Efficiency and productivity of Mongolian subnational public hospital systems, 2014–2023: annual DEA, global Malmquist and double-bootstrap analysis

Mongolia must organise public hospital capacity across regions that differ greatly in population and geography. We reanalysed subnational hospital efficiency after methodological review, using an auditable production set and inferential procedures tailored to estimated frontier measures. This retrospective longitudinal study included 22 aggregated regional public hospital systems from 2014 to 2023 (220 region-years). Inputs were beds, physicians and mid-level personnel; desirable outputs were inpatient admissions, operations and outpatient visits. Annual input-oriented constant- and variable-returns DEA described contemporaneous efficiency, with pooled DEA and a desirable-output-only slack-based measure as supplementary benchmarks. Productivity was estimated using an output-oriented global constant-returns Malmquist index. A 400-replicate smoothed trajectory frontier bootstrap generated pseudo outputs and re-estimated annual and global frontiers in every replicate. For contextual assessment, Simar-Wilson Algorithm 2 was applied to one decade-average observation per region (22 independent decision-making units), using standardised log population and log geographic area. Mean annual variable-returns technical efficiency was 0.931 (annual range 0.909–0.950); mean pooled variable-returns efficiency was 0.845. Bias-corrected global productivity declined in 2014–2015 (0.933, 95% bootstrap CI 0.911–0.955), 2015–2016 (0.957, 0.927–0.985), 2016–2017 (0.955, 0.940–0.970) and 2019–2020 (0.940, 0.922–0.956), increased in 2020–2021 (1.105, 1.070–1.145), and increased again in 2022–2023 (1.038, 1.018–1.060). In the independent-region double-bootstrap model, neither log population (coefficient −0.007, 95% CI −0.107 to 0.094) nor log area (0.120, −0.093 to 0.201) had a clearly non-zero association with inefficiency. Mongolian regional public hospital systems showed high contemporaneous technical efficiency but several statistically supported periods of productivity decline, including 2019–2020, followed by recovery in 2020–2021. The revised independent-region analysis does not support a robust regression claim that geographic area predicts inefficiency. DEA results are relative production benchmarks, not cost-effectiveness estimates or incremental cost-effectiveness ratios, and should inform monitoring and further investigation rather than punitive rankings.