The association of a common HFE gene variant with stroke disability utilizing predictive machine learning and electronic health records.

Animal studies demonstrate that the H67D HFE mutation is neuroprotective against intracerebral hemorrhage (ICH), but whether the human analog, H63D HFE, confers a similar effect is unknown. We developed a machine learning algorithm to predict discharge Modified Rankin Score (mRS ≤ 2 functionally independent vs. ≥ 3 moderate-severe disability) from electronic health records (EHR) of stroke patients. This model was applied to large genetic databases to determine the association between the H63D mutation and predicted stroke disability. Four algorithms were trained on ICD-10 codes from 6,500 stroke patient records to predict dichotomized discharge mRS. The best model was applied to UK Biobank and AllofUs stroke cohorts, and the association of predicted mRS with H63D mutation was assessed by logistic regression. Although all models performed similarly, the gradient boosting model achieved the highest performance (AUROC 86.66; AUPRC 89.51). ICH patients without the H63D mutation had significantly higher odds of predicted disability (mRS ≥ 3) compared to carriers (OR 1.42, p = 0.044). No association was observed in other stroke types. Our study presents evidence that the commonly found H63D HFE mutation is associated with decreased ICH disability. These findings highlight a potential genetic mechanism that could guide future therapeutic strategies for ICH.