Determinants of robustness to unmeasured confounding in real-world data: An E-value based meta-regression through the lens of acupuncture studies

Background This study evaluated the robustness of real-world data (RWD)–based observational studies of acupuncture to unmeasured confounding using the E-value as a quantitative index. We examined how data source, exposure specification, and methodological characteristics influence robustness across published literature. Methods A systematic MEDLINE search identified observational RWD studies of acupuncture (2000–2025) reporting effect estimates with confidence intervals. Minimum E-values were computed to quantify the strength of unmeasured confounding needed to nullify observed associations. Study-level variables—including country, data source, disease category, treatment parameters, and analytic methods—were extracted. Group comparisons used nonparametric tests, while segmented temporal regression and ordinary least squares (OLS) meta-regression identified determinants of robustness. Results Sixty-five studies met inclusion criteria, mainly using national insurance claims and medical records. Mean minimum E-values ranged from 1.2 to 2.9. Although no continuous temporal trend was observed (Spearman’s ρ = −0.07, p = 0.60), segmented regression revealed a structural break at 2021, showing gradual improvement until 2021 followed by a marked decline. In OLS models, higher minimum session counts were inversely associated with E-values (p = 0.037), suggesting potential selection bias from longer exposures. Studies using electronic medical records were more robust than claims-based studies, and specification of treatment initiation increased robustness. Propensity score methods were not significantly associated with E-values. Conclusion Robustness of acupuncture RWD studies depends more on data richness and exposure definition than analytic complexity. Post-2021 declines highlight the need for standardized exposure definitions, data integration, and design-based control strategies to strengthen causal validity in nonpharmacological RWD research.