Abstract 14875: Topological Data Analysis of Electronic Health Record Features Predicts Major Cardiovascular Outcomes After Revascularization for Acute Myocardial Infarction

Abstract

Introduction: Topological data analysis (TDA) is an emerging mathematical technique that imposes structure on a dataset and examines the shapes (or topological features) that arise. The TDA algorithm Mapper Plus is proposed to study the spatial characteristics of clinical data for the purpose of patient classification. This is a data-driven and hypothesis-free approach to clinical risk stratification. Hypothesis: The hypothesis is that Mapper Plus can classify distinctive subsets of patients receiving optimal treatments post-acute myocardial infarction (AMI). Our goal is to identify subgroups who remained at risk of having a future adverse event (AE) i.e., death, heart failure hospitalization, or recurrent MI Methods: A single-center, retrospective analysis of 31 clinical variables from the electronic health record (EHR) was conducted on 798 AMI subjects. Risk was defined as high or low in relation to the average probability of survival without AE for the entire cohort at the 1 year mark. Results: TDA identified six subsets of patients. Four subsets (n=597) had > 1-fold change on the probability of survival free of AEs; these became the low-risk subgroup. Two subsets (n=344) had < 1-fold change on the probability of survival free of AEs; these became the high-risk subgroup. Because Mapper Plus allows for subject overlap across subgroups, 143 subject (18%) were shared between the high (n=201, 25%)- and low-risk (n=454, 57%) subgroups and were extracted into a 3rd intermediate risk subgroup (see figure). Conclusions: TDA significantly stratified AMI patients into 3 subgroups with distinctive incidences of AE up to 3 years post AMI. This is a new framework for EHR-based risk stratification that requires no additional patient interaction, is agnostic to prior knowledge, and is driven by the topological features of the included data. Further studies will be needed to validate TDA across different patient cohorts before it can be applied to clinical practice.