Integrating Clinical Decision Support into EMR and PHR

Integrating Clinical Decision Support into EMR and PHR: a Case Study Using Anticoagulation Dave-Gregory CHACKERYa, Karim KESHAVJEE a,b, Kashif MIRZAa, Ahmad GHANYa, and Anne M HOLBROOKc a InfoClin Inc, b School of Health Information Science, University of Victoria c Division of Clinical Pharmacology & Toxicology, McMaster University Abstract. Clinical decision support (CDS) for atrial fibrillation is expected to ease the implementation of often-complex guidelines for atrial fibrillation and anticoagulation. Most clinical decision support systems (CDSS) for anticoagulation are stand-alone systems that do not integrate with electronic medical records (EMR). We have developed an architecture that consists of a computerized CDS that can integrate with multiple EMRs and multiple patient health records (PHRs). The design process revealed some significant issues that were resolved through systematic business/clinical analysis and creative clinical design in the diagnostic and treatment domains. Key issues identified and resolved include: 1) how to correctly allocate existing patients into various CDSS states (e.g., MAINTENANCE, HOLD, DISCONTINUE, etc), 2) identify when a patient becomes eligible for CDSS guidance over time, 3) how the CDSS maintains information about the patient’s anticoagulation state and 4) how to transform vague human-readable concepts to explicit computable concepts. The management of anticoagulation for atrial fibrillation is no easy task and we believe our architecture will improve patient care at all levels and ultimately better balance the reduction of stroke risk while minimizing harms from major bleeding. In addition, the architecture presented is scalable to other treatment guidelines and is scalable to multiple EMRs and PHRs, making it suitable for use in a platform approach. Keywords. Clinical Decision Support, Clinical Decision Support Systems, Computerized Guidelines, Electronic Health Record, Patient Health Record, Atrial Fibrillation, Anticoagulation.

Introduction It is estimated that about 350,000 Canadians live with atrial fibrillation (AF) [1]. Of these, roughly 35% (or 120,000) will go on to have a stroke during their lifetime [2]. With proper anticoagulation (AC) therapy such as warfarin or one of the new oral anticoagulants, the risk of AF related stroke can be reduced by over 60%, saving more than 70,000 people from having a stroke [3]. It has been found that as many as 28% (~100,000) of patients with AF are undertreated, leaving 35,000 patients at virtual certainty of having a stroke [4]. The treatment of AF frequently requires AC, which can potentially harm the patient by causing bleeds. Selecting the appropriate treatment for AF requires careful balancing of benefits and harm. Fortunately the CHA2DS2-

VASc tool, which quantifies the risk of stroke and the HAS-BLED tool, which quantifies the risk of bleeding allows clinicians and patients to balance the potential benefit versus harm of AC. There are many benefits that are expected from using CDSS for anticoagulating patients who have AF. These benefits include increasing the number of people with AF who are treated appropriately with AC and reducing the number of people that are at higher risk of harms than benefits [5]. A comparison of AC guidelines for AF and real-world warfarin prescription has shown that there is a high discrepancy between actual warfarin use and guideline recommendations [6]. This suggests that AC decisions are not being based on systematic evaluation of stroke and bleeding risk [6]. A clinical decision support system may improve AC by analyzing stroke and bleeding risk, providing AC information so that clinician and patient can have an informed discussion on therapy choices [6]. It has been shown that computerized CDS does lead to AC improvement in the number of patients that were able to achieve therapeutic AC [7]. There are several standalone Anticoagulation CDSSs in use or in testing [8] [9]. In an era of increasing EMR use, physicians find standalone solutions to be cumbersome and difficult to integrate into practice. We have developed an architecture for integrating CDSS into EMRs and PHRs using anticoagulation as a case study. The aim of this design project was to investigate how to integrate clinical decision support (CDS) for AF into electronic medical records (EMR) and patient health records (PHR).

1. Methods We used standard business analysis and architecture development tools for the design process [10]. The design team included a business analyst, a clinical subject matter expert, a project manager and a clinical informatics expert. A literature search in Medline was conducted to identify use of CDSS for AF. Six key business analysis artifacts were developed: 1) Work flow analysis, 2) CDS anticoagulation guideline algorithm (including Triggers Document and Data Dictionary), 3) Object-oriented domain analysis, 4) Requirements specification document, 5) Graphical user interface mockups and 6) Application Programming Interface requirements. We used an iterative process to develop the artifacts and obtained external feedback from an anticoagulation clinical expert on a periodic basis.

2. Results Our initial plan was to develop a state-less CDSS; i.e., one where the patient’s anticoagulation state would be determined in real-time, based on clinical criteria in the medical record. It turned out that this was not possible because some patient states are indistinguishable from others using just clinical data in the medical record. Our analysis identified several issues that need to be solved when integrating a CDSS into an EMR and PHR. These include: 1) Correctly allocating patients already on treatment to the right state in the CDSS (see Initializing the CDSS); 2) Knowing when a patient has become eligible for CDSS during an encounter (see Determining Eligibility below); 3) Keeping up with and managing the changing patient states during

the course of treatment. Patients do and will change states and those states need to be explicit and transparent; e.g., patient’s anticoagulant being put on HOLD and then reinitiated. (See Managing State Transitions below) Using clinical data alone to determine patient state is not possible and could expose the patient to harm that would be difficult to monitor; and, 4) Making vague terms and statements explicit. Humanreadable guidelines are peppered with terms such as ‘when stable’ and ‘increase frequency of monitoring’, which don’t have an operational definition that can be computed. 2.1. Initializing the CDSS When the CDSS is initialized, there are already many patients in the EMR database who would be eligible for CDSS recommendations. There are two main categories of patients that are detected: 1) ‘Orphan’ patients –those with the diagnosis of interest (AF in this case) but not currently on treatment or patients on treatment, but without a diagnosis on record and 2) ‘Maintenance’ patients –those already on treatment. Orphan patients fall into one of 4 categories (‘fell through the cracks’, ‘refused treatment’, ‘harms outweigh benefits’ and ‘false positive’) that are not distinguishable using only clinical data. Maintenance patients are easy to categorize, but those on HOLD cannot be easily distinguished using clinical data. To solve these issues of poor categorization, we designed a Patient Initialization Dashboard that creates a registry of patients that are most likely to be eligible for the CDSS. Physicians can categorize their patients according to their anticoagulation status and ensure that the CDSS provides the appropriate recommendations for each patient. The patient registry has the secondary benefit of allowing the physician to get a list of all of their AF patients and ensure that no one is falling through the cracks. 2.2. Determining Ongoing Eligibility for CDSS Patients who are newly diagnosed after initialization of the CDSS need to be detected in a different way. This requires a more sophisticated integration with the EMR. The EMR needs to pass information to the CDSS when new information is added to the chart. For example, upon entry of a new Problem List Diagnosis or a new Encounter Diagnosis, the CDSS needs to be informed of the new diagnosis and needs to make a determination of eligibility. To solve the problem of on-going eligibility checking, the EMR needs to provide the CDSS with a minimal dataset of information when a diagnosis is added to the problems list (OnNewProblemList) and when a diagnosis is added to an encounter (OnNewEncounterDiagnosis). The CDSS can then determine whether the patient is eligible for CDSS recommendations. 2.3. Managing State Transitions Once a patient’s state is determined, either through direct physician categorization in the Patient Initialization Dashboard or through the eligibility determination during the visit, the CDSS needs to be able to persist the information about the patient’s current state. This information may not be able to be stored by the CDSS for privacy or efficiency reasons. We thus install a ‘cookie’ on each EMR with the relevant information about the patient and require the EMR to provide it back to the CDSS

within the same package of clinical information or minimal dataset that it sends to the CDSS when the CDSS is called. State transition diagrams place patients into explicit AC states: 1) PREINITIATION, 2) MAINTENANCE, 3) HOLD and 4) DISCONTINUE. Switching is handled as a DISCONTINUE and PRE-INITIATION sequence. This same system can be used to safely put a patient’s AC on HOLD for a short period of time (e.g., for surgery or other invasive procedure). In this case, the CDSS would replace the HOLD cookie with a MAINTENANCE cookie, once the HOLD expired. This also solves an important clinical problem where patients who are put on HOLD fall through the cracks and are not re-initiated after their procedure. 2.4. Making vague terms explicit The AC guidelines do have several vague terms. The most difficult to operationalize was one called ‘stable’ INR (International Normalized Ratio –a measure of the level of AC). When does an INR result become stable? After 2 or 3 or 4 results that are within the appropriate range? The answer is, ‘it depends’. Even if INR has been stable, if a patient gets sick or is hospitalized, then the INR is not considered to be stable and needs to be monitored more frequently. This problem was solved by creating an operational definition that included additional pieces of information that could be obtained through the PHR or a patient interactive system. 2.5. The business analysis artifacts The business requirements artifacts were completed after several iterations and are being made available to EMR vendors to incorporate into their products. We have created an EMR and PHR implementation guide that explains to EMR vendors how to integrate the CDSS into their EMR and tethered PHR applications. Lastly, we have developed a web-based tool for testing and demonstration purposes. The artifacts are EMR and PHR agnostic and will allow any EMR vendor to integrate their EMR and PHR products with the CDSS engine.

3. Discussion Anticoagulation in CDSS systems is already successfully being used as standalone systems in specialized clinics. There is a need for them in general practice, but they need to be integrated into EMRs and PHRs. We have developed an architecture that describes how to integrate an AC CDSS into multiple EMRs and PHRS. The architecture and design are intended to be generic and work for multiple treatment guidelines. However, the specifications do need further prototyping and testing. This integrated CDSS will undergo clinical trials in the near future. A demonstration system that can be integrated into EMRs and PHRs has been developed and is awaiting integrations and further testing. Acknowledgements: Funding for this project was obtained through an eHealth Catalyst grant from CIHR.

4. References [1] “Statistics - Heart and Stroke Foundation of Canada” [Online]. Available: http://www.heartandstroke.com/site/c.ikIQLcMWJtE/b.3483991/k.34A8/Statistics.htm. [Accessed 05 July 2014]. [2] J. L. Blackshear, "Appendage Obliteration to Reduce Stroke in Cardiac Surgical Patients With Atrial Fibrillation," The Annals of Thoracic Surgery, vol. 61, pp. 755-759, 1996. [3] A. Wittkowsky, "Effective anticoagulation therapy: defining the gap between clinical studies and clinical practice," The American Journal of Managed Care, vol. 10, pp. S297-306, 2004. [4] R. Neiuwlaat, S. B. Olsson, G. Y. Lip, A. J. Camm, G. Breithardt, A. Capucci, J. G. Meeder, M. H. Prins, S. Levy and H. J. Crijns, "Guideline-adherent antithrombotic treatment is associated with improved outcomes compared with undertreatment in high-risk patients with atrial fibrillation. The Euro Heart Survey on Atrial Fibrillation," American Heart Journal, vol. 153, no. 6, pp. 1006-1012, 2007. [5] J. Robson, I. Dostal, R. Mathur, R. Sohanpal, S. Hull, S. Antonious, P. MacCallum, R. Schilling, L. Ayerbe and K. Boomia, "Improving anticoagulation in atrial fibrillation: observational study in three primary care trusts," The British journal of general practice: the journal of the Royal College of General Practitioners, vol. 64, pp. e275-281, 2014. [6] J. Casciano, D. Singer, W. Kwong, E. Fox and B. Martin, "Anticoagulation therapy for patients with nonvalvular atrial fibrillation: comparison of decision analytic model recommendations and real-world warfarin prescription use," American journal of cardiovascular drugs: drugs, devices, and other interventions, vol. 12, no. 5, pp. 313-323, 2012. [7] B. H. Rocha, L. H. Langford and S. Towner, "Computerized management of chronic anticoagulation: three years of experience," Studies in health technology and informatics, vol. 129(Pt 2), pp. 1032-6, 2007. [8] J. Cox, "Integrated Management Program Advancing Community Treatment of Atrial Fibrillation (IMPACT-AF)," 20 August 2013. [Online]. Available: http://impact-af.ca/. [Accessed 14 July 2014]. [9] “DAWN AC Anticoagulation Software - 4S Dawn Clinical Software.” 4S Information Systems Ltd,, [Online]. Available: http://www.4s-dawn.com/products/anticoagulation/dawnac/. [Accessed 18 August 2014]. [10] "International Institute of Business Analysis," [Online]. Available: http://www.iiba.org/. [Accessed 25 August 2014].