Assessing Population-Based Radiological Brain Health in Stroke Epidemiology: Rationale and Design

Approximately 20% of strokes in the USA are preceded by either a stroke or transient ischemic attack (TIA). Determining which stroke patients are at higher risk for recurrence allows for individualized, aggressive secondary stroke prevention. A comprehensive clinical decision tool, considering the full spectrum of radiological brain health including small vessel disease parameters, is currently lacking. Furthermore, large-scale characterization of preexisting radiological brain health may elucidate novel phenotypes. This study aims (1) to characterize imaging manifestations of brain health at a population level, and associated demographic and clinical risk factors at the time of index stroke and (2) to create a 90-day and 3-year prediction models of cerebrovascular disease recurrence (ischemic or hemorrhagic stroke) incorporating comprehensive parameters from routine clinical imaging.

Our overall cohort was estimated to consist of 4,250 patients hospitalized with stroke, including 525 with hemorrhagic and 3,725 with ischemic/TIA subtypes, ascertained in the Greater Cincinnati/Northern Kentucky Stroke Study (GCNKSS) population of 1.4 million residents from January 1, 2015, to December 31, 2015. Among 3,725 ischemic stroke/TIA patients, based on published and ongoing data collection, we estimated that approximately 16% will have a recurrent ischemic or hemorrhagic stroke over the subsequent 3 years. Among these, 80% were estimated to have MR imaging (MRI) for review. Leveraging extensive clinical and demographic data already collected in the 2015 NIH-funded GCKNSS study, we will have obtained and centrally characterized MRI, acute CT, and vascular data in patients with hospitalized stroke/TIAs. We will determine if and how preexisting imaging parameters cluster using factor analysis, and identify associated demographic and clinical risk factors in multivariable modeling. We will develop short-term (90-day) and long-term (3-year) risk prediction models using the machine learning approach of random survival forest with internal validation, and perform Cox regression models as a sensitivity analysis.

The primary outcome is recurrence defined as any stroke (ischemic or hemorrhagic) occurring after index ischemic stroke or TIA event. For index ischemic strokes, the second event must within a different vascular territory if <14 days from the index event.