Diagnostic Accuracy of Machine Learning Models in Predicting Functional Outcome of Thrombectomy for Acute Posterior Circulation Artery Occlusion: a Systematic Review and Meta-Analysis.

BACKGROUND: Mechanical thrombectomy (MT) is the standard treatment for acute posterior circulation artery occlusion (PCAO), but predicting outcomes remains challenging. Existing prognostic models combine clinical, imaging, and procedural variables but show inconsistent performance. Our objective is to evaluate the diagnostic accuracy of Machine Learning (ML) models in predicting favorable functional outcomes of thrombectomy for acute PCAO. METHODS: We conducted a systematic review and bivariate diagnostic meta-analysis of PubMed, Embase, and Web of Science. Eligible studies evaluated ML predicting favorable outcomes (Modified Rankin Scale of 0 to 3 at hospital discharge or at 90 days) after MT for PCAO. Pooled sensitivity, specificity, and area under the summary receiver operating characteristic curve (AUC) were calculated with a bivariate random-effects model. RESULTS: Five studies including 1739 patients met inclusion criteria. Pooled sensitivity was 78% (95% CI: 59-89%; I2 = 89.29%) and specificity was 80% (95% CI: 74-85%; I2 = 46.06%) for a favorable outcome. The Summary Receiver Operating Characteristic (SROC) curve yielded an AUC of 83% (95% CI: 80-86%). Subgroup analyses revealed that studies including patients with successful reperfusion (mTICI ≥ 2b) had significantly lower sensitivity and specificity. Random forest-based models achieved greater specificity, and multicenter studies demonstrated inferior specificity compared to single-center designs. CONCLUSIONS: ML models demonstrate good diagnostic accuracy in predicting functional outcomes after thrombectomy for acute PCAO. Integration of these models into clinical practice may support individualized decision-making and resource allocation, although prospective validation and improved reporting are needed before routine implementation.