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

In 30 seconds

This systematic review and bivariate diagnostic meta-analysis assessed machine-learning models for predicting favorable functional outcomes after mechanical thrombectomy for acute posterior circulation artery occlusion. Five studies involving 1,739 patients were included; favorable outcome was defined as a Modified Rankin Scale score of 0–3 at discharge or 90 days. Pooled sensitivity was 78% (95% CI: 59-89%; I² = 89.29%) and specificity was 80% (95% CI: 74-85%; I² = 46.06%), with an SROC AUC of 83% (95% CI: 80-86%). The authors judged accuracy good but called for prospective validation before routine use.

Key findings

• Across five studies and 1,739 patients, pooled sensitivity for favorable outcome was 78% (95% CI: 59-89%; I² = 89.29%).
• Pooled specificity was 80% (95% CI: 74-85%; I² = 46.06%); the SROC AUC was 83% (95% CI: 80-86%).
• Studies including patients with successful reperfusion (mTICI ≥ 2b) had significantly lower sensitivity and specificity.
• Random forest-based models had greater specificity, while multicenter studies showed lower specificity than single-center studies.

Why it matters

Outcome prediction after thrombectomy for acute posterior circulation occlusion is challenging. These pooled results suggest machine-learning models may help estimate the likelihood of favorable functional recovery and inform individualized decisions or resource allocation. However, substantial variation in sensitivity across studies and differences by reperfusion status and study setting mean that pooled accuracy should not be assumed to apply uniformly.

What to keep in mind

Only five studies were included, and sensitivity showed substantial heterogeneity (I² = 89.29%). The authors state that prospective validation and improved reporting are needed before routine implementation.