🤖 AI Summary
In random-effects meta-analyses with few studies (k < 10), the Higgins–Thompson–Spiegelhalter (HTS) prediction interval exhibits severely subnominal coverage—often below 90%—due to poor estimation of between-study heterogeneity τ². To address this, we propose a novel prediction interval construction method integrating bootstrap resampling with confidence distribution theory. Unlike conventional approaches that substitute τ² with point estimates, our method directly models the uncertainty in τ² via bootstrap resampling and derives the prediction distribution through confidence distribution inference, thereby ensuring nominal 95% coverage. Simulation results demonstrate that, for k = 3–8, the proposed method achieves stable coverage of 94.2%–95.6%, substantially outperforming HTS (72.5%–89.1%). Empirical evaluation across three medical meta-analyses further confirms its robustness and practical utility.
📝 Abstract
Prediction intervals are commonly used in meta-analysis with random-effects models. One widely used method, the Higgins–Thompson–Spiegelhalter prediction interval, replaces the heterogeneity parameter with its point estimate, but its validity strongly depends on a large sample approximation. This is a weakness in meta-analyses with few studies. We propose an alternative based on bootstrap and show by simulations that its coverage is close to the nominal level, unlike the Higgins–Thompson–Spiegelhalter method and its extensions. The proposed method was applied in three meta-analyses.
