A Comparison of Machine Learning Models for ICH Prognostication: An Analysis of ATACH-2 and Qatar Stroke Database.

INTRODUCTION: Multiple prognostic scores have been developed to predict morbidity and mortality in patients with spontaneous intracerebral hemorrhage (sICH). These scoring models were traditionally based on statistical methods involving a limited set of variables. The advent of machine learning (ML) has enabled the development of several prognostic models for sICH that can leverage much more data. METHODS: We trained ML models on two distinct datasets: (1) Qatar dataset only and (2) a combined dataset consisting of the Qatar and Antihypertensive Treatment of Acute Cerebral Hemorrhage II (ATACH-2) datasets. Model validation was conducted separately on the Qatar and ATACH test sets, providing insights into model performance within and across study populations. By incorporating inpatient variables into model development, we leveraged more information. We also compared models derived from admission-only variables with models derived from both admission and inpatient variables. RESULTS: For 90-day mortality using combined training data, XGBoost (XGB) achieved the highest area under the curve (AUC) on the Qatar test set, while Random Forest achieved an AUC of 0.916 on the ATACH test set. For 90-day functional outcomes, Random Forest and XGB achieved AUCs of 0.882, respectively. Models trained using both admission and inpatient data outperformed admission-only models. Feature importance revealed important markers of prognostication such as hematoma expansion and status of intubation. Sensitivity analyses confirmed that results were robust to assumptions regarding follow-up imaging availability. CONCLUSION: Our study design mirrors a real-world deployment scenario in which a model developed at a single center is transported to external cohorts, while still preventing any information leakage from test data.