320 A Common HFE Mutation is Associated With Decreased ICH Severity: A Machine Learning Approach Using Electronic Health Records

INTRODUCTION: Intracerebral hemorrhage (ICH) is a devastating disease with a 30-day mortality rate greater than 40%. Following hemorrhage, iron released from blood-breakdown leads to an iron-dependent form of cell-death known as ferroptosis. Animal studies on genetic variations in iron metabolism demonstrate that the H67D HFE mutation reduces ICH-associated ferroptosis. However, the impact of the commonly found human homolog, the H63D HFE mutation, has not been studied in stroke patients. METHODS: Four different machine learning algorithms were developed to predict dichotomized discharge mRS using EHR ICD-10 codes from 6,500 stroke patients at the Penn State’s Comprehensive Stroke Center. Performance was assessed using Precision-Recall curve (PR-AUC) and Receiver-Operating-Characteristic curve (ROC-AUC) in a 25% held-out test set. The best-performing model was applied to predict dichotomized mRS of ICH stroke patients in UK-Biobank (UKBB) and AllofUs datasets. Finally, a logistic regression model assessed the association between the H63D mutation and the predicted dichotomized mRS. The association results from UKBB and AllofUs were combined using a fixed-effect meta-analysis. RESULTS: Although all models achieved comparable performance, the gradient boosting machine model consistently achieved the highest ROC-AUC (86.66) and PR-AUC (89.51). When applied to UKBB and AllofUs datasets, the H63D mutation was significantly associated with lower odds of predicted dichotomized mRS =3 at discharge (OR=0.70, two-tailed p-value=0.044). CONCLUSIONS: Our study presents evidence that the H63D HFE mutation may decrease ICH severity. Further investigation through a detailed cohort-study is warranted to confirm this observation and to further re-evaluate risk stratification and personalized treatment approaches for ICH patients dependent on their HFE genotype.