Skip to content
Review Open access

Can machine learning predict favorable functional outcomes after acute posterior circulation artery occlusion thrombectomy? A systematic review and meta-analysis

Oct 2026 · ANPAbstract_CBN26 · 0 citations

Abstract

Background: Mechanical thrombectomy (MT) is the standard of care for acute posterior circulation artery occlusion (PCAO), yet predicting functional outcomes remains challenging due to the complex interplay of clinical, imaging, and procedural variables. Machine learning (ML) models have emerged as promising tools to improve prognostic accuracy, but their overall diagnostic performance has not been comprehensively quantified. Objective: This systematic review and meta-analysis aimed to evaluate the diagnostic accuracy of ML models in predicting favorable functional outcomes following MT for acute PCAO. Methods: A systematic search of PubMed, Embase, and Web of Science was conducted from inception through August 2025. Eligible studies evaluated ML models predicting favorable outcomes, defined as a Modified Rankin Scale (mRS) score of 0–3 at hospital discharge or 90 days after MT. Data were extracted to construct 2 × 2 contingency tables. Pooled sensitivity, specificity, likelihood ratios, and area under the summary receiver operating characteristic curve (AUC) were calculated using a bivariate random-effects model. Subgroup analyses and meta-regression were performed to explore sources of heterogeneity, and methodological quality was assessed using the PROBAST tool. Results: Five studies comprising 1,739 patients were included. ML models demonstrated good overall diagnostic performance, with pooled sensitivity of 78% (95% CI: 59–89%; I² = 89.29%) and specificity of 80% (95% CI: 74–85%; I² = 46.06%). The summary ROC curve yielded an AUC of 83% (95% CI: 80–86%), indicating strong discriminatory ability. A positive ML prediction (LR+ = 4) increased the probability of favorable outcome from 25% to approximately 57%, while a negative result (LR- = 0.28) reduced it to 9%. In subgroup analysis, Random Forest models achieved significant specificity (p < 0.001), while Support Vector Machine models did not. Studies restricted to successfully reperfused (mTICI ≥ 2b) patients showed statistical significant sensitivity (p < 0.001) and specificity (p < 0.05) compared with worse reperfusion. The Deeks’ funnel plot (Figure 6) showed no significant asymmetry (p = 0.74), indicating a low likelihood of publication bias among the included studies evaluating ML models in patients with acute PCAO undergoing MT. Regarding quality assessment in PROBAST, 3 studies were considered as overall low risk of bias, 1 as unclear and 1 as high. Conclusion: ML models demonstrate good diagnostic accuracy in predicting functional outcomes after thrombectomy for acute PCAO and may support individualized prognostication, treatment decision-making, and resource allocation. However, current evidence is limited by retrospective designs, heterogeneity, and insufficient external validation. Future prospective, multicenter studies with standardized reporting and validation are necessary before routine clinical implementation.

Read PDF

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.