The research team, led by Kai Dai and Yiqiang Feng of Central China Normal University and Sichuan Agricultural University, together with Ziao Chen of the Chinese Academy of Medical Sciences, assembled an annual panel covering 2015 through 2023 for 1,787 county-level units outside municipal districts. The team deliberately restricted the sample to counties whose administrative boundaries remained geometrically stable throughout the period, requiring that each county’s polygon overlap its prior-year shape by at least 90 percent, that area change stay within 5 percent, and that centroid displacement not exceed 10 kilometers. This filtering step matters more than it might appear: when county borders are redrawn, bed counts can shift mechanically between jurisdictions without any real change in service capacity, contaminating the statistics. After the geometry screen, the panel contained up to 16,083 county-year observations drawn from the China County Statistical Yearbook and companion county databases.

The central analytical challenge was to distinguish two forces that both shape where beds end up. The first is territorial planning context, a composite of population scale, geographic dispersion of settlement, and the share of the local economy tied to agriculture, which together describe the kind of service-delivery environment a county faces. The second is local fiscal-economic capacity, combining real fiscal revenue per capita, the ratio of revenue to expenditure, real GDP per capita, and household savings per capita. Because both constructs are multidimensional and their components are correlated, the researchers avoided simply comparing individual regression coefficients. Instead, they estimated planning-context-only, capacity-only, and joint fixed-effects models on identical samples, then used leave-one-block-out changes in within-R-squared and an exact two-block Shapley decomposition to allocate explained variation between the two blocks.

The headline finding is a paradox. For annual changes in total bed stocks, neither block dominated: the planning-context block received 53.1 percent of the Shapley allocation and capacity 46.9 percent, but the joint model explained less than 0.4 percent of within-county variation, so these shares describe an allocation of very little explained movement. For beds per 10,000 residents, the planning-context block appeared to dominate with 71.1 percent of the allocation. But that apparent alignment with planning context was driven almost entirely by a strongly negative population coefficient, meaning counties with larger registered populations tended to have lower bed density. Rather than evidence that beds follow need, the researchers argue, this reflects a partly mechanical relationship, since population appears both in the planning proxy and in the denominator of the density measure.

When the team removed population from the planning-context proxy, the picture reversed. The capacity block then received 57.8 percent of the Shapley allocation and was jointly significant, while the no-population planning-context block fell to 42.2 percent and was only suggestive. The no-population mismatch measure correlated 0.870 with the baseline measure and agreed on high-mismatch classification in 86.9 percent of county-years, so the underlying phenomenon was robust, but the conclusion about which force matters more turned out to be specification-sensitive. The authors are explicit that these decompositions are exploratory model-fit accounting, not definitive evidence about substantive importance.

The longer-term results are where the study delivers its most striking insight. Using long-difference models on a common sample of 1,764 counties, the team found that counties with higher initial planning-context values added more total beds between 2015 and 2023, yet saw their relative mismatch widen. Counties with higher initial fiscal-economic capacity added fewer total beds but enjoyed faster growth in beds per resident, lower relative mismatch, and better odds of escaping the high-mismatch state. The researchers call this an expansion-alignment paradox: the counties most visibly in need of territorial service capacity attracted more nominal bed construction, while counties with stronger fiscal and economic foundations were better able to convert investment into sustained, population-scaled service capacity, likely because doing so requires recurrent financing, workforce recruitment, equipment procurement, and organizational continuity rather than one-off construction spending.

The dynamics of relative disadvantage are equally sobering. The team defined a high-mismatch state as membership in the annual upper quartile of relative planning-supply mismatch, calculated by comparing each county’s planning-context rank with its bed-density rank within the same province and year. Of 3,603 county-years previously in the high-mismatch state, 84.0 percent remained there the following year, while only 16.0 percent exited. By contrast, among county-years previously outside the state, 94.6 percent stayed out and just 5.4 percent entered. In total, 572 counties occupied the high-mismatch state in at least three study years. Higher lagged capacity was associated with a higher probability of exit, and a discrete-time survival model estimated an exit hazard ratio of 1.232 per unit of capacity. But longer spells predicted progressively lower exit hazards, with counties in spells of five or more years showing a hazard ratio of just 0.258 relative to one-year spells, a pattern the authors caution may partly reflect survivor selection rather than true cumulative disadvantage.

Space matters too, though less than raw maps suggest. Global Moran’s I for relative mismatch was positive and permutation-significant in every year, declining descriptively from 0.276 in 2015 to 0.221 in 2023, and local spatial analysis identified persistent high-high and low-low clusters in both endpoint years. Yet the spatial-lag coefficient in a fixed-effects panel model fell from 0.214 under national-year effects to 0.090 once province-by-year effects were included, indicating that much of the apparent spatial dependence reflected common regional and temporal shocks rather than genuine spillovers. Adjacent-county pair models, covering 36,551 pair-years with standard errors clustered on both county endpoints, showed that larger capacity gaps were associated with larger bed-density gaps, and that the higher-capacity county had higher bed density in 63.3 percent of pair-years. Notably, crossing a prefecture boundary did not significantly amplify these gaps, suggesting that capacity gradients, not administrative borders, were the dominant spatial signal.

The authors are careful about what their design can and cannot claim. The study is descriptive and associational throughout; no coefficient is interpreted as a causal effect of planning context, fiscal capacity, or administrative boundaries. The planning-context proxy captures demographic scale, settlement dispersion, and rural production structure, but not morbidity, aging, or disease burden, and the high-mismatch state is a relative distributional classification rather than a clinically verified bed shortage. Robustness checks varied the high-mismatch threshold, the planning-context weights, and the supply measure, and the positive capacity-exit association generally held, though it reversed under a Health Resource Density Index specification that gives greater weight to geographic area, a boundary the authors highlight rather than hide. The COVID-19 pandemic period also attenuated the capacity association with bed-density adjustment, though the planning-context association was unchanged.

The policy implications are concrete. Aggregate bed targets, the authors argue, should be evaluated alongside population-scaled supply and relative territorial alignment, with allocation formulas incorporating population scale, settlement dispersion, existing supply, direct need indicators where available, and prior spell duration. Capital investment should be paired with recurrent financing, workforce support, and operational assistance for lower-capacity counties, since construction alone appears insufficient to convert beds into durable service capacity. Monitoring systems should track entry into, exit from, and duration within the high-mismatch state rather than relying on annual bed counts alone. And because functional service regions may extend across administrative boundaries, coordinated planning should rest on travel times, referral relationships, and patient-flow data rather than contiguity alone. In a health system where nearly 1,800 counties expanded simultaneously yet the relative geography of disadvantage barely moved, the message is clear: the hardest part of health equity is not building more, but building where it counts.

Subject of Research: Territorial equity of hospital and health-center bed allocation across Chinese counties

Article Title: Hospital and health-center bed expansion and persistent relative mismatch across geometry-stable non-district Chinese counties, 2015–2023

Article References: Dai, K., Chen, Z., Wang, L., Lin, Z., & Feng, Y. (2026). Hospital and health-center bed expansion and persistent relative mismatch across geometry-stable non-district Chinese counties, 2015–2023. Frontiers in Public Health, 14, Article 1970301. https://doi.org/10.3389/fpubh.2026.1970301

Image Credits: AI Generated

DOI: 10.3389/fpubh.2026.1970301

Keywords: hospital beds, health equity, China, county health systems, fiscal capacity, spatial analysis, health policy, territorial planning, mismatch persistence, public health, Hospital, health-center