How a Chinese City Rewired Hospital Payments to Push Care Back to the Community

When health systems get sick, the symptoms often show up in the data long before patients notice anything wrong. Hospitals become magnets for every complaint, however minor, while the community clinics meant to be the front door of care sit half empty. In China, this imbalance became one of the most stubborn problems in health policy: a hospital-centred delivery model that rewarded volume, technology, and drug sales rather than prevention and everyday care. A new study from the city of Sanming, published in the International Journal for Equity in Health, offers one of the most detailed quantitative portraits yet of what happened when a single city decided to attack that model at its financial roots.

Sanming, a prefecture-level city in Fujian province, became famous in Chinese health-policy circles for a sequence of reforms that began in the early 2010s and progressively changed how doctors are paid, how hospitals earn revenue, and how money flows through the insurance system. The new analysis, led by Zhe Liu of Fudan University and Tongji Medical College with colleagues from Nankai University, Fudan’s School of Public Health, and Union Hospital, set out to measure whether those reforms actually moved care from hospitals back to primary care, and whether they changed the internal revenue composition of the institutions themselves. The answer, based on twelve years of monthly data, is a cautiously qualified yes.

The study’s engine is a statistical design known as interrupted time-series analysis, or ITSA. Rather than comparing Sanming to a matched control city, the researchers assembled 144 consecutive monthly observations spanning January 2013 to December 2024 and modelled the level and trend of each outcome before and after three policy interruptions: August 2015, January 2018, and October 2021. Each interruption corresponds to a distinct stage of Sanming’s reform programme, allowing the team to ask not simply whether things changed, but when they changed and in which direction. Segmented regression estimates a baseline trend, a level shift at each intervention point, and a change in slope afterwards, which together capture both immediate shocks and gradual drifts in behaviour.

The methodological care is notable. Only genuinely missing values were imputed using statistical modelling; observed zeros and statistically extreme values were deliberately retained rather than cleaned away, a choice that avoids silently erasing real-world events. Standard errors were computed with Newey-West corrections using a three-month lag to handle autocorrelation and heteroscedasticity, common afflictions of long monthly health series. The authors then stress-tested their findings across alternative lag orders, different transition windows around each intervention, and alternative seasonal specifications. Results were broadly stable across all of these checks, which matters because interrupted time-series findings can be fragile when the timing of a policy effect is uncertain or when seasonal illness cycles masquerade as policy signals.

The headline result concerns where people go for care. Mean monthly outpatient and emergency visits at primary care institutions rose from 160,018 in 2013 to 516,885 in 2024, more than a threefold increase, while the primary care share of all such visits climbed from 27.5 percent to 47.7 percent. In a system where hospitals had long absorbed the majority of routine encounters, nearly half of outpatient and emergency care now flowing through primary care represents a structural shift, not a statistical rounding error. The most striking inflection came after the August 2015 intervention: the monthly trend in total visits accelerated by 8,451 visits per month (95 percent confidence interval 4,773 to 12,129), the trend in primary care visits accelerated by 7,284 visits per month (95 percent CI 3,820 to 10,747), and the primary care visit-share trend rose by 0.442 percentage points per month (95 percent CI 0.245 to 0.640), all statistically significant at P less than 0.001.

The January 2018 stage tells a different and arguably more interesting story. After that interruption, the trends in inpatient admissions, total visits, provider-specific visits, the overall hospitalization rate, and the primary care visit share all declined. In other words, the second reform phase appears to have cooled an expansion that the first phase had ignited, tempering growth in hospital activity and even slowing the momentum of primary care uptake. This pattern is consistent with reforms that tightened accountability for admissions and discouraged unnecessary hospitalization, though the study’s design cannot assign the change to any single lever. What it does show is that payment architecture can bend utilization curves in both directions, which is precisely what a policy instrument should be able to do.

The October 2021 stage produced a further increase in the primary care visit-share trend, suggesting that later reforms continued to nudge the system toward community-based care. However, the inpatient findings from this later period were attenuated once the researchers adjusted for the COVID-19 pandemic, a reminder that the pandemic scrambled utilization patterns worldwide and that any before-and-after comparison touching 2020 through 2022 must contend with an unprecedented confounder. The authors are appropriately careful here: the pandemic adjustment softened some conclusions without overturning the core pattern of primary care gains.

Beyond utilization, the study examined how the money itself moved. The researchers tracked hospital and primary care medical revenue, and within each sector decomposed revenue into medical service revenue, pharmaceutical revenue, medical consumables, and examination and laboratory revenue. These shares are diagnostic gold in Chinese health policy because the old fee-for-service regime famously rewarded drug sales, high-tech tests, and consumables over the cognitive work of medicine. A shift in revenue composition away from pharmaceuticals and consumables toward medical services is the financial fingerprint of Sanming’s remuneration reforms, which tied physician pay to service value rather than drug margins. The study reports that successive reform stages were associated with changes in institutional revenue composition, indicating that the payment redesign did not merely redirect patients but also restructured what institutions earned money for.

The authors are explicit about the limits of causal inference in their design. A single-group interrupted time series without a comparison area cannot rule out secular trends, concurrent policies, or broader economic forces as alternative explanations. Their conclusion is deliberately measured: the temporal associations are consistent with, but do not demonstrate, a role for aligned payment accountability, remuneration arrangements, and primary care capacity. That honesty is refreshing in a field where reform success stories are often oversold, and it does not diminish the practical significance of the findings. Replicating Sanming’s trajectory elsewhere would require the same alignment of insurance payment rules, physician salary structures, and grassroots clinical capacity, and the study suggests that misalignment of any one component could blunt the effect.

For the international audience, Sanming’s experiment speaks to a universal question: can you redesign incentives to make primary care the default rather than the afterthought? The Sanming data suggest that when payment accountability, physician remuneration, and community capacity are pushed in the same direction, patient flows follow, with the primary care share of outpatient and emergency visits nearly doubling over the study period. The caveats, no control region, pandemic contamination of the later years, and the inherent ambiguity of before-and-after inference, are real. But the scale and persistence of the shift, documented month by month across twelve years with robust sensitivity analyses, make this one of the most compelling natural experiments in health-system reform to date. As countries from the United States to the United Kingdom wrestle with hospital overload and underfunded primary care, a mid-sized Chinese city has quietly produced a dataset worth studying in every health ministry on earth.

Subject of Research: Provider payment reform and primary care utilization in China's Sanming health reform

Article Title: Payment reform, remuneration incentives, and changes in service utilization and institutional revenue composition: an interrupted time-series analysis from Sanming, China

Article References: Payment reform, remuneration incentives, and changes in service utilization and institutional revenue composition: an interrupted time-series analysis from Sanming, China. (n.d.). https://doi.org/10.1186/s12939-026-03032-5

Image Credits: AI Generated

DOI: 10.1186/s12939-026-03032-5

Keywords: payment reform, remuneration incentives, primary care, interrupted time series, Sanming, China, health policy, service utilization, hospital revenue, health equity, health economics, health system reform