Abstract MP12: Drivers of Rescue Therapy Efficacy for Vasospasm Following Aneurysmal Subarachnoid Hemorrhage: A Propensity-Score Matched Analysis With Machine Learning
Introduction: One of the foremost challenges in subarachnoid hemorrhage (SAH) management is understanding which patient characteristics and treatment decisions lead to good outcomes. This study uses novel game theory-based methods in explainable machine learning (ML) and propensity-score matching to elucidate the clinical factors driving outcomes following rescue therapy for post-SAH vasospasm. Methods: Data for patients with post-SAH angiographic or symptomatic vasospasm were obtained from six clinical trials and observational studies in the Subarachnoid Hemorrhage International Trialists (SAHIT) repository. Gradient boosting ML models were constructed for each patient to predict the probability of receiving rescue therapy and 3-month Glasgow Outcome Scale (GOS) scores. Shapley Additive Explanation (SHAP) values were calculated to quantify feature importance and interaction effects. Variables with high SHAP importance in predicting rescue therapy were used in a propensity score-matched analysis of rescue therapy and 3-month GOS scores. Results: A total of 1,532 patients were included. SAH characteristics and neurological sequelae, but not admission neurological scores, heavily influenced the probability of receiving rescue therapy. Comparing feature importances showed cerebral ischemia/infarction was invariably linked to poor outcome, while other important predictors of outcome varied by rescue type. Higher blood pressures and fewer postoperative days until vasospasm treatment were more important for predicting worse outcome following interventional rescue, while high admission WFNS grade and pneumonia were important predictors of worse outcome for non-interventional rescue. Finally, in a propensity score-matched analysis guided by SHAP-based variable selection, rescue therapy was associated with higher odds of 3-month GOS of 4-5 (odds ratio: 1.63; 95%CI 1.22-2.17; p=0.001). Conclusion: Rescue therapy for vasospasm was associated with good functional outcomes. Future randomized trials focusing on preventative or therapeutic interventions may be able to demonstrate improvements in clinical outcomes. Insights from these models may help improve patient selection criteria and trial designs for rescue therapy.