Abstract 153: Novel Machine Learning Proves Stroke Risk is Not Linear

Abstract

Background: Current stroke risk assessment tools presume the impact of risk factors is linear and cumulative. However, both novel risk factors and their interplay influencing stroke incidence are difficult to reveal using traditional linear models. Objective: To improve upon the Revised-Framingham Stroke Risk Score and design an interactive non-linear Stroke Risk Score (NSRS). Our work aimed at increasing the accuracy of event prediction and uncovering new relationships in an interpretable user-friendly fashion. Methods: A two phase approach was used to develop our stroke risk score predictor. First, clinical examinations of the Framingham offspring cohort were utilized as the training dataset for the predictive model consisting of 14,196 samples where each clinical examination was considered an independent observation. Optimal Classification Trees (OCT) were used to train a model to predict 10-year stroke risk. Second, this model was validated with 17,527 observations from the Boston Medical Center. The NSRS was developed into an online user friendly application in the form of a questionnaire (http://www.mit.edu/~agniorf/files/questionnaire_Cohort2.html). Results: The algorithm suggests a key dichotomy between patients with or without history of cardiovascular disease. While the model agrees with known findings, it also identified 23 unique stroke risk profiles and introduced new non-linear relationships; such as the role of T-wave abnormality on electrocardiography and hematocrit levels in a patient’s risk profile. Our results in both the training and validation populations suggested that the non-linear approach significantly improves upon the existing revised Framingham stroke risk calculator in the c-statistic (training 87.43% (CI 0.85-0.90) vs. 73.74% (CI 0.70-0.76); validation 75.29% (CI 0.74-0.76) vs 65.93% (CI 0.64-0.67), even in multi-ethnicity populations. Conclusions: We constructed a highly predictive, interpretable and user-friendly stroke risk calculator using novel machine-learning uncovering new risk factors, interactions and unique profiles. The clinical implications include prioritization of risk factor modification and personalized care improving targeted intervention for stroke prevention.