The notes they The next step make are minimal and usually act more as reminders The next step in the process of understanding how of what they have read. Often the notes will consist of clinical coders make decisions is to conduct a study to a list of codes and abbreviations. This is not a memo- test whether the models presented above are valid. risation process, but rather signifies increased lateral Such a study will be piloted later in 2004 and then, it experience. The advanced clinical coder happily solves is hoped, extended in 2005. problems never seen before, a state described by The findings from this study will assist the design- Oliver et al. (1999: 2) as ‘unprecedented clinical prob- ers of training and quality management programs for lems with multiple foci’. These clinical coders can per- clinical coders to determine how best to direct their form at similar levels of expertise with classifications programs. Both of these areas are of interest to this that are new to them (Department of Health and Fam- author. ily Services 1997; Grozef et al. 1999). An interesting part of the study, if it is possible to A further part of the intention phase for advanced expand it, will be to compare clinical coders from the coders is that they automatically check for information different Australian states and territories, New Zealand that they know arises from the information they al- and Singapore. Does the region in which a clinical ready have; for example, the diagnosis of hyperpara- coder works make a difference in how they make deci- thyroidism on a surgical case will trigger a search for sions? Or is it dependent on which training course they documentation of the cause. This is because the inci- undertook? dence of secondary hyperparathyroidism (particularly Another area that would benefit from an under- renal) is much higher than that of primary. standing of the decision-making process for clinical The choice phase at this level is frequently a simple coders is decision support. The big challenge in deci- one based on the integrated knowledge of the classifi- sion support is ‘working out what to give people so cation. that each person individually can learn what they New information is integrated readily into the need to learn, when they need to learn it, how they knowledge scripts an advanced coder has already de- want to learn it’ and where they want to learn it veloped. (Peers 1999: 4). The review phase, similarly to the competent level, A clinical coding decision support system needs to has two aspects. The first is the regular auditing of combine knowledge of best practice in all clinical areas their work by other coders. The second, the automatic with expert knowledge of the classification in use by scanning of the list of codes selected for anomalies. the clinical coder. Ideally it would involve Artificial In- telligence so that the system itself would learn how to present information to individual clinical coders and be Why do clinical coders need to make self-auditing. accurate decisions? There are several implications of accurate decisions by Conclusion clinical coders. The two most important centre on money and research. A clinical classification is an arbitrary structure that is The Clinical Coding Service of any hospital is ex- laid over the top of the ‘real world’. In other words, pected to provide data so that the optimal level of the real world is much more complicated than can be funding for each patient event is received. The more represented by any of the currently available disease accurate a clinical coder, the more sure the hospital classifications, and indeed procedure classifications. can be that it has received the correct funding for an Even the procedure classification developed for future event. There are, of course, other influential factors, use in the United States of America, The International including the quality of documentation and the nature Statistical Classification of Diseases and Related of the contracts a hospital has with its funders. How- Health Problems, Tenth Revision, Procedure Classifica- ever, the first point of responsibility is, in this author’s tion (ICD-10-PC), with its very large number of poten- experience, the coding. The funding issue can be tial codes (52 x 109), still does not permit some of the starkly expressed in terms of a coder’s salary. If a finer nuances of surgical work to be captured. This coder was to not abstract, and therefore not code, 10 necessitates the employment of trained people who pacemaker insertions over the course of a year, then can interpret the real world in terms of the classifica- that coder’s salary will have been lost to the hospital. tion and the classification in terms of the real world.

Health Information Management 2004 ISSN 1322-4913 Vol 33 No 3 Page 82 Reviewed articles

Simon’s four-phase model of how the interpretative NCCH (2000). The International Statistical Classification of decisions are made by these people (clinical coders) is Diseases and Related Health Problems, 10th Revision, limited because of the omission of the important Australian Modification, Volume 5. Australian Coding nd phases of intention and information extraction and Standards for ICD-10-AM, 2 edn. Sydney, National Centre for Classification in Health. integration. The intention phase has been shown to be involved in working out what information needs to be Norman, D.A. (1984). Stages and levels in human-machine coded and what can be left out. The information ex- interaction. International Journal of Man-Machine Studies 21: traction and integration phase has been shown to be 365-375. important for refining the process for future decisions. Oliver, M., Naidu, S., Koronios, A., McIntyre, Y. and Stalley, As Oscar Wilde (1966) once said: ‘It is a very sad A. (1999). A multimedia-based clinical decision support sys- thing that nowadays there is so little useless informa- tem for graduate nurses entering the workforce. Proceedings of HIC 1999. Hobart, Society of Australia: tion’ (p. 1203). 262-267. Patel, V.L., Arocha J.F. and Kaufman, D.R. (1984). Diagnostic References reasoning and medical expertise. The Psychology of Learning and Motivation 31: 187-252. Allen, V.G., Arocha, J.F., and Patel, V.L., (1998). Evaluating evidence against diagnostic hypotheses in clinical decision Peers, J. (1999). Turning information into knowledge and making by students, residents and physicians. International knowledge into more productive habits. Proceedings of HIC Journal of Medical Informatics 51: 91-105. 1999. Hobart, Health Informatics Society of Australia: 47-52. Armbruster, B.B. and Armstrong, J.O. (1993). Locating in- Rose, C. and Nicholl, M.J. (1997). Accelerated learning for formation in text: a focus on children in the elementary the 21st Century. New York, Dell Publishing. grades. Contemporary Educational Psychology 18: 139-161. Rosenberg, W.M. and Sackett, D.L. (1996). On the need for Department of Health and Family Services (1997). ICD-10- evidence-based medicine. Therapie 51(3): 212-217. AM Impact Assessment Project. Final Report. Australia, Coo- Simon, H.A. (1977). The New Science of Management Deci- pers and Lybrand Consultants. sions, 3rd edn. Englewood Cliffs, NJ, Prentice Hall. Dreher, M.J. (1992). Searching for information in textbooks. Turban, E. and Aronson, Y.E. (2001). Decision Support Sys- Journal of Reading 35: 364-371. tems and Intelligent Systems, 6th edn. Englewood Cliffs, NJ, Eddy, D. (1990). Clinical decision making: from theory to Prentice Hall. practice: anatomy of a decision. Journal of the American Wilde, O.F.O. (1966). Complete Works of Oscar Wilde. Lon- Medical Association 263(3): 441-443. don, Collins. Edwards, D. (1995). Sacks and psychology. Theory and Psy- Wilson, T. and Walsh, C. (1996). Information behaviour: an chology 5: 579-596. interdisciplinary perspective. British Library Research and Edwards, D. (1997). Discourse and cognition. London and Innovation Report 10. , Wetherby, West Beverly Hills, Sage. LS23 7BQ. U.K., British Library Research and Innovation Cen- tre. Friedman, C., Gatti, G., Elstein, A., Franz, T., Murphy, G. and Wolf, F. (2001). Are clinicians correct when they believe they Wissmann, S. (1999). Clinical coding structure. Auckland NZ, are correct? Implications for medical decision support. Pro- Auckland District Health Board (internal document). ceedings of MEDINFO 2001. Amsterdam, IOS Press: 454- 458. Grozef, A., Thorpe, C. and Wooding, A. (1999). Orthopaedic Mapping Tables between MBS-E and CPT-4. Wellington, New Andrew Wooding GradDipBus (HlthInfo) Zealand Health Information Service (unpublished). Team Leader – Clinical Coding (Medical, Surgical & Child) Guthrie, J.T. and Mosenthal, P. (1987). Literacy as multidi- Auckland City Hospital mensional: locating information and reading comprehension. Auckland District Health Board Educational Psychologist 22: 279-297. Private Bag 92-024 National Centre for Classification in Health (NCCH) (1998). Auckland, New Zealand Clinical Coder’s Creed. Coding Matters 5(3): 11-12. Email: [email protected]

Health Information Management 2004 ISSN 1322-4913 Vol 33 No 3 Page 83