meta-analytic synthesis

This paper addresses the limited efficacy of conventional meta-regression methods under dual heterogeneity—both geographic and temporal. We systematically evaluate traditional approaches (random-, fixed-, and mixed-effects models) against seven novel specifications, including our proposed “location-fixed-effects + time-trend” hybrid model. Through large-scale simulation studies, we quantitatively compare methods across three key dimensions: statistical power, estimation bias, and model robustness. Results demonstrate that jointly modeling dual heterogeneity substantially outperforms unidimensional modeling: under high heterogeneity, it improves statistical power by 32% and reduces bias by 41%. Our hybrid specification achieves an optimal balance of flexibility and robustness. Furthermore, we provide a practical model-selection framework grounded in information criteria and robustness diagnostics. This work advances methodology for causal inference in settings characterized by multi-source spatiotemporal heterogeneity.

Traditional random-effects meta-analysis suffers from substantial bias in heterogeneity estimation and overly wide confidence intervals when applied to small-scale evidence syntheses (2–4 studies). To address this challenge in small-sample medical meta-analysis, we propose a novel evidence synthesis framework leveraging within-trial subgroup data. Our approach jointly models subgroup-level effect estimates and introduces two improved DerSimonian–Laird–type heterogeneity estimators. We further enhance uncertainty quantification by incorporating Henmi–Copas variance correction and adaptive degrees-of-freedom t-distribution inference. Simulation studies demonstrate that the method markedly improves accuracy of heterogeneity estimation and achieves superior balance between confidence interval coverage probability and width across diverse subgroup effect structures and prevalence scenarios. Empirical application to real-world clinical examples confirms its robustness and practical utility.

This study addresses key challenges in synthesizing causal evidence from heterogeneous sources—namely, individual-level and aggregate-level data—including poor integrability of treatment effects across target populations, limited transportability, and insufficient personalization. To this end, we propose a target-population-characteristic-aware weighted evidence synthesis framework. Our method innovatively introduces a sample-bounded weighting scheme to enable customized meta-analysis; develops a principled approach to identify studies whose populations deviate from the target, thereby enhancing robustness and interpretability; and integrates causal inference with meta-analytic principles, ensuring asymptotic normality of the estimator under multiple consistency conditions. Simulation studies and real-data analyses demonstrate that the proposed weighted estimation method significantly improves both accuracy and personalization of treatment effect estimation, outperforming conventional regression-based meta-analytic approaches.

This study addresses the inconsistency in causal effect estimates between observational studies and randomized controlled trials (RCTs) by proposing the first unified framework for decomposing causal effect heterogeneity. The framework systematically identifies and quantifies three sources of heterogeneity: differences in covariate distributions, variation in mediating pathways, and shifts in outcome-generating mechanisms. Methodologically, it formally defines effect decomposition across data types (observational vs. experimental), integrating causal inference, sensitivity analysis, and decomposition modeling, while enabling robust parameter estimation under multiple hypotheses. Evaluated through simulation studies and an empirical analysis of the “Moving to Opportunity” experiment, the framework demonstrates improved interpretability, robustness, and policy generalizability in synthesizing evidence from heterogeneous data sources.

This study addresses the challenge of sparse 2×2 contingency tables in medical meta-analyses caused by rare events, where existing methods often yield unreliable inferences. The authors propose a novel unified inference framework that integrates confidence distributions with Poisson-pair modeling, enabling optimal estimation of treatment effects and their ratios under both fixed- and random-effects models. By introducing Poisson-pair modeling into confidence-distribution-based meta-analysis for the first time, the method substantially enhances the accuracy and reliability of statistical inference in rare-event settings. Empirical evaluation on real-world datasets demonstrates its superior performance compared to conventional approaches.

This study addresses the frequent under-identification and inadequate interpretation of outliers, non-replicable findings, and highly influential studies in meta-analyses, which often compromise the robustness of conclusions. It clarifies conceptual distinctions among these three types of problematic studies and proposes a systematic diagnostic framework that integrates robust statistical methods, graphical diagnostic tools, and advanced modeling techniques accounting for sampling variance dependencies. This approach enables more accurate detection of anomalous studies while leveraging visualization to facilitate interpretation of their potential sources. By synthesizing recent methodological advances, the work offers meta-analysts practical diagnostic strategies and cautious interpretive guidance, substantially enhancing the reliability and transparency of meta-analytic results.

This study addresses the lack of clear guidance in selecting network meta-regression (NMR) models, where misspecification can induce bias in treatment effect estimates. Through 120 simulated evidence network scenarios, it systematically compares standard network meta-analysis with four NMR variants—differing by common or independent, and consistency or inconsistency interaction terms—across varying levels of heterogeneity, network density, and multi-arm trial structures. The work reveals, for the first time, how specific network characteristics influence NMR bias and confidence interval coverage, demonstrating that omitting effect modification leads to overestimation of treatment effects. Independent-interaction NMR performs robustly in dense networks, whereas consistency-interaction models are better suited for networks containing multi-arm studies. The study advocates aligning model assumptions with underlying network features to ensure reliable evidence synthesis for informed decision-making.

Traditional meta-analyses struggle to quantify the strength of evidence for the presence or absence of an effect and cannot adequately assess the sensitivity of conclusions to publication bias or small-study effects. This work proposes a Bayesian evidence auditing framework tailored to meta-analytic corpora, integrating bias-aware models with unbiased baseline specifications through Bayesian model averaging and Bayes factors. It introduces “rigor” as a composite metric that jointly evaluates the strength of evidence for an effect and robustness to bias—allowing null effects to achieve high rigor scores when supported by strong evidence. Built upon Bayesian random-effects models, the approach employs simulation and resampling strategies within the ADEMP framework, including synthetic data generation, registered-report resampling, and contour-enhanced funnel weighting. Applied to nutritional intervention studies, the method frequently attenuates conventional effect estimates, revealing that many nominally significant findings lack robust evidential support. Full reproducible resources are publicly released.

This study addresses the imprecise inference in subgroup interaction meta-analyses under sparse data, which stems from the absence of empirical prior distributions tailored to interaction heterogeneity. Leveraging over 3,000 interaction meta-analyses from the Cochrane Database of Systematic Reviews, we construct the first treatment-by-subgroup interaction–specific empirical prior distribution, revealing that such interaction heterogeneity is typically substantially smaller than that of overall treatment effects. By integrating a Bayesian random-effects model with large-scale data mining and predictive prior derivation, the proposed prior markedly improves estimation accuracy in sparse-data settings compared to conventional heterogeneity priors, thereby offering a more reliable evidentiary foundation for subgroup analyses.

In high-dimensional clustered data, when covariates exhibit heterogeneous distributions across clusters, conventional marginal LASSO may erroneously treat them as sparse proxies for latent cluster effects, leading to biased estimation and incorrect variable selection. This work proposes the Synthetic Heterogeneous Effects LASSO (SHEL), which, for the first time, integrates cluster-level synthetic variables into a fixed-effects penalized regression framework to explicitly model latent heterogeneity and correct estimation bias. SHEL enables accurate variable selection and valid post-selection inference in high-dimensional settings. Theoretical analysis establishes its desirable asymptotic properties under high dimensionality, while simulations demonstrate substantial improvements over existing methods. The approach is successfully applied to longitudinal RNA-seq data from neutrophils of COVID-19 patients, illustrating its practical utility.