Shanghai’s Fever Clinics Are Running at a Third of Their Potential, Study Finds

When a mysterious respiratory illness swept through a Chinese city in late 2002 and early 2003, the country’s health system responded by building a new kind of frontline defense: the fever clinic. The idea was simple and elegant. Anyone arriving at a hospital with an elevated temperature would be routed into a dedicated, segregated pathway where triage, screening, testing, and referral could happen in one place, away from the general patient population. The model worked well enough that it became a permanent fixture of Chinese public health infrastructure. Two decades later, in 2020, the system was pushed one level deeper into the medical hierarchy, extending down to the primary healthcare centers, the small community clinics that form the very first point of contact between citizens and the state’s medical apparatus. Now, a new study has taken a hard, quantitative look at how that grassroots expansion actually performs on an ordinary Tuesday, and the numbers it uncovered are striking enough to make health economists around the world sit up.

The research, published in BMC Health Services Research by a team led by Yun-yun Yan of Huashan Hospital at Fudan University, examined the fever clinics attached to 177 primary healthcare centers across all 16 districts of Shanghai. Using a multi-stage stratified sampling design and a purpose-built questionnaire, the team collected cross-sectional data on what each clinic consumed and what it produced. On the input side, the analysis tracked seven indicators, including the usable area of the clinic rooms, the cost of hardware configuration, staffing, and the number of operating days. On the output side, four measures captured real clinical work: the total number of patient visits, the total number of confirmed SARS-CoV infections detected, the total number of confirmed influenza A and B infections, and the total number of patients referred upward to hospitals. This input-output framing is the essential raw material for one of the most widely used tools in health services research: data envelopment analysis, or DEA.

DEA is a mathematical technique that treats each clinic as a decision-making unit and constructs an efficiency frontier, the boundary beyond which no unit can produce more output without consuming more input. The method comes in two classic flavors. The CCR model, named after its creators Charnes, Cooper, and Rhodes, assumes constant returns to scale, asking whether a clinic converts its resources into services as well as the best performers in the sample, regardless of size. The BCC model, from Banker, Charnes, and Cooper, relaxes that assumption to allow variable returns to scale, isolating what researchers call pure technical efficiency, the skill of management and clinical practice stripped of the distortions caused by operating at the wrong size. Comparing the two yields a third measure, scale efficiency, which reveals how much of a clinic’s shortfall is purely a matter of being too big or too small for its workload. The Shanghai team ran both models across all 177 clinics and then fed the resulting scores into a Tobit regression, a statistical approach designed for outcomes that are bounded between zero and one, to test how eight external factors shaped the results.

The headline finding is a tale of two efficiencies. The mean comprehensive technical efficiency score across the 177 fever clinics was just 0.34, with a standard deviation of 0.32. In plain terms, the average clinic could theoretically have produced its actual outputs with roughly a third of the resources it consumed, if it had operated as efficiently as the frontier clinics in the sample. Only 20 clinics, about 11.3 percent, achieved a perfect score of 1. Yet when the scale distortion was removed, the picture flipped dramatically. Mean pure technical efficiency came in at 0.84, and mean scale efficiency collapsed to 0.38. The clinics, in other words, are not badly run. Their staff triage competently, their testing pipelines work, their referral pathways function. What is broken is the match between the size of the operation and the volume of demand flowing through it. The inefficiency is architectural, not managerial.

This distinction matters enormously for anyone deciding where the next round of investment should go. The fever clinic system in Shanghai absorbed substantial funding aimed at improving infrastructure, staffing, and equipment, yet preliminary interviews conducted by the research team had already hinted at a paradox: periodic resource depletion coexisting with periodic wastage. Some clinics ran out of capacity during infection surges while others sat idle for weeks. The DEA results formalize that intuition. A clinic can be excellent at its job and still waste most of its budget if it was built for a pandemic-scale patient flow that only materializes a few weeks per year. The study’s authors attribute the observed inefficiency not to a lack of technical capability but to a mismatch between operational scale and demand, a diagnosis that reframes the entire policy conversation from how much to spend to how to size the spending.

The Tobit regression added a second layer of insight by identifying which external factors actually moved the efficiency needle. After removing two variables that suffered from multicollinearity, checked using variance inflation factors, the team correlated five input variables with the four outputs. Two investments showed consistent positive correlations with performance: the usable area of the fever clinic rooms and the cost of hardware configuration. The strongest association appeared with the detection of confirmed SARS-CoV infections, with correlation coefficients of 0.276 for clinic area and 0.346 for hardware spending. Space and equipment, it turns out, translate into real diagnostic yield. But one intuitive investment failed the test entirely. Extending the number of operating days showed no correlation with increased efficiency. Keeping the lights on for more days per week did not generate proportionally more visits, more detections, or more referrals. It simply spread the same thin demand across a longer calendar.

Equally revealing were the factors that made no difference at all. The efficiency of a fever clinic was influenced by objective structural conditions, such as the permanent population of the community it serves and the available building area of the parent primary healthcare center. But it was unrelated to geographical location within the city, to the presence of an on-site X-ray machine, or to the accessibility of CT scanners. This is a quietly radical finding for equipment procurement policy. Expensive imaging assets, often treated as proxies for a clinic’s seriousness, contributed nothing measurable to fever clinic performance. A clinic in a dense neighborhood with adequate space and decent diagnostic hardware outperformed a clinic in a prestigious location stocked with imaging gear it could not efficiently deploy. The lesson is that context, population density and physical capacity, beats prestige and location every time.

From these results the authors draw recommendations that read like a blueprint for post-pandemic respiratory surveillance everywhere, not just in China. In regions with limited resources, or during high-incidence influenza seasons, they argue, point-of-care testing for antigen monitoring should be prioritized. POCT, as it is known, moves diagnostic capability directly to the bedside or the triage desk, delivering results in minutes rather than hours, and it is far cheaper to scale than centralized laboratory infrastructure. More provocatively, the team calls for a hierarchical configuration standard that would allow differentiated and flexible arrangements of hardware and clinic operating times based on community population density and the historical risk of infectious diseases in each district. In other words, stop building identical fever clinics everywhere and start building the right fever clinic for each neighborhood, resizing and rescheduling them as epidemiology and demographics shift.

The broader significance of the study extends well beyond Shanghai’s city limits. Fever clinic networks, or their functional equivalents, now exist in many health systems that spent the early 2020s building frontline respiratory screening capacity. The Shanghai data demonstrate that the default outcome of such rapid, politically driven expansion is a fleet of facilities that are individually competent but collectively oversized, operating at roughly a third of their potential efficiency. The methodological pairing used here, DEA to locate the frontier and Tobit regression to explain the distance from it, offers any health system a replicable audit tool for its own screening infrastructure. The study received funding from the Shanghai Municipal Health Commission’s Medical Nova program and related municipal research schemes, and it drew on the cooperation of the Shanghai Community Healthcare Association and the primary care physicians who completed the surveys.

What makes the findings genuinely viral in their implications is the uncomfortable honesty at their core. Billions were spent globally on frontline screening capacity in the name of pandemic preparedness, and much of that capacity now sits in the same limbo as Shanghai’s: staffed, equipped, clinically sound, and waiting for a surge that comes and goes with the seasons. The Shanghai study suggests the answer is neither to dismantle the network nor to keep it running at full peacetime cost, but to make it elastic, shrinking and stretching with population density, historical risk, and the influenza calendar, with cheap point-of-care antigen testing as the flexible backbone. A fever clinic built for the worst week of a pandemic year should not have to justify its existence during the other fifty-one. The frontier, as the efficiency mathematicians would say, is not a place. It is a shape that must be redrawn every season.

Subject of Research: Operational efficiency of fever clinics in Shanghai's primary healthcare centers measured with DEA and Tobit regression

Article Title: Efficiency of fever clinics in Shanghai’s primary healthcare centers: a DEA-Tobit analysis

Article References: Yan, Y.-Y., Sun, X., Huang, J., Zhu, Z., Liu, M., Fan, T., Du, Z., Huang, Y., & Ying, X. (2026). Efficiency of fever clinics in Shanghai’s primary healthcare centers: a DEA-Tobit analysis. BMC Health Services Research. https://doi.org/10.1186/s12913-026-15867-3

Image Credits: AI Generated

DOI: 10.1186/s12913-026-15867-3

Keywords: fever clinics, Shanghai, primary healthcare centers, data envelopment analysis, Tobit regression, scale efficiency, health services research, pandemic preparedness, point-of-care testing, resource allocation, influenza surveillance, SARS-CoV