Data‐Driven Analysis Reveals Cortical Infarction Patterns Correlated With Inflammation and Prognosis: A Retrospective, Multicenter Cohort Study

Abstract

Background We aim to identify the distinct lesion patterns and regions associated with functional outcome and inflammation in patients with acute ischemic stroke, and investigate whether the association between lesion patterns and functional outcome was mediated by inflammation. Methods and Results We performed nonnegative matrix factorization to derived low‐dimensional lesion patterns (atoms), and Bayesian linear regression models were applied to explore the associations of lesion patterns with inflammatory factors including high‐sensitivity C‐reactive protein and interleukin‐6, as well as functional outcome (defined as modified Rankin Scale score at 3 months). The difference distribution mean and 95% highest probability density interval (HPDI) were calculated. Mediation analysis was used to examine the mediating effects of inflammation on the relationships between lesion patterns and functional outcome. Seven lesion patterns were derived from 5914 patients with acute ischemic stroke. Lesion patterns distributed in the cortical regions were associated with inflammatory response, including atom 1 (interleukin‐6: mean, 0.113 [95% HPDI, 0.073–0.162]; high‐sensitivity C‐reactive protein: mean, 0.082 [95% HPDI, 0.038–0.123]) and atom 4 (interleukin‐6: mean, 0.113 [95% HPDI, 0.071–0.167]; high‐sensitivity C‐reactive protein: mean, 0.108 [95% HPDI, 0.058–0.165]). These lesion patterns were also significantly associated with functional outcome (atom 1: mean, 1.958 [95% HPDI, 1.538–2.383]; atom 4: mean, 2.245 [95% HPDI, 1.773–2.741]). Mediation analysis suggested that interleukin‐6 explained 15.34% and 7.47% in the association of atom 1 and atom 4 with functional outcome, respectively. Conclusions Certain lesion patterns that are associated with both inflammation and functional outcome of acute ischemic stroke, especially cortical infarction, may play a role in functional outcome through modulating inflammatory reactions.

Keywords

• functional outcome
• inflammatory response
• ischemic stroke
• lesion patterns
• machine learning