Evaluating Outcome Prediction Models in Endovascular Stroke Treatment Using Baseline, Treatment, and Posttreatment Variables

Background Statistical models to predict outcomes after endovascular therapy for acute ischemic stroke often incorporate baseline (pretreatment) variables only. We assessed the performance of outcome prediction in an iterative fashion using baseline, treatment‐related, and posttreatment variables. Methods Data from ESCAPE‐NA1 (Safety Efficacy Nerinetide [NA‐1] Subjects Undergoing Endovascular Thrombectomy Stroke) trial were used build 4 multivariable logistic regression: model 1 included available before treatment decision making, 2 additional treatment‐related variables, 3 that become early (within 24–48 hours), later (beyond 48 hours) therapy. The primary was functional independence (90‐day Modified Rankin Scale score 0–2). Model compared area under receiver operating characteristic curve (AUC). Shapley values determine marginal contributions variance regression models. Results Among 1105 patients, achieved by 666 (60.3%). When only (model 1), AUC 0.74 (95% CI, 0.71–0.77); this iteratively improved when added 2: AUC, 0.77; 95% 0.74–0.80; 3: 0.80; 0.77–0.83; 4: 0.82; 0.79–0.85). With alone, 26% patients who erroneously classified as not achieving independence. Even with most comprehensive model, 19.8% misclassified such. Patient age contributed (Shapley value, 0.28), followed severe adverse events including pneumonia (0.16) intracranial hemorrhage at 24‐hours imaging (0.13). Conclusions A substantial contribution comes factors unrelated currently collected patient One‐fifth independence, even model. Our findings suggest achievable accuracy current is limited, caution should be applying them clinical practice.