Using clinical and thrombus characteristics to predict the etiology of ischemic stroke: An analysis of the INSIGHT registry.

BACKGROUND AND PURPOSE: Ischemic stroke comprises about 87% of all stroke cases in the US. 20% of these have a cardioembolic (CE) etiology, and 25% are still considered embolic strokes of undetermined sources (ESUS). ESUS strokes are associated with worse functional outcomes and survival rates, underscoring the need for understanding the clinical differences in stroke etiology.

METHODS

We analyzed the Investigation of Clot in Ischemic Stroke and Hematoma Evacuation (INSIGHT) registry (NCT04693767, https://clinicaltrials.gov/study/NCT04693767), which includes n=388 patients total (n=208 CE, n=142 AFIB, n=180 non-CE). We analyzed 123 data features, spanning across domains including demographics, past medical history, baseline blood lab values, intraoperative data, and thrombus characteristics. We found correlations, univariate analyses, and LASSO machine learning models to predict etiology.

RESULTS

We found several associations between clinical variables and stroke etiology. Higher baseline prothrombin time and INR (only among patients not on anticoagulants), baseline use of antiarrhythmics and antihypertensives, and a higher baseline mRS were associated with an increased risk of both overall CE and AFIB-specific etiologies. A higher baseline platelet and RBC count was associated with a decreased risk of AFIB stroke etiology. All of these results were significant (p<0.05) across multiple analysis methods. Lastly, our LASSO models predicted CE and AFIB-specific etiologies with an AUC of 0.71 and 0.73, respectively.

CONCLUSIONS

Several clinical features are strongly correlated with stroke etiology, and these can be combined in a machine learning model to predict stroke etiology.