Development of treatment-specific outcome prediction models for intracranial aneurysm after coiling or clipping

Treatment-specific outcome prediction after intracranial aneurysm repair remains limited in resource-constrained settings, where complex case and access to endovascular care may differ among centers. We aimed to develop preliminary treatment-specific prediction models for in-hospital outcomes after endovascular coiling or microsurgical clipping in an comprehensive stroke center.

Consecutive patients treated with coiling or clipping were analyzed. In-hospital endpoints were mortality, good functional outcome defined as Glasgow Outcome Scale-Extended (GOSE) 5-8, and recovery of Glasgow Coma Scale (GCS) to premorbid baseline. Separate multivariable logistic regression models were developed for each treatment-outcome pair after multiple imputation, bidirectional Akaike information criterion selection, and multicollinearity assessment. Discrimination was quantified using area under the curve (AUC), and bootstrap internal validation (500 resamples) was used to estimate optimism-corrected performance.

The retrospective cohort (n = 436; 224 coiling, 212 clipping) was predominantly ruptured (86.9 %). In-hospital mortality was 11.2 % after coiling and 16.5 % after clipping. Good GOSE occurred in 47.3 % and 37.7 %, respectively, and GCS recovery occurred in 79.0 % and 65.1 %, respectively. Apparent AUCs for coiling and clipping were 0.826 and 0.815 for mortality, 0.707 and 0.793 for good GOSE, and 0.768 and 0.765 for GCS recovery. Notably, the retained predictor structure differed by treatment modality: coiling mortality retained age, hypertension, World Federation of Neurosurgical Societies (WFNS) grade, and dome height, whereas the better-performing clipping models incorporated GCS < 15, Intraoperative Monitoring (IOM) availability, aneurysm location, daughter aneurysm, and neck-related morphology. Bootstrap correction preserved the same overall performance ranking, with mortality models remaining the strongest and the coiling good-GOSE model the weakest.

The National Brain Center - General Aneurysm Risk Utility for Decision Aid (NBC-GARUDA) provides preliminary treatment-specific in-hospital risk estimates, with the strongest performance observed for outcome prediction. However, discrimination was heterogeneous and weaker for the coiling favorable-outcome models; therefore, the calculator should be regarded as exploratory pending independent temporal and external validation.