Integrative Midlife Profiling to Anticipate Stroke and Silent Cerebrovascular Injury
Description: Accurately predicting stroke risk remains a significant challenge. Existing models (e.g. CHADS2VASC, PREVENT) rely heavily on vascular risk factors (VRF), and have only moderate predictive power. They perform poorly in younger populations where lifestyle changes could be most impactful, and they provide little value in identifying those deemed traditionally “lower-risk”. Moreover, they offer no insight into clinically meaningful patient subtypes (e.g. ischemic vs hemorrhagic stroke risk profiles) and do not consider imaging measures of cerebrovascular injury, despite the known strong linkage to clinical stroke and dementia. Therefore, a better approach is needed to improve early identification of clinical and subclinical brain injury. Using biclustering methods, we will create and describe clusters of individuals in midlife that predict stroke with the ultimate goal of identifying personalized prevention strategies based on these identified riskgroups. Clustering is a powerful tool that allows for nonlinear, heterogenous relationships to be uncovered. We will base our study in large, population-based cohort studies [primary cohort: Atherosclerosis Risk in Communities (ARIC) Study, replication cohort: Multi-ethnic Study of Atherosclerosis (MESA)]. Both ARIC and MESA have ongoing physician adjudication of stroke events. Using unsupervised machine learning, we will cluster participants without prevalent stroke in midlife based on their VRFs, social factors and proteomics (~5,000 individual proteins already collected). Proteins serve as biologic mediators in the body that are closer to disease than the upstream clinical risk factors. We believe these proteins can help to improve stroke prediction over traditional risk factors alone. We will determine which clusters predict incident stroke and compare the clusters’ ability to predict stroke when including, versus not including the proteomic data (Aim 1). We will also determine which clusters predict subclinical cerebrovascular disease measured on brain magnetic resonance imaging (Aim 2). Finally, we will then compare the clusters’ ability to distinguish between distinct stroke characteristics, such as stroke type and age at stroke (Aim 3). This project aims to change the landscape of stroke prediction by using clustering methods to analyze multidimensional data, including proteomics, to unveil previously overlooked stroke risk markers, all in an effort to start effective prevention strategies even earlier.
Funder: American Heart Association
Amount: $99,990
Funding type: research
Funder scheme: Rapid Impact Research Award
Start date: 2026-04-01
End date: 2028-03-31
Investigators: Michelle Johansen, Michelle Johansen
Institution: Johns Hopkins University School of Medicine
Funded outputs count: 0
Funder award ID: 26RIRA1642481
OpenAlex ID: G6289222663