Computerized Genetic Risk Assessment and Decision Support in Primary Care Andrew S. Coulson1, David W. Glasspool1, John Fox1 and Jon Emery2 1. Imperial Cancer Research Fund, Advanced Computation Laboratory, PO Box 123, London WC2A 3PX. 2. Imperial Cancer Research Fund, General Practice Research Group, Division of Public Health and Primary Health Care, Institute of Health Sciences, Old Road, Headington, Oxford OX3 7LF. Correspondence to Andrew Coulson, Advanced Computation Laboratory, Imperial Cancer Research Fund, PO Box 123, London WC2A 3PX. Email: asc@acl.icnet.uk. This research was supported by an award from the UK Economic and Social Research Council (award no L127251011) under the Cognitive Engineering Programme. Jon Emery is supported by the Cancer Research Campaign. -1- Abstract Public awareness of the availability of genetic testing threatens to put severe strain upon genetics clinics in the near future. General practitioners (GPs) could help avert this problem by making an initial genetic risk assessment and acting as gatekeepers to specialist services. However, studies in the United Kingdom suggest that few GPs feel they have the requisite skills for taking family history details and making an appropriate referral decision. They are also poorly served by computer-based pedigree programs, which do not cater to the specific needs of a general practice consultation. To address these issues, a new computer application called RAGs (Risk Assessment in Genetics) has been designed. The system allows a doctor to create family trees and assess genetic risk of breast cancer. RAGs possesses two features that distinguish it from similar software: (a) a user-centred design, which takes into account the requirements of the doctor-patient encounter; (b) risk reporting using qualitative evidence for or against an increased risk, which the authors believe to be more useful and accessible than numerical probabilities are. In that the system allows for any genetic risk guideline to be implemented, it can be used with all diseases for which evaluation guidelines exist. The software may be easily modified to cater for the amount of detail required by different specialists. -2- Introduction One of the results of the revolution in molecular genetics is the increasing availability of tests for the presence of genetic abnormalities which predispose certain individuals to contracting particular diseases. To date, genetic markers for over 50 diseases have been identified, and tests are available or being developed for conditions such as cystic fibrosis, Tay-Sachs, Huntington’s, certain forms of Alzheimer’s and muscular dystrophy, and various rare cancers such as retinoblastoma, Wilms’ tumour and Li-Fraumeni syndrome. Predisposing genes for more common cancers have also been located, such as those for breast cancer (the BRCA1 and BRCA2 genes) and certain forms of colon cancer (MLH1, MSH2). Coupled with these advances is the increased media attention upon genetic testing, which has resulted in a heightened awareness of the subject by the general population. Such public knowledge, although desirable, is likely to create a steady growth in the demands for genetic screening 1. In that there are only one or two consultant geneticists per million people in the UK 2, this increase threatens to put considerable strain upon genetics clinics in the near future. Much of the problem here will stem from clinics’ having to deal with people who are at low genetic risk, yet who have still been referred because one or two members of their family had the disease in question3. The situation is further compounded by the current unreliability of genetic screening. Direct mutation testing is available for the genes listed above, but interpretation of the test results is complicated by the large number of mutations in some of these genes (e.g. BRCA1 / BRCA2) and by uncertainty about their clinical meaning4. Thus in the UK for example, where as many as 30,000 women are estimated to carry a mutation in BRCA1, general population screening may not at present constitute a responsible course of action. There is, however, a simple practical solution to the problem of restricting advanced genetic services only to appropriate members of the population. This involves all patients being properly evaluated in primary care, with general practitioners (GPs) acting as the gatekeepers to specialist services. In such cases, doctors would examine family histories in order to identify those for whom genetic testing may be worthwhile5. It would appear that GPs accept this gatekeeper role in genetics, believing it is appropriate for them to take detailed family pedigree information in order to make the relevant referral decisions6 7 8. -3- Initially therefore, this whole state of affairs seems promising. An increased public demand for genetic risk evaluation will mean a greater probability of identifying those who are genuinely at risk. At the same time, GPs acting as the first point of contact will reduce the pressure on genetics clinics by filtering out obviously low risk cases. In addition, the use of family pedigree analysis can offset the current uncertainty associated with large scale indiscriminate genetic testing. A major barrier to this process, however, is that most GPs are unfamiliar with genetic risk assessment and are consequently unprepared to provide such a service for their patients. Emery et al. 7, for example, found that even doctors who had attended courses on cancer genetics felt themselves to be poor at evaluating genetic risk and were uncomfortable doing so. GPs admitted that their information gathering techniques were often incomplete and haphazard, and that they did not have the skills to draw proper pedigrees. There was also concern about what could be termed “guideline chaos”: practitioners were inundated with paper guidelines of varying quality, which even if retrieved at the appropriate time would only serve to confuse the risk assessment process. As a result of such problems, GPs would either (a) refer all patients, (b) assess risk using a simple heuristic based on an incomplete recall of referral guidelines, or (c) simply reassure patients and do nothing. None of these actions could be said to be desirable ‘gatekeeping’ behaviours, but practitioners were usually left with little choice. It has also been noted that GPs may be unwilling to do genetic risk assessment because of the sheer length of time involved in constructing a pedigree, evaluating the results and giving advice 8 9. Computerized Assistance Suchard et al.8 concluded that the problems of inadequate genetic expertise and time pressures on GPs could be alleviated by “practice-enabling” strategies10, such as the use of computerized aids. Software that can both draw pedigrees and assess genetic risk therefore has the potential to restore a competent gatekeeping role for GPs in this field of medicine. In our view, designing such a system for practitioners entails thinking about at least two factors that are not addressed by current genetics software: (a) an appropriate user-interface design; (b) an effective means of communicating risk and uncertainty. Appropriate user-interface design. Given the time pressures upon GPs, it is unreasonable to expect them to have to deal with a complex piece of software in order to create family trees and make genetic risk assessments. In addition, although demand for genetic advice may increase in the future, currently only one to two patients per month present to a GP about their family history of a common disease 11. Therefore, since a -4- genetic risk evaluation program is unlikely to be used very often, it must be simple enough to require as little relearning as possible. Here existing pedigree-drawing software is inappropriate12. Systems such as Cyrillic13 and Progeny14, for example, are excellent state-of-the-art commercial pedigree-drawing packages. However, these and similar applications are primarily designed as tools for professional geneticists, and thus have a full complement of menu items and menu icons which require familiarization over time. The use of such systems by GPs is impractical: practitioners simply do not have the time to learn and continually top up their skills on complex applications that will only be accessed relatively rarely. Time pressure on GPs during consultations also necessitates an application that is quick to use. For these reasons, we view the appropriate software as requiring a design and presentation that is easily comprehended, affords fast (re)learning and allows the rapid construction of pedigree charts. Effective risk communication. A major issue in the design of decision support software is the notorious difficulty of effective communication about risk and uncertainty. People apparently find uncertainty information difficult to use. We believe that this is because they do not use probability theory when making everyday decisions, while risk is generally expressed in terms of numerical probability values (by programs such as Cyrillic, for example). Instead, people may think more naturally in terms of arguments for or against particular courses of action. Degrees of risk and uncertainty may thus be more easily comprehensible (and more readily related to appropriate action) when numerical values are augmented with or replaced by qualitative arguments or ‘reasons to believe’ alternative possibilities15 16. Argumentation17 18 19 is a qualitative logical procedure that allows decisions to be made by weighing up arguments for or against a particular proposition. This powerful and formally well-developed technique for risk assessment in the absence of precise numerical estimates of uncertainty has already proved its worth in a number of practical applications20, including drug prescription21 and medical imaging22. The use of argumentation can allow a decision support system to communicate in a more naturalistic way about degrees of uncertainty. We now describe an application designed to meet the deficiencies in current genetic risk software outlined above. -5- Overview of the RAGs Software RAGs (Risk Assessment in Genetics) is a decision support application that is part of a larger medical decision making system (“PROforma”) being developed at the ICRF. RAGs is designed to assist general practitioners in assessing and communicating risk information to women who are worried about their genetic risk of breast cancer. The software allows the user to construct a pedigree diagram of the patient’s family, incorporating information on the known incidence of cancers. This pedigree information is evaluated via a risk calculation protocol that assesses genetic risk level, makes recommendations for patient management and provides an argument-based explanation for its conclusions. The specific program features are outlined below. Data entry. Relevant information for all family members is entered via a standard data entry form (Figure 1). Shaded fields on the form indicate information which is required before a risk assessment can be made. The user is prevented from adding a person to the tree until all such information is entered. Family tree icons. Once a person’s information has been entered, a representative icon appears in a graphics window. Icons are colour coded according to cancer type, with colours mixed appropriately if a person has more than one form of cancer. Two types of pedigree representation are available. The “Standard” representation consists of ‘small people’ icons which take patient sensitivity into account, e.g. by not having lines drawn across icons for deceased relatives. The alternative, “Expert” representation uses standard genetic pedigree symbols. Building the tree. Family members are added by clicking on any of the person icons currently on the screen. A form is presented that gives the user options to change that person’s details, add family members or remove that person from the pedigree. When a family member is to be added, the relevant data entry form appears. In this way one can quickly create an entire pedigree. -6- Figure 1: Standard data entry form in RAGs. Risk assessment and communication. RAGs evaluates genetic risk for a given patient by using the PROforma argumentation-based software system20 23 24 25. This is a generic technology for implementing clinical decision support and ‘intelligent’ clinical guidelines. For the present version of RAGs, a guideline was devised for evaluating the genetic risk of breast cancer in women. The protocol is expressed as a set of 23 rules, each of which relates to a person’s increased or decreased genetic risk. For example: Each first degree relative (i.e. parent, child or sibling) with breast cancer increases the presenting patient’s genetic risk. -7- Genetic risk is decreased if the presenting patient is over 40 and has not yet developed cancer. The rules allow an overall risk score to be computed. Based on this score a patient is put into a low, moderate or high genetic risk category, and the appropriate referral advice is given, specifically: Low risk: patient can be reassured and managed in primary care Moderate risk: refer to breast clinic High risk: refer to clinical geneticist Both the assignment of risk categories and the referral advice are, like the guideline, components of the program that can be altered as required. A sensitivity analysis of the system was performed by comparing the risk assignments in RAGs with equivalent assignments based on values given by Cyrillic13, which calculates the genetic risk of breast cancer as a numerical probability using the Claus dataset26. In all 50 test cases, RAGs agreed with Cyrillic in its assignment of patients into ‘low’, ‘moderate’ and ‘high’ genetic risk categories. Figure 2 shows the family tree for a hypothetical patient, Karen, after a risk assessment has been performed. RAGs computes a risk score for every possible path of inheritance and then chooses the highest-risk path on which to base its decision. This path is highlighted in the program. Some of the results of the risk assessment are also displayed in the figure. Three kinds of information are available: 1. General Explanation, which gives a standard overview of what the software does and how it performs its risk assessment. 2. Referral Advice, which provides a short, simple referral recommendation based upon the risk calculation. 3. Reasons for Advice, the text for which is shown in the figure. This button presents the user with an explanation of how the system came to its decision. It does so by providing the arguments for and against believing a cancer-predisposing gene to be present in the -8- Figure 2: Explanation generated by RAGs for its risk assessment and referral advice. patient’s family. These arguments can be directly cross-referenced with the pedigree diagram. The user may continue to alter the tree after a risk assessment has been calculated, and then request a new risk assessment if desired. Discussion There are a number of pedigree drawing and genetic risk assessment programs available commercially, but RAGs has certain features that distinguish it from similar software. -9- Interface Design As noted in the Introduction, not all GPs are adept at the use of computers, yet most pedigree software tends to have a large array of options, usually represented by menu items or menu icons. In contrast, the RAGs system has been designed to keep choices and procedures to a minimum, with those available always indicated on screen. The application can be mastered within a few minutes even by those with minimal computer experience, and requires little investment in time either for initial learning or for re-familiarization. Such features are essential for infrequently used software. There are two more positive effects of this simplicity. First of all, it will help the doctor maintain an air of competence. Insofar as the software itself is not contributing an added level of confusion and stress to the consultation, the GP can relax and focus his/her attention on the patient. This in turn gives the patient an increased confidence in the way the doctor is addressing her concerns. Secondly, the simple and informal nature of the system encourages practitioners to share the screen with their patients. This activity appears to promote a sense of joint effort and shared understanding 7, and our own experiences in trials with volunteer GPs suggest that it is also likely to enhance the quality of the data entered. Effective program development. The RAGs system is being refined through feedback by GPs, who have been using it in mock consultations with actors taking the part of patients. This method of program enhancement by testing it in a real work setting has been termed an “action design” approach to software development27. The technique has already raised issues relating to data input and presentation that would have been missed had we just requested a simple ‘wish list’ of features 7. For example, a prototype version had the risk information present itself on screen as soon as the user clicked on the “Create Report” button. This feature proved to be undesirable in the mock consultations, as it did not afford GPs the opportunity to absorb the information before discussing it with the patient, resulting in the loss of a relaxed and empathic doctor-patient interaction. The current software version simply gives a message that the report is ready, and waits for one of the three information buttons (Figure 2) to be clicked. Thus, once the practitioner or qualified assistant has entered the family tree details, the results can be accessed in a more controlled fashion. For example, some GPs might prefer to read the report in the absence of the patient and discuss the results in a second consultation. The doctor therefore has the chance to adopt the appropriate attitude in light of the results, and talk about them in an informed manner. To our knowledge, other programs in general do not take into account factors such as patient sensitivity in their design. - 10 - Risk Assessment and Reporting The RAGs software centres around a qualitative reason-based decision procedure. This allows the outcome of an assessment to be communicated in what we believe to be a more useful form than simple numerical risk estimates. The most immediate benefit of such a process is that the clinician and the patient can understand how the program is arriving at its conclusions, and even have the opportunity to take issue with the presented arguments. Unlike other genetic risk assessment programs, therefore, RAGs allows its users to make an informed choice as to what to do with the advice it gives. Paradoxically such advantages may be particularly clear in the case of those who are not at increased risk due to genetic factors. Breast cancer is sufficiently common that many women have a number of affected relatives, yet genetic predisposition is relatively rare. Most women approaching their GPs with concerns about genetic predisposition are thus unlikely to be at increased risk. In such situations it can be difficult to reassure the patient, and here the argumentation-based explanations provided by the RAGs system may be considerably more effective than a simple predicted risk level. The fact that one can see what arguments the program is using may have further, more general benefits. For example, different national or international authorities may employ different guidelines for genetic risk, making it even more imperative that the user knows exactly what rules are being used in an assessment. In that the PROforma risk assessment part of the software is a separate system, it can easily be adapted to implement whatever guideline the user wishes. As a result, any required changes to a guideline are simple and quick. Future Developments Emery et al.7 found that GPs’ responses to the RAGs system as a whole were positive. Nevertheless, a number of improvements and refinements were suggested. In particular, unpublished results from this study revealed that practitioners wanted RAGs to function at two levels, depending on the type of care to be provided and the amount of information the patient wanted: 1. ‘Gatekeeper’ mode - where patients are triaged into risk categories and are given the appropriate advice. This is what RAGs currently does, with its main purpose being to allow the GP to identify those patients that can be reassured and those who require referral to specialist services. - 11 - 2. ‘Primary care geneticist’ mode - for GPs who wish to become more involved in genetic counselling. This would entail the program being capable of providing much more detailed information than it does at present. For both these levels, the practitioners still felt it was important to have access to the evidence supporting the risk assessment and management suggestions. With regard to the ‘primary care geneticist’ mode of functioning, we are investigating how to adapt the RAGs program to cater for the needs not only of GPs, but for genetics associates and other specific groups of users. Changes here may include giving information on screening and prevention, explaining the role of genetic testing, and providing numerical risk estimates to supplement the argument-based conclusions. Also as discussed, the implementation of alternative guidelines is very straightforward, and we are currently looking into producing a version of the program that gives referral advice relating to colorectal cancer. It is also hoped that all these modifications would be refined in the same way as with the original system, i.e. as a result of detailed evaluation of the software in an actual clinical environment. References 1. Bell, J. The new genetics: The new genetics in clinical practice. British Medical Journal 1998; 316: 618-620. 2. Kinmonth, A.L., Reinhard, J., Bobrow M. & Pauker, S. The new genetics: Implications for clinical services in Britain and the United States. British Medical Journal 1998; 316: 767-770. 3. Harris, R. & Harris, H. Primary care for patients at genetic risk. British Medical Journal 1995; 311: 275-276. 4. Collins, F. BRCA1 – Lots of mutations, lots of dilemmas. New England Journal of Medicine 1996; 334: 186-188. 5. Scheuner, M.T., Wang, S., Raffel, L.J., Larabell, S.K. & Rotter J.I. Family history: a comprehensive genetic risk assessment method for the chronic conditions of adulthood. American Journal of Medical Genetics 1997; 71: 315-324. 6. Fry, A., Campbell, H., Gudmundsdottir, H., Rush, R., Porteous, M., Gorman, D. & Cull, A. General practitioners’ views on their role in cancer genetics and current practice. Family Practice, in press. - 12 - 7. Emery, J., Walton, R., Coulson, A., Glasspool, D., Ziebland, S. & Fox, J. Computer support for recording and interpreting family histories of breast and ovarian cancer in primary care (RAGs): qualitative evaluation with simulated patients. British Medical Journal 1999; 319: 32-36. 8. Suchard, M.A., Yudkin, P. & Sinsheimer, J.S. Are general practitioners willing and able to provide genetic services for common diseases? British Journal of General Practice 1999; 49: 45-46, and Journal of Genetic Counseling, in press. 9. Watson, E., Shickle, D., Qureshi, N., Emery, J. & Austoker, J. The ‘new genetics’ and primary care: GPs’ views on their role and their educational needs. Family Practice 1999; 16: 420-425. 10. Davis, D.A., Thomson, M.A., Oxman, A.D. & Haynes, B. Evidence for the effectiveness of CME: A review for 40 randomized controlled trials. Journal of the American Medical Association 1992; 268: 1111-1117. 11. Emery, J., Watson, E., Rose, P. & Andermann, A. A systematic review of the literature exploring the role of primary care in genetic services. Family Practice 1999; 16: 426-445. 12. Emery, J. Computer support for genetic advice in primary care. British Journal of General Practice 1999; 4: 572-575. 13. Chapman, C. Cyrillic pedigree drawing software (v. 3.0.1). Cherwell Scientific Publishing: Oxford, 1999. www.cyrillicsoftware.com (www.cherwell.com.uk). 14. Progeny Software. Progeny pedigree drawing software (v. 2.0.06). Genetic Data Systems, LLC: Indiana, 1998. www.progeny2000.com/ 15. Fox, J., Hardman, D., Krause, P., Ayton, P. & Judson, P. Risk assessment and communication: A cognitive engineering approach. In Proceedings of Expert Systems ’95. Cambridge University Press, 1995. 16. Krause, P., Fox, J. & Judson, P. Is there a role for qualitative risk assessment? In Proceedings of the 11th Conference on Uncertainty in Artificial Intelligence 1995: Montreal Quebec, Canada. 17. Fox, J., Krause, P. & Ambler, S. Arguments, contradictions and practical reasoning. In B. Neumann (Ed.), Proceedings of the 10th European Conference on AI, 1992: 623-627. ECAI92, Vienna, Austria. 18. Fox, J. On the necessity of probability: Reasons to believe and grounds for doubt. In G. Wright & P. Ayton (Eds.), Subjective Probability. John Wiley: Chichester, 1994. 19. Parsons S. & Fox, J. Argumentation and decision making: A position paper. In Proceedings of the International Conference on Formal and Applied Practical Reasoning. Springer-Verlag: Berlin, 1996: 705-709. - 13 - 20. Fox, J. & Thomson, R. Decision support and disease management: A logic engineering approach. IEEE Transactions on Information Technology in Biomedicine 1998; 2(4). Special Issue, December. 21. Walton, R.T., Gierl, C., Yudkin, P., Mistry, H., Vessey, M.P. & Fox, J. Evaluation of computer support for prescribing (CAPSULE) using simulated cases. British Medical Journal 1997; 315: 791-795. 22. Taylor, P., Fox, J. & Todd-Pokropek, A. The development and evaluation of CADMIUM: A prototype system to assist in the interpretation of mammograms. Medical Image Analysis, To appear. 23. Fox, J., Johns, N., Lyons, C., Rahmanzadeh, A., Thomson R. & Wilson, P. PROforma: A general technology for clinical decision support systems. Computer Methods and Programs in Biomedicine 1997; 54: 59-67. Information also available at www.acl.icnet.uk. 24. Fox, J., Johns, N. & Rahmanzadeh, A. Disseminating medical knowledge: The PROforma approach. Artificial Intelligence in Medicine 1998; 14(1-2): 157-182. 25. InferMed. Arezzo guideline authoring language. InferMed Ltd.: London, 1999. www.InferMed.com.uk. 26. Claus, E., Schildkraut, J., Thompson, W.D. & Risch, N. The genetic attributable risk of breast and ovarian cancer. Cancer 1996; 77: 2318-2324. 27. Timpka, T., Rauch, E. & Nyce, J.M. Towards productive knowledge-based systems in clinical organizations: A methods perspective. Artificial Intelligence in Medicine 1994; 6(6): 501-519. - 14 -