e-HAIL Event

Building and Deploying Interpretable AI Models to Predict Chronic Disease Exacerbations in VA

Shirley Cohen-Mekelburg, M.D., MSAssistant Professor of Internal MedicineMedical SchoolAkbar Waljee, M.D., MScProfessor of Internal MedicineMedical School

Chronic disease exacerbations affect over 50% of patients with conditions such as chronic obstructive pulmonary disease, rheumatoid arthritis, and inflammatory bowel disease. Early intensive treatment is key to preventing these exacerbations and disease progression to improve health outcomes. However, there is a tradeoff between the benefits and potential harms of intensive treatment, so this strategy is not appropriate for all patients. There is a critical need for developing high-performing clinical prediction models to identify Veterans with these chronic conditions who are at risk of disease exacerbation when they would benefit most from intensive treatment. However, we have identified critical barriers to successfully deploying prediction models in clinical practice:

  • Existing models lack transparent reasoning behind predictions leading to skepticism from users.
  • Risk models are often developed from office visit data rather than patient-generated health data (PGHD), but Veterans may be seen in the clinic only 1-2 times per year.
  • The lack of information on user acceptance and the value of models in improving outcomes and reducing costs limits our understanding of the logistics of model deployment.

We propose to address these gaps to develop an interpretable ML/AI model for predicting exacerbations in Veterans with IBD and lay the groundwork for incorporating interpretable risk models to support early intensive treatment of chronic diseases to improve Veterans’ health.

Zoom information will be shared with e-HAIL members.


J. Henrike Florusbosch