Comparison of pragmatic post-thrombectomy prognostic scores not based on advanced imaging in a large national stroke registry

Background: Nearly half of the patients who received endovascular thrombectomy (EVT) for large vessel occlusion experience poor functional outcomes. Reliable tools for early post-procedural prognostication are needed. We aimed to assess and compare the performance of existing, pragmatic post-EVT prognostic scores in a large national multicenter cohort. Methods: We conducted a systematic literature search to identify pragmatic post-thrombectomy prognostic scores predicting 90-day functional outcomes. Models relying on advanced imaging, small derivation samples, or machine learning were excluded. We analyzed data from the IRETAS registry-a prospective, multicenter Italian cohort of stroke patients treated with EVT. Inclusion criteria were pre-stroke modified Rankin Scale (mRS) <= 2 and available 90-day mRS. The primary outcome was good functional outcome (mRS <= 2). Prognostic performance was assessed using c-statistics in the samples where each individual score was measurable. Scores were compared using DeLong tests in the subset of patients for whom all scores were measurable. Results: Three scores were identified: HERMES-24, BET, and SNARL. Among 22,768 patients in the registry, 18,408 (89.1%) had a measurable HERMES-24 score, 13,593 (59.7%) had a measurable BET score, and 19,007 (83.5%) had a measurable SNARL score. Median age was 75 years (IQR 65-82), and 11,528 (50.6%) were female. In the subset in which each test was measurable, HERMES-24 showed the best performance for predicting mRS <= 2 (c-statistic = 0.889), followed by BET (c-statistic = 0.794) and SNARL (c-statistic = 0.762) (p < 0.001). In the subset of 12,233 patients for whom all three prognostic scores were calculable, a head-to-head comparison confirmed the superior performance of the HERMES-24 model: HERMES-24 score versus BET score (c-statistic difference = 0.098 [95% CI = 0.092-0.105]; p < 0.001) and HERMES-24 score versus SNARL score (c-statistic difference = 0.124 [95% CI = 0.116-0.132]; p < 0.001). Conclusions: In this large, multicenter, national cohort, the post-EVT HERMES-24 score-which accounts only for age and 24-h NIHSS-demonstrated the highest prognostic performance among existing, pragmatic post-EVT scores. Its simplicity and robust performance support its routine adoption in clinical practice.