Associations of DRG-Based Payment Reforms With Hospitalization Costs and Length of Stay: A Systematic Review and Meta-Analysis
Jiaman Liao, Changhai Wu, Xiaoling Deng, Qianhua Huang, Danni Zheng, Yawei He, Yiqiang ZhanIntroduction
To systematically evaluate the associations of diagnosis-related group (DRG)-based payment reforms with hospitalization costs and length of stay (LOS).
Methods
We conducted a systematic review and meta-analysis in accordance with the PRISMA 2020 statement. Quantitative studies evaluating DRG-based payment reforms were eligible. Evidence was synthesized according to study design, outcome, effect estimate, and effect scale. Random-effects models were used to pool comparable adjusted difference-in-differences (DID) estimates and interrupted time series (ITS) level and slope changes. Studies without comparable policy-effect estimates were synthesized narratively. This systematic review and meta-analysis was retrospectively registered with INPLASY (registration number: INPLASY202690066; DOI: 10.37766/inplasy2026.9.0066).
Results
Twelve studies were included, including 11 from China and one from Switzerland. Three DID studies reporting comparable hospitalization-cost estimates yielded a pooled log-scale coefficient of −0.0526 (95% CI, −0.1215 to 0.0163; P = 0.135; I 2 = 90.3%), corresponding to an approximately 5.1% reduction, although the estimate was not statistically significant. ITS findings varied by effect measure, with inconsistent evidence for post-intervention trends. The pooled proportional immediate change in LOS was −0.0178 (95% CI, −0.0231 to −0.0124), whereas the pooled proportional slope change was −0.0020 (95% CI, −0.0087 to 0.0046; I 2 = 95.4%). None of the pooled estimates remained statistically significant under modified Knapp–Hartung small-sample sensitivity analyses. Narrative evidence suggested heterogeneous findings, with some studies reporting reduced hospitalization costs and mixed LOS results.
Conclusions
DRG-based payment reforms may be associated with lower hospitalization costs and, in some settings, shorter LOS; however, the magnitude and consistency of these associations vary across study designs and implementation contexts. Findings should be interpreted cautiously given the limited number of comparable quasi-experimental studies and substantial heterogeneity in some analyses.