Covert Cerebrovascular Disease Detected by Artificial Intelligence (C2D2AI): A Platform for Pragmatic Evidence Generation for Stroke and Dementia Prevention

Tufts Medical Center, Boston MA

Investigators

Linked publications, trials & patents

Abstract

Project Summary It is a common clinical occurrence that neuroimaging scans obtained in the course of routine clinical care discover covert cerebrovascular disease (CCD), comprising covert brain infarction (CBI) and white matter disease (WMD), in patients with no history of stroke or transient ischemic attack. Indeed, epidemiologic studies indicate that covert CBI are far more common than clinically-evident strokes and these imaging findings are strong, independent risk factors for future stroke and dementia. However, there are no proven preventive treatments or guidelines for initiating risk factor-modifying therapy. While there is strong evidence that antiplatelet therapy and statin therapy are effective in preventing recurrent stroke in patients with prior stroke, it is unclear the degree to which these results apply to patients with CCD. Additionally, patients and providers are rarely aware of these findings, even when they are detected. As part of our previous grant (R01-NS102233), we developed a natural language processing (NLP) algorithm to identify incidentally discovered (id-) CCD from neuroimaging reports, which we ported into a large integrated healthcare system. We identified a cohort of almost a quarter million patients over age 50 who received either a head CT or MRI and were stroke- and dementia- free at the time of the index scan. Key findings of our analyses include: NLP can identify id-CBI and id-WBD from neuroimage reports as well as human readers; that id-CCD is present in about one-third of these scans in an age- and vascular risk factor dependent manner; that id-CCD increases the risk of future stroke and future dementia by approximately 2- to 3-fold; that NLP is able to extract additional important prognostic information on WMD severity from routinely obtained imaging reports; and finally, these patients are generally not given risk factor modifying treatment following the discovery of id-CCD. Given the difficulty of recruiting this at risk population, we now propose to leverage this NLP system as a platform to plan and conduct prospective randomized comparative effectiveness studies to identify optimal treatment strategies for id-CCD. Thus, our aims are: Aim 1: To inform the enrollment criteria of prevention clinical trials and ensure consistency of findings, we will expand the cohort to Kaiser Permanente Northern California and further characterize patients with id-CCD regarding their future stroke and dementia risk. Aim 2: To determine optimal treatment algorithms, we will leverage established simulation models to estimate the treatment effects of different risk factor modification algorithms in patients with id-CCD on future stroke and dementia. Aim 3: To determine optimal recruitment strategies in demographically diverse populations, we will examine the feasibility of recruiting this novel population based on NLP-identified findings both prospectively (i.e. concurrent with clinical identification) and retrospectively (as identified from pre-existing scans). Aim 4: Based on the above findings we will plan a multicenter clinical trial for the prevention of stroke and dementia in this population with CCD.