Our overall research mission is to improve the accuracy of medical diagnosis in frontline healthcare settings, particularly the emergency department (ED). The long-range goal of this research program is to develop and validate a computer-based, diagnostic decision support system for the ED that reduces critical misdiagnosis (stroke, myocardial infarction, etc.) of patients with undifferentiated medical symptoms (dizziness, chest pain, etc.). To succeed, the system must fit within the ED workflow. This issue has been largely ignored in prior attempts to develop similar systems (which rely on physician data entry), resulting in physician non-use. We propose a workflow-sensitive approach, using a patient-centered system for Automated Medical Interviewing for Diagnostic Decision Support (AMIDDS) capable of symptom-specific, adaptive diagnostic history-taking in a pre-encounter setting such as the ED waiting area. The interview would be self-administered (e.g., using a touch-screen kiosk) and would provide physicians a diagnostic summary with probabilistic assessments of the likelihood of common and critical diagnoses, with suggestions for possible bedside or laboratory test strategies. This would permit physicians to practice medicine in the way they are most comfortable, without additional time on their part, but with evidence-based suggestions to help avoid critical misdiagnosis. There would also be an interview transcript that could be used as documentation, similar to a paper pre-encounter history form. Building such a system will require a sequential, modular, symptom-specific approach to diagnosis (e.g., """"""""dizzy"""""""" module, """"""""chest pain"""""""" module, etc.), with modules rigorously developed, tested and validated against clinical outcomes. We have chosen dizziness as our first symptom module because (1) dizziness is common (>2.5 million ED visits per year US), (2) we have a great deal of experience working with this symptom; and (3) there is considerable room for improvement (e.g., dizziness is the symptom most often associated with missed stroke in the ED). The first step in developing this novel approach to decision support is to establish a firm scientific basis for automated history-taking. The goal of this application is to develop and validate a """"""""dizzy"""""""" module prototype capable of accurately, reliably, and efficiently obtaining a history of patients' dizziness symptoms. Over the past 5 years, we have extensively field-tested a large databank of interviewing questions and have confirmed that patients are capable of self administering such automated interviews.
Our Specific Aims for this 1-year project are to validate (relative to structured human interview) that an automated interview can (1) discern which ED waiting room patients are dizzy; and (2) elicit key aspects of their dizziness symptom history. Completion of these Aims will enable creation of a decision support system to improve dizziness diagnosis in the ED. Future studies will establish the clinical and cost-effectiveness of such decision support in reducing critical misdiagnosis. These projects will serve as a logical prelude to extending this patient-centered, workflow-enhancing approach for decision support to multiple symptoms in multiple healthcare contexts. ? ? ?
(Relevance) Critical misdiagnosis is common, resulting in the preventable deaths of 40-80,000 annually in the US, and probably substantial unmeasured morbidity (e.g., from missed opportunities to prevent major, disabling stroke after missed transient ischemic attack or minor stroke). Initiatives in patient safety have principally focused on improving treatment, not diagnosis. This work will form the cornerstone of a symptom-by-symptom and setting-by-setting translational research effort to bring science to the art of bedside diagnosis, in pursuit of a long-term solution to the major public health problem of misdiagnosis.