Double Machine Learning Analysis of Treatment Heterogeneity in the EARLYDRAIN Trial: Precision Insights into Lumbar CSF Drainage Efficacy

Introduction: Subarachnoid hemorrhage (SAH) is a severe sequelae of aneurysms leading to poor outcomes. The EARLYDRAIN trial illustrated that prophylactic lumbar cerebrospinal fluid (CSF) drainage can improve long-term neurological outcomes in patients with aneurysmal SAH, although not uniformly for all patients. Our study builds upon the EARLYDRAIN findings by examining how aneurysm-related features impact the effectiveness of LD on patient outcomes through causal inference analysis.

Methods: Using published data from the EARLYDRAIN trial, we investigated treatment heterogeneity in LD efficacy across the aneurysm treatment method (clipping vs. coiling), aneurysm circulation, and hemorrhage presence. We applied Double Machine Learning (DML) with Random Forest models to estimate the (Conditional) Average Treatment Effect (ATE and CATE) of LD on the 6-month modified Rankin Scale (mRS) score, focusing on subgroups defined by aneurysm characteristics.

Results: The effectiveness of LD varied by aneurysm features. In general, consistent with the EARLYDRAIN trial, LD had an overall ATE of -0.35 on mRS, suggesting improved neurological outcomes. Specifically, patients with anterior circulation aneurysms and no intraventricular hemorrhage benefited the most from LD (CATE = -0.76). However, patients with posterior circulation aneurysms and concurrent intraparenchymal hemorrhage notably had potentially negative effects from LD (CATE = +0.1). There was a statistically significant difference in LD effectiveness between coiling and clipping treatments (p = 0.042).

Conclusion : This extended analysis supports the efficacy of LD in improving long-term neurological outcomes post-aneurysmal SAH, yet highlights the benefits are causally context-dependent, with specific aneurysm characteristics influencing outcomes. LD appears most beneficial for patients with anterior circulation aneurysms and specific hemorrhage profiles. Future studies should incorporate additional morphological features of aneurysms to assess when LD may be most effective.