Sustained decline in healthcare-associated infections despite stabilized antimicrobial use: a 10-year study in Southwestern China

Abstract

Healthcare-associated infections (HAIs) and antimicrobial use (AMU) remain major patient safety challenges worldwide, particularly in low- and middle-income settings. However, long-term population-based evidence describing their joint temporal evolution in relation to diagnostic stewardship remains limited. We conducted a 10-year repeated cross-sectional study using annual point-prevalence survey (PPS) data from secondary and tertiary hospitals and other voluntarily participating institutions in Yunnan Province, China, between 2015 and 2024. Temporal trends in HAI prevalence, AMU prevalence, overall culture rate, and pre-prescription culture rate were evaluated using interrupted time series (ITS) analysis as the primary method, supplemented by exploratory segmented regression. Hospital size–stratified analyses were performed to assess differential improvement trajectories across facilities. A total of 1,135,037 inpatients from up to 381 hospitals were included. Overall HAI prevalence declined from 2.11% to 0.92%, representing a 56.4% relative reduction (p for trend < 0.001), with an accelerated decrease observed after 2020. In contrast, AMU prevalence stabilized around 33% after 2020, showing a biphasic pattern with pre-pandemic decline followed by post-pandemic stabilization. The overall culture rate increased from 59.6% in 2015 to 74.5% in 2024. Meanwhile, the pre-prescription culture rate increased substantially, from 49.1% in 2017 to 68.4% in 2024. This divergence between HAI and AMU trends was accompanied by a shift toward more rational prescribing patterns, characterized by higher proportions of therapeutic use and single-agent regimens. Improvements were more pronounced among smaller hospitals, resulting in a narrowing of inter-hospital disparities. Sustained reductions in HAI prevalence were achieved despite stabilization of overall antimicrobial consumption. These findings suggest that pre-prescription culture testing was associated with improved AMU quality and reduced HAI prevalence and strengthened infection prevention efforts. This 10-year real-world evidence supports prioritizing diagnostic stewardship as a core component of long-term infection control strategies, especially in resource-constrained settings.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-46061-4.

Introduction

Healthcare-associated infections (HAIs) represent one of the most significant patient safety threats among hospitalized patients worldwide. The burden is particularly severe in low- and middle-income countries (LMICs), where HAI prevalence has been reported to be more than twice that observed in high-income countries^1,2, with even greater disparities in intensive care units³. Antimicrobial resistance further exacerbates this problem, increasing mortality among patients with HAIs by two- to three-fold^2,4. Although LMICs bear a disproportionate share of the global HAI burden, surveillance capacity remains limited, with only 16% of developing countries having established functional national surveillance systems⁵. More than a decade later, the 2024 WHO Global Report on Infection Prevention and Control reported that only 39% of countries had fully implemented national IPC programs, highlighting the persistent gap in HAI surveillance capacity⁶.

Point prevalence surveys (PPSs) are an important method for assessing the burden of HAIs and patterns of antimicrobial use (AMU)⁷. Since 2011, the European Centre for Disease Prevention and Control (ECDC) has coordinated multiple Europe-wide PPSs, with the 2016–2017 survey covering 28 countries and 1,209 hospitals⁸. National PPSs have also been conducted in Belgium and Scotland, allowing for the evaluation of temporal trends^9,10. However, most international studies are based on single surveys or short-term monitoring over two to three years, and national longitudinal datasets spanning more than five years remain rare¹¹.

In China, a nationwide campaign to regulate the clinical use of antimicrobials was launched in 2011, followed by the release of the National Action Plan to Contain Antimicrobial Resistance (2016–2020) in 2016¹². Nevertheless, existing evidence has largely been derived from single-center, short-term surveillance studies^13,14 or nationwide annual cross-sectional surveys¹⁵. Although the national healthcare-associated infection surveillance network has continued to provide macro-level prevalence benchmarks, continuous provincial-level longitudinal data—particularly from western regions with uneven healthcare resource distribution—remain limited. Overall, both domestic and international studies share notable limitations, including short observation periods, restricted geographic coverage, insufficient long-term evaluation of policy effects, and limited attention to differences across hospital size categories^13,14,16.

Against this background, several key questions remain insufficiently addressed: (1) What long-term trends have occurred in HAI prevalence, AMU prevalence, and culture rates over the decade following policy implementation? (2) How did the 2016 national action plan and the COVID-19 pandemic in 2020 influence these indicators? (3) Do hospitals of different sizes differ in their responsiveness to policy initiatives and the pace of quality improvement, and do related quality indicators exhibit convergence over time^17,18? Addressing these questions is essential for evaluating policy effectiveness and informing future interventions³.

Yunnan Province, a representative region of healthcare development in southwestern China, has conducted annual HAI point prevalence surveys since 2015, accumulating ten years of longitudinal surveillance data covering more than 1.13 million hospitalized patients. The objectives of this study were to: (1) systematically describe ten-year trends (2015–2024) in HAI prevalence, AMU prevalence, and culture rates among participating hospitals in Yunnan Province; (2) assess the impact of the COVID-19 pandemic in 2020 using interrupted time series analysis, with the 2016 national antimicrobial stewardship policies considered as background context and segmented regression^19,20; (3) compare improvement trajectories across hospitals of different sizes, stratified by bed capacity, and evaluate convergence in quality indicators¹⁸; and (4) provide empirical evidence to inform HAI prevention and antimicrobial management policies in resource-limited settings. To our knowledge, this study represents one of the largest and longest longitudinal investigations of HAI and AMU trends in resource-constrained regions, addressing a critical gap in long-term policy evaluation.

Methods

Study design and data sources

This study was designed as a repeated cross-sectional study using data from annual single-day point prevalence surveys (PPSs) organized by the Yunnan Provincial Infection Control Quality Control Center between 2015 and 2024. Each participating institution independently selected one survey day within the designated annual window period (April–November) and conducted one hospital-wide PPS.

All inpatients hospitalized between 00:00 and 24:00 on the survey day were included, including patients discharged on the same day. Patients newly admitted on the survey day and outpatients or emergency department patients were excluded. Eligible institutions included secondary and tertiary hospitals as well as other healthcare institutions that voluntarily participated in the surveys, covering institutions of different levels.

Survey data were reviewed at both hospital and prefecture (municipal) levels before being uploaded to the provincial surveillance platform, where de-identified annual datasets were compiled for analysis. Participating institutions could partially overlap across years, whereas patient-level data represented independent cross-sectional samples for each survey year.

Data collection and quality control

All participating hospitals applied a uniform HAI case definition based on the Diagnostic Criteria for Healthcare-Associated Infections (Trial Version) (Ministry of Health of China, 2001)²¹. Data were collected using a standardized bedside survey form with consistent variable definitions across sites. These indicator definitions and data collection procedures remained consistent throughout the 2015–2024 study period. Bedside data collection was jointly conducted by trained full-time infection control professionals and clinical physicians, with verification through medical record review.

Provincial-level standardized training was provided prior to each annual survey. Logical validation rules were embedded in the provincial data entry system, and records flagged as abnormal were returned to participating institutions for verification and correction. A three-level quality control system (hospital, prefecture, and provincial levels) was implemented, and participating institutions were required to achieve an on-site verification rate of ≥ 96%. Changes in terminology or survey items across years were harmonized during data cleaning to ensure longitudinal comparability. Antimicrobial resistance indicators were not included in the present analysis.

Variable definitions and calculations

The primary outcome indicators included:

• (i)HAI prevalence (%) = number of HAI cases on the survey day ÷ total number of inpatients on the survey day × 100;
• (ii)Antimicrobial use (AMU) prevalence (%) = number of inpatients receiving any systemic antimicrobial therapy on the survey day ÷ total number of inpatients on the survey day × 100. Topical agents alone, anti-tuberculosis drugs, antiviral agents, antifungal agents, and other agents not counted under the AMU definition were excluded;
• (iii)Pre-prescription culture rate (%) (2017–2024) = number of patients in whom specimens were collected and submitted for culture prior to the first antimicrobial dose ÷ number of patients eligible for testing × 100. The eligible population was defined as patients receiving therapeutic antimicrobials or combined therapeutic and prophylactic antimicrobials, whereas patients receiving prophylactic antimicrobials only were excluded from the denominator^22,23.

Secondary indicators included:

• (iv)Overall culture rate (%) = number of eligible patients who underwent bacterial culture testing (regardless of timing relative to antimicrobial use) ÷ number of eligible patients (as defined in (iii) above) × 100;
• (v)Distribution of antimicrobial use purpose (therapeutic, prophylactic, and combined therapeutic plus prophylactic, %);
• (vi)Combination antimicrobial therapy patterns (monotherapy, dual therapy, triple or higher-order combinations, %);
• (vii)Surgical site infection (SSI) rates, stratified by incision class (Class I–IV, %).

All indicators were calculated using pooled numerators and denominators across all participating hospitals within each stratum (i.e., total events ÷ total population at risk), consistent with standard PPS aggregation methodology^24,25. Annual numerators and denominators for all main indicators are provided in Table S4.

Statistical analysis

Descriptive analysis

Continuous variables were summarized using medians and interquartile ranges (IQRs), and categorical variables were summarized as frequencies and percentages (%). Annual trends in primary outcome indicators (HAI prevalence, AMU prevalence, and culture rates) were visualized using line charts. For hospital size–stratified analyses, grouped line plots were used to illustrate temporal trajectories.

Interrupted time series (ITS) analysis

Segmented linear regression models were applied to assess the impact of the COVID-19 pandemic in 2020 on study indicators. The model was specified as:

where Time represents a continuous time variable centered at the intervention point (Time = Year − 2020, such that 2015 = − 5 and 2024 = 4), and Intervention denotes the intervention indicator (0 for years before 2020 and 1 for 2020 onward). Centering at the intervention point ensures that β₂ is interpretable as the immediate level change at 2020. β1 estimates the pre-intervention baseline trend, β2 represents the immediate level change at the time of intervention, and β3 reflects the change in trend (slope) following the intervention. Models were fitted using ordinary least squares with Newey-West heteroscedasticity and autocorrelation consistent (HAC) robust standard errors. Confidence intervals were constructed using t-distribution critical values with model-appropriate degrees of freedom. The year 2020 was selected a priori as the ITS intervention point because the onset of COVID-19 represented a major exogenous disruption to hospital operations, infection control practices, and patient case-mix, including intensified hand hygiene, universal masking, visitor restrictions, and altered admission patterns. National antimicrobial stewardship policies (2011 Special Rectification, 2016–2020 National Action Plan) were treated as background context given their gradual rollout and cumulative effects.

Exploratory breakpoint analysis

As an exploratory supplement to the primary ITS analysis, segmented regression was conducted using the segmented package in R to identify data-driven breakpoints in time series trends. The model iteratively searched for optimal breakpoints and estimated segment-specific annual slopes. These results are presented as exploratory findings to complement the hypothesis-driven ITS analysis.

Hospital size–stratified analysis

Hospitals were stratified into four categories based on the actual number of operational beds (< 300, 300–599, 600–899, and ≥ 900 beds). Interaction terms between time and hospital size were included in regression models to assess differences in improvement rates across hospital size strata. Changes in outcome differences between the largest (≥ 900 beds) and smallest (< 300 beds) hospitals were calculated to evaluate convergence in healthcare quality indicators. Size-stratified analyses used GLS with AR(1) correlation structure; likelihood ratio tests compared nested models with and without the size × time interaction term.

Structural change and subgroup trend analyses

Secondary outcomes were analysed using linear regression for temporal trends in antimicrobial use patterns and the Cochran–Armitage trend test for SSI rates by incision class (2019–2024).

Statistical software

All statistical analyses were performed using R software (version 4.5.1), with the sandwich package for Newey-West robust standard errors and the segmented package for exploratory breakpoint analysis. All tests were two-sided, and a P value < 0.05 was considered statistically significant.

Ethics statement

The study protocol was approved by the Ethics Committee of the First People’s Hospital of Yunnan Province. This observational study used anonymised routine surveillance data and did not affect clinical care. The requirement for informed consent was waived by the Ethics Committee of the First People’s Hospital of Yunnan Province due to the retrospective nature of the study and the use of de-identified aggregated data. The study protocol was conducted in accordance with the principles of the Declaration of Helsinki.

Results

Study population and overall characteristics

Between 2015 and 2024, a total of 1,135,037 hospitalized patients were included in the analysis. The number of participating hospitals increased from 116 in 2015 to 381 in 2024. The distribution of hospitals across four bed-size categories (< 300, 300–599, 600–899, and ≥ 900 beds) changed over time, with a continuous expansion in overall coverage throughout the study period (Table S1).

Ten-year trends in primary indicators

Healthcare-associated infection prevalence

HAI prevalence declined from 2.11% to 0.92% (56.4% reduction) (Fig. 1A). ITS analysis indicated a significant pre-pandemic decline (β₁ = −0.126% points/year, P < 0.001), a modest level increase at 2020 (β₂ = 0.110, P = 0.044), and a further acceleration in decline after 2020 (slope change β₃ = −0.046, P = 0.002). Exploratory segmented regression identified two breakpoints (2019, 2021), corroborating the pattern of stage-specific accelerated decline (Table 1).

Fig. 1.

Ten-year trends of key indicators in province-wide participating hospitals. Four-panel line chart showing HAI prevalence (Panel A), AMU prevalence (Panel B), overall culture rate (Panel C), and pre-prescription culture rate (Panel D) with 95% confidence intervals. Vertical dashed line marks 2020 COVID-19 pandemic onset. Segmented regression lines indicate trend breakpoints.

Table 1.

Interrupted time series and segmented regression analysis of main surveillance indicators (2015–2024).

Indicator	Analysis / parameter	Coefficient (per year)	95% CI	P-value
HAI prevalence rate (%)	ITS (Primary)
	Pre-COVID trend (2015–2019)	−0.126	(− 0.144, − 0.108)	< 0.001
	Level change at 2020	0.110	(0.004, 0.216)	0.044
	Slope change after COVID-19	−0.046	(− 0.068, − 0.024)	0.002
	Post-COVID trend (2020–2024)	−0.172	—	—
	Segmented (exploratory)
	2015–2019 (Segment 1)	−0.126	(− 0.168, − 0.084)	0.003
	2019–2021 (Segment 2)	−0.085	(− 0.268, 0.098)	0.107
	2021–2024 (Segment 3)	−0.187	(− 0.226, − 0.148)	0.002
	Breakpoints: 2019, 2021; AIC = − 32.4
AMU Rate (%)	ITS (Primary)
	Pre-COVID trend (2015–2019)	−0.726	(− 1.094, − 0.358)	0.003
	Level change at 2020	0.764	(− 1.745, 3.273)	0.484
	Slope change after COVID-19	1.021	(0.017, 2.025)	0.047
	Post-COVID trend (2020–2024)	0.295	—	—
	Segmented (exploratory)
	2015–2019 (Segment 1)	−0.726	(− 1.394, − 0.058)	0.041
	2019–2024 (Segment 2)	0.258	(− 0.468, 0.984)	0.379
	Breakpoint: 2019; AIC = 33.4
Overall Culture Rate (%)	ITS (Primary)
	Pre-COVID trend (2015–2019)	0.204	(− 0.297, 0.706)	0.357
	Level change at 2020	2.405	(0.156, 4.654)	0.040*
	Slope change after COVID-19	2.315	(1.367, 3.263)	< 0.001***
	Post-COVID trend (2020–2024)	2.519	—	—
	Segmented (exploratory)
	2015–2019 (Segment 1)	0.204	(− 1.312, 1.721)	0.697
	2019–2024 (Segment 2)	2.511	(1.712, 3.311)	< 0.001***
	Breakpoint: 2019; AIC = 40.7
Pre-prescription Culture Rate (%)	ITS (Primary)
	Pre-COVID trend (2017–2019)	−0.060	(− 1.148, 1.028)	0.886
	Level change at 2020	−5.090	(− 17.291, 7.111)	0.311
	Slope change after COVID-19	6.313	(2.441, 10.185)	0.011
	Post-COVID trend (2020–2024)	6.253	—	—
	Segmented (exploratory)
	2017–2020 (Segment 1)	−2.686	(− 9.342, 3.970)	0.225
	2020–2024 (Segment 2)	6.253	(2.388, 10.118)	0.014
	Breakpoint: 2020; AIC = 48.0

Presents ITS (primary) and exploratory segmented regression results for HAI prevalence, AMU prevalence, overall culture rate, and pre-prescription culture rate. Coefficients represent annual percentage point changes. ITS used 2020 as intervention point; segmented regression identified optimal breakpoints via AIC. * P < 0.05; ** P < 0.01; *** P < 0.001.

Coefficients represent annual percentage-point changes. ITS = interrupted time series (primary analysis), with time centered at 2020 so that the level-change coefficient represents the immediate shift at the a priori intervention point. Newey–West heteroscedasticity- and autocorrelation-consistent (HAC) robust standard errors were used; 95% CI calculated with t-distribution (df = n − 4). Segmented regression was used as an exploratory analysis to identify data-driven breakpoints via the R segmented package. ‡ Only 2 data points in this segment; CI and P-value not estimable. CI = confidence interval; AIC = Akaike information criterion.

Antimicrobial use prevalence

AMU prevalence decreased from 35.62% to 32.17%, corresponding to a relative reduction of 9.7% (Fig. 1B). ITS analysis showed a significant pre-pandemic decline (β₁ = −0.726% points/year, P = 0.003), no significant level change at 2020 (β₂ = 0.764, P = 0.484), and a significant slope change (β₃ = +1.021, P = 0.047), indicating attenuation of the downward trend. Exploratory segmented regression identified 2019 as a breakpoint, consistent with the ITS findings (Table 1).

Pre-prescription culture rate

The overall culture rate increased from 59.56% in 2015 to 74.51% in 2024 (Fig. 1C). ITS analysis showed no significant pre-pandemic trend (β₁ = 0.204, P = 0.357), a significant level increase at 2020 (β₂ = 2.405, P = 0.040), and a further significant slope acceleration after 2020 (β₃ = +2.315, P < 0.001). Exploratory segmented regression identified 2019 as the optimal breakpoint (Table 1). The pre-prescription culture rate increased from 49.11% in 2017 to 68.35% in 2024 (Fig. 1D). ITS analysis showed no significant pre-pandemic trend (β₁ = −0.060, P = 0.886), no significant level change at 2020 (β₂ = −5.090, P = 0.311), but a significant slope acceleration after 2020 (β₃ = +6.313, P = 0.011). Exploratory segmented regression identified 2020 as the optimal breakpoint, with the annual rate of increase shifting from − 2.69% points per year (2017–2020, P = 0.225) to 6.25% points per year (2020–2024, P = 0.014) (Table 1).

Differences across hospital size categories and convergence effects

Hospital size–related differences in HAI prevalence

In 2015, HAI prevalence was higher in large hospitals (≥ 900 beds: 2.69%) than in small hospitals (< 300 beds: 1.30%). By 2024, the gap between the largest and smallest hospitals narrowed from 1.39 to 0.90% points, a convergence of 0.49% points (P for size×time interaction = 0.013). Stratified trends are shown in Fig. 2A; Table 2.

Fig. 2.

Stratified trends of HAI prevalence, antimicrobial use, and pre-prescription culture rate by hospital size (2015–2024). Three-panel ribbon chart displaying HAI prevalence (Panel A), AMU prevalence (Panel B), and pre-prescription culture rate (Panel C) across four hospital size categories (< 300, 300–599, 600–899, ≥ 900 beds). Semi-transparent colored bands show convergence patterns over time.

Table 2.

Stratified Analysis by Hospital Bed Size (2015–2024).

Indicator	Hospital size (beds)	Baseline year	Baseline (%)	2024 (%)	Change (pp)	Size × time P
HAI prevalence (%)	< 300	2015	1.30	0.43	−0.87
	300–599		1.61	0.55	−1.06
	600–899		1.59	0.75	−0.84
	≥ 900		2.69	1.33	−1.36
	Gap convergence†				0.49	0.013
AMU rate (%)	< 300	2015	39.37	37.12	−2.25
	300–599		37.40	32.61	−4.79
	600–899		39.98	29.36	−10.62
	≥ 900		32.51	30.91	−1.60
	Gap convergence†				0.65	0.381
Pre-prescription culture rate (%)	< 300	2017	34.40	55.55	21.15
	300–599		42.66	68.71	26.05
	600–899		48.77	74.71	25.94
	≥ 900		58.75	72.36	13.61
	Gap convergence†				7.54	0.289

Compares baseline (2015/2017) and 2024 values for HAI prevalence, AMU rate, and pre-use testing rate across four hospital size categories. Shows absolute change and gap convergence between largest and smallest hospitals. Size×Time P-value tests for differential trends across categories.

pp = percentage points. † Gap convergence = reduction in absolute difference between ≥ 900-bed and < 300-bed hospital categories from baseline to 2024. Size × Time P-value from likelihood ratio test comparing GLS models with and without size × time interaction term (AR(1) correlation structure). Baseline year is 2015 for HAI and AMU, 2017 for pre-prescription culture rate (earliest available data).

Hospital size–related differences in antimicrobial use prevalence

AMU prevalence declined across all hospital size categories over the study period, although the magnitude of reduction varied (Table 2). Medium-sized hospitals (600–899 beds) showed the largest decline. However, convergence between large and small hospitals was minimal (0.65% points), with no significant difference in improvement rates across size categories (P for size×time interaction = 0.381) (Fig. 2B).

Hospital size–related differences in pre-prescription culture rate

In 2017, substantial disparities were observed in the pre-prescription culture rate across hospital size categories, with small hospitals (< 300 beds) at 34.40% and large hospitals (≥ 900 beds) at 58.75%. By 2024, this gap had narrowed substantially, demonstrating a convergence of 7.54% points, although this difference in improvement rates was not statistically significant (P for size×time interaction = 0.289). Smaller hospitals showed faster improvement rates, narrowing the initial disparity (Table 2; Fig. 2C).

Changes in antimicrobial use patterns and related infection outcomes

Indications for antimicrobial use and combination therapy patterns

Over the study period, the proportion of antimicrobials prescribed for therapeutic purposes increased, whereas the proportion prescribed for prophylactic use remained relatively stable. With respect to prescribing patterns, the proportion of patients receiving monotherapy increased from 83.4% to 89.1%, while no substantial change was observed in the proportion of triple or higher-order combination therapy (Table S2).

Trends in surgical site infection rates

During the surveillance period for surgical site infections (2019–2024), a significant downward trend was observed for Class I (clean) surgical incisions (Cochran–Armitage trend test: Z = − 3.60, P < 0.001). In contrast, no significant temporal trends were observed for SSI rates associated with Class II, III, or IV incisions (Table S3).

Discussion

Principal findings and international comparison

Based on ten years of point prevalence surveillance data from participating hospitals in Yunnan Province (2015–2024), this study systematically characterized long-term trends in healthcare-associated infection (HAI) prevalence, antimicrobial use (AMU) prevalence, and the pre-prescription culture rate. Overall, HAI prevalence declined continuously, with a cumulative reduction of 56.4%. The HAI prevalence observed in 2024 (0.92%) was lower than the average reported in the European PPS (6.5%)⁴ and the results of a recent multicenter PPS conducted in China (1.27%)¹⁵. However, direct comparisons across studies should be interpreted with caution due to differences in case definitions, hospital types, and patient case mix¹⁰. Similar declining trends have been reported in several European countries^10,26, suggesting that sustained and systematic infection prevention efforts may yield consistent effects across different healthcare systems. Notably, this contrasts with findings from the United States and Europe, where device-associated HAI rates increased significantly during the COVID-19 pandemic—one of the most consequential emerging infectious disease events of the 21st century^27–29, suggesting that the sustained decline observed in our study may partly reflect concurrent improvements in diagnostic stewardship.

AMU prevalence exhibited a biphasic pattern, characterized by a decline before the COVID-19 pandemic and a subsequent attenuation of this downward trend thereafter. This contrasts with findings from Sicily, Italy, where three consecutive PPSs reported an increase in AMU prevalence from 50.5% to 53.7%³⁰. Notably, AMU prevalence in the present study remained substantially lower than that reported in a multi-state PPS in the United States, in which 49.5% of hospitalized patients received antimicrobials in 2015³¹.

The overall culture rate reached 74.51% by 2024, indicating substantial improvement in pathogen testing coverage among eligible patients. In contrast, the pre-prescription culture rate increased steadily throughout the study period, an indicator that has been infrequently reported in international PPS studies. The ECDC PPS protocol does not mandate reporting of pre-prescription testing⁴ and the U.S. NHSN system primarily focuses on specific infection types. Existing evidence suggests that obtaining culture specimens prior to antimicrobial initiation improves pathogen identification and supports targeted therapy decisions^32,33. By incorporating this clinical quality indicator into a province-wide longitudinal surveillance framework, our study provides population-level real-world evidence to inform the evaluation of diagnostic stewardship–related policies³⁴.

Hospital size differences and convergence effects

This study observed clear convergence effects across hospital size categories for both HAI prevalence and the pre-prescription culture rate. Smaller hospitals demonstrated faster improvement trajectories than larger hospitals, resulting in a gradual narrowing of inter-hospital disparities and suggesting a catch-up effect. Such convergence patterns have rarely been systematically examined in the PPS literature, where most studies have relied on cross-sectional stratified comparisons rather than analyses of long-term changes in between-hospital gaps. Although PPS studies from Sicily³⁰ and Scotland¹⁰ reported results stratified by hospital size, neither documented clear evidence of convergence over time.

The experience in Yunnan Province may reflect the benefits of coordinated support from the provincial quality control center, the dissemination of standardized procedures, and the implementation of unified hospital accreditation and evaluation standards. Nevertheless, size-related differences in HAI prevalence were not completely eliminated. In 2024, large hospitals continued to exhibit higher HAI prevalence than small hospitals, which likely reflects differences in patient case mix rather than inferior infection control performance. Large hospitals typically care for more critically ill patients and perform more complex surgical procedures, factors that independently increase the risk of HAIs³⁵.

Contrasting trends in AMU and HAI after 2020

Notably, the temporal trends of AMU prevalence and HAI prevalence appeared to diverge during the later years of the study. ITS analysis showed an acceleration in the decline of HAI prevalence after 2020, whereas the downward trend in AMU prevalence plateaued (Table 1). This phenomenon does not undermine the value of diagnostic stewardship; rather, it may reflect improvements in the quality of antimicrobial use rather than further reductions in overall prescribing volume. Although AMU prevalence remained around 33%, the sustained increase in the pre-prescription culture rate suggests that a growing proportion of patients received antimicrobials guided by culture evidence. China’s national stewardship campaign, launched in 2011, was associated with substantial reductions in antibiotic consumption at the national level³⁶. Although subsequent policy documents reinforced the requirement for pre-prescription culture testing, the marked acceleration in implementation observed after 2020 suggests that the COVID-19 pandemic may have served as a catalyst for translating longstanding policy requirements into actual clinical practice.

Concurrently, this study observed substantial optimization in antimicrobial prescribing patterns, including an increased proportion of therapeutic use and a higher prevalence of monotherapy. This shift from empiric combination therapy toward targeted single-agent therapy may help explain why HAI risk continued to decline despite relatively stable overall AMU prevalence. These findings align with recent evidence from the United States emphasizing that antimicrobial stewardship should focus not only on reducing prescription volume but also on improving prescribing appropriateness³⁷. Our results suggest that quality improvement initiatives centered on increasing pre-prescription culture may facilitate a transition from volume-based to precision-oriented antimicrobial management. This is further supported by evidence that diagnostic stewardship interventions targeting culture practices can reduce both inappropriate prescriptions and reportable HAI rates³⁸.

Given the ecological design of this study and the absence of patient-level data, causal inferences cannot be directly drawn. As highlighted by the Sicilian PPS study, the cross-sectional nature of PPS limits causal interpretation³⁰. Although the observed temporal patterns provide indirect support for causal hypotheses, whether improvements in pre-prescription culture directly contributed to reductions in HAI prevalence requires confirmation through patient-level cohort studies. Interestingly, PPS data from the Netherlands reported an opposite trend during the COVID-19 pandemic, with HAI prevalence increasing from 6.4% before the pandemic to 7.4% during the pandemic, largely driven by increases in ventilator-associated pneumonia and central nervous system infections³⁹. This contrast with our findings may reflect differences in healthcare system responses to the pandemic.

Study limitations

This study has several limitations. First, the ecological design based on aggregated annual data precludes adjustment for individual-level confounders, such as patient case mix and comorbidity severity. Changes in the composition of participating hospitals (from 116 to 381 institutions) and potential variations in diagnostic practices over time may also affect comparability. Although the marked decline in Class I surgical site infection rates suggests that the overall reduction in HAI prevalence is unlikely to be solely attributable to changes in case mix, cautious interpretation remains warranted. Second, antimicrobial resistance data were not included, preventing direct evaluation of the full hypothesized pathway linking increased pre-prescription testing to optimized therapy, reduced resistance pressure, and lower HAI risk. Third, pooled aggregate rates may mask between-hospital heterogeneity. Individual hospital-level time series data were not available for multilevel modeling or heterogeneity quantification; the bed-size stratification partially addresses this limitation but cannot fully capture facility-level variation. Fourth, device-specific HAI rates (e.g., ventilator-associated pneumonia, catheter-associated urinary tract infection) could not be calculated, as device-day denominators were not collected in the PPS design; the distribution of HAI by infection site is provided as a supplementary analysis (Table S5). Fifth, the findings are based on data from a single province, and the generalizability to other regions or countries should be interpreted cautiously.

Policy and practice implications

Based on these findings, several implications for antimicrobial stewardship and infection prevention policies can be proposed. First, stratified promotion of pre-prescription culture may represent a key strategy for reducing HAIs. The temporal evidence suggests that incorporating this indicator into core performance evaluation frameworks could be beneficial. Policy targets may be tailored by hospital size, with enhanced laboratory capacity and rapid diagnostic support prioritized for smaller hospitals, and refined management of critically ill patients and complex cases emphasized in larger hospitals.

Second, greater emphasis should be placed on the quality of antimicrobial use rather than solely on prescribing volume. Once AMU prevalence reaches relatively low levels, further reductions may risk delaying necessary therapy. Our findings support a policy shift toward optimizing antimicrobial use quality through standardized pre-prescription testing, rational agent selection, appropriate treatment duration, and timely de-escalation, with the pre-prescription culture rate serving as a practical leverage point.

Third, vigilance is needed to prevent post-pandemic relaxation of infection control and stewardship policies. Although HAI prevalence declined more rapidly after the pandemic, this trend may partially reflect intensified but potentially unsustainable measures. Institutionalizing effective practices implemented during the pandemic—particularly maintaining high adherence to pre-prescription testing—may help sustain long-term gains.

Conclusions

Using ten years of real-world surveillance data, this study provides a comprehensive description of long-term trends and interrelationships among HAI prevalence, AMU prevalence, and the pre-prescription culture rate in participating hospitals in Yunnan Province. The findings suggest that precision-oriented antimicrobial therapy guided by culture-based diagnostics may have potential value in reducing HAI burden and underscore the policy relevance of incorporating pre-prescription testing into core infection prevention and control metrics. Despite the limitations inherent to an ecological study design, this work offers valuable real-world evidence to support the integration of antimicrobial stewardship and infection prevention strategies.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

L.X. and Y.Y. conceptualized the study design. L.X., L.Z., and H.Z. performed data collection and quality control. L.X. conducted the statistical analyses. L.X. and Y.Y. wrote the original manuscript draft. J.L., X.F., and Y.Y. provided critical review and editing. X.F. and Y.Y. supervised the project and acquired funding. All authors reviewed and approved the final manuscript.

Funding

This work was supported by the Science and Technology Project of the First People’s Hospital of Yunnan Province (Grant No. KHYJ-2025-6-08), the High-level Talent Research Grant of the First People’s Hospital of Yunnan Province (Grant No. 2025-KHRCBZ-B05), Yunnan Provincial Academician and Expert Workstation (202505AF350054), and Yunnan Provincial Major Science and Technology Project – Biomedical Special Project (202502AS080003).

Data availability

The datasets generated and/or analysed during the current study are availablefrom the corresponding author upon reasonable request.

Competing interests

The authors declare no competing interests.