Answers are grounded in this paper and cite their sources.
Standardized mean differences (SMDs) depend on within-group standard deviations (SDs). We examined dispersion plausibility and conversion provenance and propose a screening workflow for biomarker meta-analysis. A post hoc audit included 377 two-group comparisons from 140 of 146 pediatric pneumonia studies. We compared coefficients of variation (CVs), Hedges g and ratios of means (RoMs) between directly reported mean ± SD and quantile-derived data. Reporting groups were not exchangeable; their contrast was descriptive. The CV < 0.15 trigger arose during the audit; we examined thresholds from 0.10 to 0.25. Mechanism checks, source-label inspection, and denominator-sensitivity syntheses were supplemented by an illustrative COVID-19 review and scale-structure comparators. Directly reported comparisons had lower median CVs than quantile-derived comparisons (0.214 versus 0.719); 98 of 99 flags were directly reported. Direct reports accounted for 97.5%–99.4% of flags across thresholds. Median signed g was 1.58 versus 0.29, with RoMs of 1.41 versus 1.27. Source-page inspection located SD or conventional mean ± SD notation in all 66 sources contributing flagged direct comparisons, but did not validate numerical dispersion. Neither √n rescaling nor slope tests identified an error mechanism. In the external review, flags occurred in 22/30 median/IQR-sourced and 4/5 mean(SD)-sourced comparisons. In 52/57 evaluable IQR arms, review SDs were arithmetically compatible with Hozo range-based formulas using IQR endpoints and review sample sizes (absolute tolerance 0.011). Arithmetic compatibility did not confirm the algorithm; source-to-review mappings remained unresolved. Conventional pooled SMDs for four cytokines (1.94–2.47) fell to 0.92–1.80 after flagged exclusions. Pooled RoMs were 1.48–1.74; prediction intervals for both strategies remained wide. In this post hoc audit, reporting form was associated with relative dispersion, and pooled SMDs were sensitive to denominator assumptions. A low CV should prompt source verification and sensitivity analysis, not automatic exclusion, rescaling or correction. The workflow is an exploratory screening heuristic with unknown diagnostic performance. Its threshold, derived from pediatric pneumonia biomarkers, should not be applied in other fields without independent calibration. Not applicable.
No takes yet. Share an insight, caveat, or question.
Li et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: