Reimagining Medical Diagnosis with Bayesian Analysis
Dany Accilien · Emergency Medicine News · 2024
Figure: Bayesian analysis, EPs, chest pain, coronary artery disease, ECG, acute ischemia, HEART score, cardiology, TIMI, statistical analysis, catheterization, acute occlusive diseaseFigureA 74-year-old woman presented from her cardiologist's office with chest pain. She had a history of coronary artery disease and had had a stent placed in one of her coronary arteries years earlier. She mentioned she had been having intermittent, sharp, left-sided, nonradiating pressure in her chest for a week, along with intermittent shortness of breath when walking moderate distances. The patient reported no additional symptoms. Her two ECGs, two high-sensitivity troponin results, and chest radiograph showed no signs of acute ischemia or abnormalities. Considering her extensive history and that she had a HEART score of 5, cardiology asked to admit her for a catheterization the next morning. She refused and wanted to go home instead. Was she making the right disposition decision? Medical training continues to be guided by fundamental Hippocratic and Oslerian principles: Perform a thorough history and physical examination, gathering a unique collection of signs and symptoms. This information is then intended to be organized and analyzed to reach a likely diagnosis and guide appropriate treatment. We have refined our history-taking and physical examination skills over the years. We added labs and imaging to our diagnostic toolkit. We developed evidence-based guidelines and risk-stratification tools such as the TIMI and HEART scores in hopes of enhancing objectivity in making a patient diagnosis. Physicians use deductive reasoning to draw diagnostic conclusions every day, but deduction requires knowledge of all the required steps in a logical form. What if we are not synthesizing all the available data correctly? Prior Probabilities Probability is an extension of logical reasoning. It can allow us to assign a quantitative range to the plausibility of a proposition given certain background information. You can update your probability based on observations. This is not novel; in fact, it's the purest form of the art of medicine. We want the probability of a hypothesis we have generated, not the probability of the data we have been given. Experience, observation, and clinician gestalt all matter. Yet it's difficult to know how to integrate these factors into our decision-making or how to update these beliefs to make diagnoses appropriately. What if we could quantify and incorporate these essential aspects of diagnosis that we use every day? This would help physicians reach highly accurate disease diagnoses in a standardized, reproducible, and understandable manner. Bayesian statistical analysis allows us to reimagine how we view medical diagnosis. Using simple data from patient history and examination, we can assign prior probabilities and integrate them into our diagnostic approach. Through Bayesian analysis, we can not only determine the likelihood of a major adverse cardiac event occurring within the next six weeks but also understand how factors like race, ZIP code, and the presence (or absence) of specific signs and symptoms may influence this probability. Bayesian analysis incorporates prior knowledge and updates it with new evidence to create a posterior probability. In simple terms, it combines what we already know with what we learn from a patient's history and physical examination. The method allows us to adjust our initial assumptions (priors) based on the evidence provided by the patient. This helps us generate more accurate diagnoses and treatment strategies by taking into account the uniqueness of each patient's situation. Tailoring Treatment More than a month after her initial visit, my patient returned to her cardiology office and subsequently completed the catheterization procedure. It showed no acute or worsening occlusions. Looking back, using Bayesian analysis of her presentation at the time, it was telling me to send her home and that the likelihood of an acute occlusive disease requiring intervention was extremely low. Her decision in the ED was not without risk, but she was right. Osler is famously credited with the quote, “Listen to your patient; he is telling you the diagnosis,” and it remains relevant today. We have always been listening, but we now have the opportunity to see the diagnosis more clearly with the help of Bayesian analysis. We can improve the precision of our diagnoses by quantifying prior beliefs and integrating them with our current diagnostic tools. This is an essential step in tailoring treatment strategies to individual patients' needs. Bayesian analysis has the potential to revolutionize the way we approach and teach medical diagnostics. It provides a more comprehensive and personalized method of understanding patients' symptoms and conditions. Recent advances in artificial intelligence will cause a paradigm shift in how we approach medical diagnosis. As we continue to refine our skills and integrate advanced AI and statistical methods into our practice, we move closer to realizing Osler's vision of not only truly listening to our patients but providing the best diagnosis. DR. ACCILIEN is an emergency physician at Wellstar Kennestone in Marietta, GA. He currently serves as the chief medical officer and medical director of RelyMD, a telehealth platform. He has interests in evidence-based medicine, health care technology, artificial intelligence, and process improvement. Share this article on X and Facebook. Access the links in EMN by reading this on our website: www.EM-News.com. Comments? Write to us at [email protected].