Gray Matter Volume Abnormalities in Schizophrenia: Comparisons Between P -Value and Effect Size Inference Frameworks Using Mega- and Meta-Analyses

BACKGROUND AND HYPOTHESIS: Identifying generalizable brain imaging markers from large multi-center datasets remains challenging due to varying statistical aggregation approaches and p-hacking with increasing big data. We hypothesized that effect size (ES) inference surpasses P-value-based inference in reliably identifying core brain damage of schizophrenia, regardless of whether Mega- or Meta-analyses are used. STUDY DESIGN: We examined voxel-wise inter-group differences in gray matter volume (GMV) based on individual data from 976 schizophrenia patients and 801 healthy controls across 16 datasets, along with published coordinates data from 103 studies involving 5151 patients and 5438 controls, using Mega-analysis (Mega), Image-Based Meta-analysis (IBMA), and Coordinate-Based Meta-analysis (CBMA) under P-value and ES inference frameworks, respectively. We then compared the performances of different statistical aggregation (Mega, IBMA, and CBMA) and statistical inference (P-value and ES) strategies in revealing brain abnormalities in schizophrenia. STUDY RESULTS: P-value Mega identified significant GMV abnormalities in nearly all gray matter voxels (94.85%) with high sensitivity to sample size; in contrast, ES Mega detected core abnormalities in only 24.63% of voxels that had large ES and manifested higher resistance to sample size. ES IBMA and CBMA also demonstrated superior detection performance and were less affected by sample size than P-value ones. Finally, IBMA exhibited comparable performance with the Mega-analysis and superior performance than all types of CBMAs. CONCLUSIONS: These results underscore the advantages of using ES inference in multi-center statistical aggregation and highlight the potential of IBMA for enhanced detection of brain structural abnormalities in schizophrenia.