The effect of mandatory regulatory policies on unsafe sexual behaviours among female sex workers: an evaluation study Authors: Wangjin Liang, Jian Chen, Yaqin Pang, Ruchen Zhang, Xujiao Li, Lanfen Tan, Jianying Liao, Xu Liang Published: 2026 ABSTRACT AbstractBackgroundSocial policies shape individual behaviours and population health. Human Immunodeficiency Virus (HIV) prevention governance measures targeting female sex workers (FSWs) remain controversial. Existing evidence mainly relies on qualitative or cross-sectional self-reported data, limiting objective population-level assessments. Using longitudinal data from China’s HIV sentinel surveillance system, we evaluated whether mandatory regulatory measures have influenced unsafe sexual behaviours among FSWs.MethodWe treated the 2015 policy implementation in Baise City, Guangxi, China, as a natural experiment. Using data from eleven counties covered by the National HIV Sentinel Surveillance System (2010–2020), we applied the Cochran–Armitage trend test and interrupted time series analyses to evaluate changes in AIDS knowledge awareness rate (AKAR), condom use rate at last commercial sexual encounter (CUR-LCSE), and active syphilis positivity rate (ASPR). Deviation of ASPR from the expected level was estimated using the event study method, and the policy effect was validated by a robustness test.ResultsAmong 19 220 observations, AKAR and CUR-LCSE showed significant upward trends from 2010 to 2020 (both p < 0.01). ASPR remained stable before 2015 (p = 0.53) but declined markedly thereafter (p < 0.01). Event study analyses detected no systematic deviation between observed and expected ASPR; robustness tests indicated that the post-2015 decline was random (t = − 0.956; p = 0.39). Overall, mandatory regulatory policies were not associated with measurable changes in unsafe sexual behaviour.ConclusionMandatory regulatory policies showed no clear evidence of elevated syphilis risk among FSWs. The stable prevalence of active syphilis and consistent behavioural indicators further confirmed that these policies did not inadvertently promote unsafe sexual behaviours, and serological findings indicated a lack of substantial behavioural changes in this population. This highlights an urgent need to clarify how structural and social determinants shape unsafe sexual behaviours to inform the development of more integrated and equitable syphilis prevention strategies.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12889-026-28083-9.Keywords: Female sex workers (FSWs), Mandatory regulatory policy, Unsafe sexual behaviours, Interrupted time series, Event Study Method Introduction According to a 2025 World Health Organization report, an estimated 39.9 million people were living with HIV in 2023, with 1.3 million new infections and 630 000 HIV-related deaths recorded that year [1]. Unsafe sexual behaviours substantially increase the risk of HIV and other sexually transmitted infections (STIs), serving both as a major driver of HIV acquisition among female sex workers (FSWs) and an important determinant of onward HIV and STI transmission [2, 3]. In this study, FSWs refer to individuals who were assigned female sex at birth. FSWs and their clients are an important bridge population in HIV epidemics, creating key viral transmission pathways across otherwise disconnected social groups. High mobility and contact with multiple concurrent partners generate complex sexual networks that increase the contribution of FSWs to maintaining HIV transmission chains [4]. From a population health perspective, a primary objective of public health interventions is to reduce the total volume of high-risk sexual encounters. Theoretically, mandatory regulatory measures may lower this volume by increasing the economic and social costs associated with commercial sex transactions, including heightened legal risks, operational barriers, and elevated transaction prices for clients, which can shrink the FSW population and reduce the frequency of high-risk contacts, thereby curbing the broader transmission of STIs [5–7]. Reducing unsafe sexual behaviours and high-risk interactions within this group remains a global priority. Targeted interventions are therefore essential—not only to mitigate infection risks for FSWs themselves but also to break the chain of HIV transmission to the wider community. Since the introduction of the “health promotion” framework as articulated in the Ottawa Charter for Health Promotion in 1986, there has been a greater global emphasis on the pivotal role of governmental and social governance structures in shaping population health. Substantial evidence shows that legal and policy interventions can influence health outcomes by reshaping the broader social environment [8, 9]. Although commercial sex work is illegal in most countries, the use of mandatory public health control measures targeting FSWs remains highly contested. The central debate revolves around two competing structural pathways. On the one hand, strengthening law enforcement may increase risks by driving sex trade activities into more concealed environments, making it harder for health services to access this population. On the other hand, mandatory regulatory policies might reduce STI prevalence by limiting the operational space for commercial sex and decreasing the overall frequency of transactions [10, 11]. A key issue, however, is whether these measures inadvertently lead to an increase in unsafe sexual behaviours, thereby raising the risk of HIV infection. Although the prevailing public health perspective suggests that these policies may shape FSWs’ risk behaviours, this view is largely informed by qualitative and cross-sectional studies [12–14]. Furthermore, this perspective conflicts with governance principles endorsed by many countries, complicating the alignment of HIV prevention strategies with national policy frameworks and contradicting core health promotion tenets, namely, intersectoral collaborative governance and shared risks and responsibility between society and individuals [15]. Previous research has predominantly used qualitative designs or single-year, cross-sectional data, with limited long-term serological or behavioural monitoring and few cohort studies capable of establishing causality. Consequently, robust evidence on whether mandatory regulation increases HIV risk among FSWs remains insufficient. Current Chinese law prohibits commercial sex and includes policies aimed at safeguarding population health, preventing infectious disease transmission, and maintaining social stability. From a social governance perspective, these measures partly help to reduce the risk of STI epidemics [16, 17]. To further strengthen social order and sustain the decline in crime, the Chinese government issued the “Opinions on Strengthening the Construction of the Social Security Prevention and Control System” in 2015 (hereafter referred to as the “Opinions”). This policy aimed to influence STI transmission by strengthening legal measures, implementing real-name registration requirements, and expanding collaborative governance. These actions effectively increased transaction costs and limited the operational space of the sex trade. As a result, the policy reduced illegal activities, including commercial sex work [18]. This policy shift created a natural experiment suitable for causal inference in the present study. We utilized China’s HIV/AIDS Case Reporting and Information Management System, which has been operational since 2009, to assess whether structural governance changes have led to a reduction in the overall frequency of unsafe sexual behaviour or inadvertently increased risks. This system systematically collects long-term serological and behavioural data on key populations, including FSWs. This comprehensive and continuous surveillance infrastructure provides a robust foundation for evaluating the population-level effects of policy changes. We combined a community-level context with long-term systematic FSW sentinel surveillance data from 2010 to 2020 in Baise City, Guangxi. Baise was selected because its eleven contiguous counties each maintain stable, well-established surveillance infrastructure and have sustained collaborative partnerships, which offered operational convenience. We used interrupted time series (ITS) analysis and the event study method (ESM) to assess the effect of regulatory measures implemented from a public affairs management perspective on unsafe sexual behaviours and syphilis prevalence trends among FSWs. Measures included AIDS knowledge awareness rate (AKAR), behavioural indicator, and active syphilis positivity rate (ASPR). Based on the logic of social governance, we hypothesised that mandatory regulatory policy would lead to a decrease or stabilisation in syphilis prevalence at the population level. This is expected because the policy increases the economic and social costs of commercial sex, thereby significantly reducing the overall frequency of high-risk transactions. The study addressed two questions: (1) whether regulatory policies influence trends in AIDS knowledge, behavioural indicators, and serological outcomes before and after implementation; and (2) whether these policies affect the prevalence of unsafe sexual behaviours within the target population. Methods Data sources We analysed HIV sentinel surveillance data from Baise City, Guangxi (2010–2020), collected via China’s voluntary HIV surveillance system in accordance with the “National HIV Sentinel Surveillance Implementation Plan”. Field surveys were conducted at community sites frequented by FSWs across all eleven counties. Data included sociodemographics, AIDS knowledge, condom use rate at last commercial sexual encounter (CUR-LCSE) as a behavioural indicator, and syphilis serological indicators (ASPR). AIDS knowledge was evaluated through standardised questionnaires of the National HIV Surveillance System in China. The questionnaire mainly assessed the basic knowledge related to AIDS, including transmission routes, prevention methods, and common misconceptions. An additional AIDS knowledge item file shows this in more detail [see Supplementary Materials 1]. Serological results were laboratory-confirmed, and behavioural and knowledge data were self-reported. Data were collected annually between April and June. According to national guidelines, active syphilis was defined as positivity on both enzyme-linked immunosorbent assay and toluidine red unheated serum test. To ensure the temporal comparability of this large-scale, long-term surveillance data, this definition has been consistently applied. Given that the indicator is based on cross-sectional, anonymous sentinel surveillance data, it should be interpreted primarily as a population-level epidemiological indicator that reflects overall active syphilis trends, rather than as definitive evidence of recent individual infections. Study design We used a combined analytical strategy integrating the Cochran–Armitage test for trend (CAT), ITS analysis, and the ESM. Given the nationwide implementation of the policy without a conventional external control group, this study was designed as a natural experiment. A key assumption underlying the ITS and event study framework is that pre-intervention trends can approximate the counterfactual trajectory in the absence of the policy. To strengthen the plausibility of this assumption, we assessed potential time-varying confounders. No major new HIV prevention programmes targeting FSWs were introduced during the study period beyond routine services, and no abrupt economic changes coincided with the policy implementation. However, unmeasured factors—such as changes in surveillance practices or population mobility—may still introduce residual confounding. China issued the “Opinions” in April 2015. Allowing for the necessary period of administrative adjustment and enforcement alignment, the effective implementation year was defined as 2016. Because annual surveillance data, collected from April to June each year, capture syphilis cases identified since the preceding survey, the effect of the “Opinions” on local FSWs was assumed to begin in 2016. Accordingly, 2016 was assigned as the event time point (T = 0), with the event window as 2016–2020 and the estimation window as 2010–2015. This temporal structure was applied consistently across the ITS and CAT analyses. ITS analysis evaluates the causal effect of an intervention by comparing changes in the trajectory of a continuous outcome before and after the intervention is implemented [19, 20]. We used annual data, defining 2016 as the interruption point and dividing the timeline into the pre-policy period (2010–2015) and the post-policy period (2016–2020). We examined changes in both the level and slope of three indicators—AKAR, CUR-LCSE, and ASPR—to independently assess the immediate and sustained effects of the long-term policy intervention. Given that ITS is limited in its ability to distinguish between true intervention effects and random fluctuations within the short post-intervention period, ESM was adopted as a counterfactual-based validation tool. ESM constructs expected trajectories based on pre-intervention trends and quantifies the deviation between observed and expected outcomes after intervention, thereby assessing whether actual changes deviate significantly from the range of random fluctuations. The central role of the ESM in this study is to assess whether the resulting deviations reflect systematic policy effects or random variation, thereby providing an independent robustness test of ITS results. ITS and ESM play complementary but distinct roles in the analysis. ITS provides estimates of level and slope changes based on models, while ESM validates the observed deviations within a short intervention window through an alternative framework. The ESM provides a robust framework for sociological causal inference, enabling the identification of policy-induced effects by comparing trajectory changes before and after a policy shock while simultaneously accounting for underlying secular trends and potential confounders [21]. By combining these methods, we harness the advantages of long-term trend monitoring alongside the estimation of immediate effects, which improves control over potential temporal confounding variables. This integrated approach enhances the validity of causal inference, providing a more robust and comprehensive framework for policy effect evaluation. We used ASPR as a serological outcome indicator. As syphilis is a curable sexually transmitted infection, a positive serological result often reflects a recent infection typically acquired within the previous 4 weeks to 6 months [22]. Because syphilis is mainly transmitted through sexual contact and its acquisition risk is strongly associated with condomless sex, its prevalence may serve as an indirect proxy reflecting potential changes in unsafe sexual behaviours among FSWs [23]. Nevertheless, ASPR does not serve as a direct metric of unsafe sexual behaviour, which necessitates a clear distinction between ASPR and the self-reported behavioural indicator. The CUR-LCSE captures individual behavioural intentions and actual practices during the most recent commercial sexual encounter, whereas ASPR represents a biological outcome shaped by multiple factors beyond individual sexual behaviour. These confounding factors include testing coverage, prior treatment history, and syphilis reinfection, among others. As noted earlier, ASPR is a population-level surveillance indicator and should not be overinterpreted as direct evidence of individual-level behavioural change. In this context, ASPR provides a relatively sensitive marker of short-term, population-level variations in risk exposure. In contrast, HIV antibody positivity cannot reliably distinguish between recent and long-standing infections, limiting its utility for capturing short-term behavioural changes within a population [24]. Statistical analysis To describe short-term and long-term trends in AIDS knowledge, behavioural indicators, and serological outcomes (Objective 1), we used CAT and ITS analysis. CAT was conducted for each outcome, with survey year as the row variable and the following as column variables: adequate AIDS knowledge, CUR-LCSE, and ASPR. A significant linear-by-linear association indicated the presence of a secular trend. For ITS analysis, stationarity and autocorrelation were assessed using the KPSS test and the Durbin–Watson statistic, respectively. A KPSS p value > 0.05 indicated a stationary series; non-stationary series were log-transformed to achieve stationarity. When autocorrelation was absent, parameters were estimated using ordinary least squares. If autocorrelation was detected, iterative generalised least squares were used, and a segmented regression model was fitted. The ITS model was specified as: Where Y denotes the outcome variable; β₀ is the intercept; β₁ represents the pre-intervention slope (baseline trend); Time is a continuous variable coded sequentially from 1 to 11; β₂ captures the immediate level change following policy implementation; Intervention is coded 0 before the intervention and 1 thereafter;β₃ represents the change in slope after implementation, with β₁+β₃ denoting the post-intervention slope; Posttime is coded 0 for all pre-intervention years and for the intervention year and coded 1–4 thereafter; and ε is the random error term. A two-sided significance threshold of α = 0.05 was used for all regression coefficient tests. To elucidate the causal relationship between the policy intervention and subsequent changes in unsafe sexual behaviour (Objective 2), we implemented an event study design. Abnormal positivity rate (ARt) and cumulative abnormal positivity rate (CARt) were used as primary indicators to quantify the extent to which the intervention affected the target population. ARt was defined as the difference between the observed positivity rate (Rt) and its expected value (ERt), such that ARt = Rt−ERt. The CARt captured the aggregated deviation over time and was calculated as . To estimate the expected trend in active syphilis prevalence during the event window, we fitted an autoregressive (AR) time-series model to forecast future values based on the historical trajectory of the outcome variable [25]. In this context, ASPR serves as a timely marker of population-level behavioural shifts. Model adequacy was evaluated using the Ljung–Box Q test under the hypotheses H₀: residuals follow a white-noise process with no autocorrelation; H₁: residuals exhibit significant autocorrelation. If residuals met the white-noise assumption, the AR model was retained. The general AR(p) specification used to estimate the expected active syphilis positivity rate from 2010 to 2015 was: Where yt denotes the predicted positivity rate at time t, c is the constant term, ϕ1, ϕ2, ⋯, ϕp are AR coefficients, p indicates model order, and 𝜀𝑡 is the white-noise error term. To assess whether deviations in syphilis positivity represented random fluctuations or systematic changes, we conducted a robustness test on CARt. The hypotheses were defined as follows: H₀: CARt = 0, indicating that observed changes reflect random variation; H₁: CARt ≠ 0, indicating non-random deviations attributable to the intervention. Because of the small sample size, normality of CARt was tested using the Shapiro–Wilk test. CARt was treated as the test variable. Parametric t-tests were used when normality was satisfied; otherwise, non-parametric rank-sum tests were used. The significance threshold was set at α = 0.05. CAT, AR modelling, and Ljung–Box diagnostics were conducted in SPSS AU (Beijing Qingsi Technology Co., Ltd., Beijing, China). ITS analysis, event study estimations, and robustness tests were performed in R version 4.5.0 (R Foundation for Statistical Computing, Vienna, Austria). Ethical statement As this study only used the annual AKAR, CUR-LCSE, and ASPR data from FSWs, which were obtained from the Baise CDC’s anonymous monitoring system in accordance with the Infectious Disease Prevention and Control Law, without involving personal privacy or other information, informed consent was waived for this study. Approval was obtained from the Ethics Committee of Youjiang Medical University for Nationalities (Approval No: 2024071201). Results The samples of HIV sentinel surveillance from 2010 to 2020 From 2010 to 2020, a total of 19 220 HIV sentinel surveillance records from FSWs were collected (Fig. 1). Annual sample size overview of FSW surveillance in Baise City, Guangxi, from 2010 to 2020 Annual trend analysis Annual trend analysis from 2010 to 2015 The CAT showed that AKAR increased from 66.12% in 2010 to 94.77% in 2015, and self-reported CUR-LCSE rose from 88.01% to 98.00% over the same period. In contrast, ASPR remained stable throughout the study period (Table 1). Table 1Trends in various study variables among FSWs from 2010 to 2015 [Na (%)]Variable2010(N = 1535)2011(N = 2104)2012(N = 2095)2013(N = 2195)2014(N = 1980)2015(N = 2197)χ2P-valueAKARb1015(66.12)1671(79.42)1904(90.88)2050(93.39)1854(93.64)2082(94.77)772.667< 0.01CUR-LCSEc1351(88.01)1962(93.25)2050(97.85)2161(98.45)1951(98.54)2153(98.00)270.157< 0.01ASPRd107(6.97)192(9.13)76(3.63)159(7.24)98(4.95)176(8.01)0.3870.53aN, the total number of female sex workers included in the surveillancebAKAR, AIDS knowledge awareness ratecCUR−LCSE, condom use rate at last commercial sexual encounterdASPR, active syphilis positivity rate Trends in various study variables among FSWs from 2010 to 2015 [Na (%)] aN, the total number of female sex workers included in the surveillance bAKAR, AIDS knowledge awareness rate cCUR−LCSE, condom use rate at last commercial sexual encounter dASPR, active syphilis positivity rate Annual trend analysis from 2016 to 2020 From 2016 to 2020, AKAR among FSWs increased from 82.15% to 96.45%. Self-reported CUR-LCSE showed a small but statistically significant upward trend, from 98.88% to 99.59%. In contrast, ASPR declined over the same period, from 7.68% to 3.15% (Table 2). Table 2Trends in various study variables among FSWs from 2016 to 2020 [Na (%)]Variable2016(N = 1602)2017(N = 1809)2018(N = 1532)2019(N = 1186)2020(N = 985)χ2P-valueAKARb1316(82.15)1602(88.56)1337(87.27)1121(94.52)950(96.45)151.664< 0.01CUR-LCSEc1584(98.88)1779(98.34)1510(98.56)1181(99.58)981(99.59)7.423< 0.01ASPRd123(7.68)90(4.98)114(7.44)20(1.69)31(3.15)37.290< 0.01a N, the total number of female sex workers included in the surveillanceb AKAR, AIDS knowledge awareness ratec CUR−LCSE, condom use rate at last commercial sexual encounterd ASPR, active syphilis positivity rate Trends in various study variables among FSWs from 2016 to 2020 [Na (%)] a N, the total number of female sex workers included in the surveillance b AKAR, AIDS knowledge awareness rate c CUR−LCSE, condom use rate at last commercial sexual encounter d ASPR, active syphilis positivity rate Interrupted time series analysis and outcome ITS analysis showed that before policy implementation (2010–2015), AKAR increased steadily. For CUR-LCSE, although it rose from 88.01% to 97.85% between 2010 and 2012, it subsequently plateaued at approximately 98% until 2015, yielding a statistically non-significant overall linear trend. ASPR remained stable during this period. After policy implementation (2016–2020), AKAR displayed short-term fluctuations but no significant long-term change. Although ASPR decreased in absolute terms, the change was not statistically significant; similarly, no meaningful change was observed in CUR-LCSE. The ITS model incorporated annual outcome values, a continuous time variable (time), an intervention indicator (intervention), and a post-intervention time variable (posttime) (Table 3). Table 3Construction of interrupted time series model sequences for each variable among FSWs from 2010 to 2020VariableAKARa(%)CUR-LCSEb(%)ASPRc(%)TimeInterventionPosttime201066.1288.016.97100201179.4293.259.13200201290.8897.853.63300201393.3998.457.24400201493.6498.544.95500201594.7798.008.01600201682.1598.887.68710201788.5698.344.98811201887.2798.567.44912201994.5299.581.691013202096.4599.593.151114aAKAR, AIDS knowledge awareness ratebCUR−LCSE, condom use rate at last commercial sexual encountercASPR, active syphilis positivity rate Construction of interrupted time series model sequences for each variable among FSWs from 2010 to 2020 aAKAR, AIDS knowledge awareness rate bCUR−LCSE, condom use rate at last commercial sexual encounter cASPR, active syphilis positivity rate Based on KPSS and Durbin–Watson tests, maximum likelihood estimation was selected to model AKAR, generalised least squares for CUR-LCSE, and ordinary least squares for ASPR. For AKAR, the fitted interrupted linear regression model for 2010–2020 was y = 100.38 + 5.30×time − 22.15×intervention − 1.65×posttime + ε. Before policy implementation (2010–2015), the annual increase in awareness was significant (β₁=5.30; t = 6.66; p < 0.01). After implementation (2016–2020), the slope decreased to 3.65 (β₁+β₃). In 2016, the year the policy was introduced, awareness dropped by 22.15% points relative to the pre-intervention trend (β₂=−22.15; t = − 4.96; p < 0.01) (Table 4; Fig. 2). Table 4Interrupted time series model coefficients for variables related to FSWs from 2010 to 2020VariableEstimateSEa95%CIbtP-valueAKARcβ0100.382.17(95.25,105.51)46.31< 0.01β15.300.80(3.41,7.19)6.66< 0.01β2-22.154.47(-32.72,-11.58)-4.96< 0.01β3-1.651.32(-4.77,1.47)-1.250.25CUR-LCSEdβ069.2616.98(29.11,109.41)4.08< 0.05β12.682.57(-3.40,8.76)1.040.33β21.341.61(-2.47,5.15)0.830.43β30.332.40(-5.35,6.01)0.140.89ASPReβ07.032.01(2.28,11.78)3.50< 0.05β1-0.110.52(-1.34,1.12)-0.210.84β22.302.75(-4.20,8.80)0.840.43β3-1.130.85(-3.14,0.88)-1.320.23a SE, standard errorb CI, confidence intervalc AKAR, AIDS knowledge awareness rated CUR-LCSE, condom use rate at last commercial sexual encountere ASPR, active syphilis positivity rate Interrupted time series model coefficients for variables related to FSWs from 2010 to 2020 a SE, standard error b CI, confidence interval c AKAR, AIDS knowledge awareness rate d CUR-LCSE, condom use rate at last commercial sexual encounter e ASPR, active syphilis positivity rate Segmented regression plot of AKAR before and after policy implementation among FSWs from 2010 to 2020. The hollow circles represent the observed annual awareness rates; the red dashed line indicates the onset of the intervention; the blue and green solid lines represent the fitted regression trends before and after the intervention, respectively For CUR-LCSE, the fitted model was y = 69.26 + 2.68×time + 1.34×intervention + 0.33×posttime + ε. Before policy implementation, the slope was not statistically significant (β₁=2.68; t = 1.04; p = 0.33). Following implementation, the post-intervention slope was 3.01 (β₁+β₃). In 2016, condom use increased by 1.34% points compared with the pre-intervention period, although this change was not significant (β₂=1.34; t = 0.83; p = 0.43) (Table 4; Fig. 3). Segmented regression plot of CUR-LCSE before and after policy implementation among FSWs from 2010 to 2020. The hollow circles represent the observed annual condom use rates; the red dashed line indicates the onset of the intervention; the blue and green solid lines represent the fitted regression trends before and after the intervention, respectively For ASPR, the fitted model was y = 7.03 − 0.11×time + 2.30×intervention − 1.13×posttime + ε. In the pre-intervention period, the slope showed no evidence of change (β₁=−0.11; t = − 0.21; p = 0.84). After implementation, the slope shifted to − 1.24 (β₁+β₃). In 2016, the positivity rate increased by 2.30% points relative to the pre-intervention period, although the change was not significant (β₂=2.30; t = 0.84; p = 0.43) (Table 4; Fig. 4). Segmented regression plot of ASPR before and after policy implementation among FSWs from 2010 to 2020. The hollow circles represent the observed annual positivity rates; the red dashed line indicates the onset of the intervention; the blue and green solid lines represent the fitted regression trends before and after the intervention, respectively Event study method analysis and outcomes The event study analysis indicated that the policy did not have a significant effect on unsafe sexual behaviours among the target population. Using data from 2010 to 2015, a predictive model was developed. Based on the Akaike information criterion and Bayesian information criterion, a first-order AR model (AR [1]) was identified as the optimal model. The fitted model was y(t) = 10.168–0.554×y(t − 1). At this stage, AIC = 28.046 and BIC = 27.421. The Ljung–Box Q test showed no evidence of autocorrelation in the residuals (Q = 2.539; p = 0.77), supporting the assumption of white noise and indicating good model fit and predictive validity. An additional diagnostic plot (ACF, PACF, and residual distribution) shows this in more detail [see Supplementary Materials 3]. Based on the AR(1) model, the expected ASPRs, ARt, and CARt are shown below (Table 5; Fig. 5). Table 5Current and cumulative abnormal syphilis positivity rates among FSWs from 2016 to 2020VariableYearTActive positive rate(Rta%)Expected active positive rate(ERtb%)Abnormal active positive rate(ARtc%)Cumulative abnormal active positive rate(CARtd%)ASPRe201607.685.741.941.94201714.986.99-2.01-0.07201827.446.301.141.07201931.696.68-4.99-3.92202043.156.47-3.32-7.24 aRt, return bERt, expected return cARt, abnormal return dCARt, Cumulative Abnormal Return eASPR, active syphilis positivity rate Current and cumulative abnormal syphilis positivity rates among FSWs from 2016 to 2020 aRt, return bERt, expected return cARt, abnormal return dCARt, Cumulative Abnormal Return eASPR, active syphilis positivity rate Actual and expected trend of active syphilis positivity rate among FSWs from 2010 to 2020 The normality test for the CARt indicated that the data were normally distributed (W = 0.902; p = 0.42). A one-sample t-test showed no significant deviation from zero (t = − 0.956; p = 0.39), and the null hypothesis was not rejected. These findings suggest that variations in the CARt from 2016 to 2020 were consistent with random fluctuation rather than systematic policy effects. Discussion This study found no clear evidence that regulatory policies substantially altered unsafe sexual behaviour among FSWs. We used ASPR from 2010 to 2020 HIV sentinel surveillance data in Baise as an evaluative indicator to assess their effect on unsafe sexual behaviours. ITS analysis showed that the ASPR was consistent following the policy adjustment. Additionally, the ESM confirmed that ASPR fluctuations occurred randomly. Although trend analysis suggested a decline in ASPR between 2016 and 2020, such short-term patterns are not sufficient to reflect true behavioural change. Given the event study findings, the ASPR decline is most likely a random phenomenon arising from substantial disruptions to community activity among the target population following policy adjustment, which may have affected sampling stability during surveillance. These findings are in sharp contrast to much of the existing evidence, likely because previous work has mainly used qualitative or cross-sectional designs. Such studies typically depend on self-reported behavioural indicators, which are prone to subjectivity and social desirability bias, and lack objective serological measures, limiting their validity. To address these limitations, we have combined longitudinal sentinel surveillance data and serological outcomes to provide a more reliable and objective assessment of policy effects. These results highlight the potential value of incorporating improved longitudinal survey and surveillance methods to better capture underlying trend changes. FSWs in the study area have long exhibited a persistent “knowledge–attitude–practice” disconnect, a pattern consistent with findings from other regions both within China and internationally [26, 27]. Prior to policy implementation, HIV-related awareness and self-reported condom use among FSWs in Baise City increased markedly. While HIV-related awareness fluctuated in certain years during the study period, ASPR remained stable. The mechanisms underlying the continued stability of unsafe sexual behaviour within this population remain unclear. From a public health perspective, sustained interventions delivered through the local healthcare system may have helped maintain relatively high levels of prevention knowledge and willingness to use condoms, which may partly explain the stable rate of active syphilis. However, this persistent pattern highlights potential limitations of behaviour-change frameworks that assume rational, autonomous decision-making, as evidence from behavioural economics and communication research indicates that human behaviour is strongly shaped by cognitive biases, heuristics, and social influences rather than purely rational deliberation [28]. External contextual pressures, such as unequal bargaining power, economic incentives, or vulnerability to coercion in certain transactional settings, may also contribute to behavioural choices in specific circumstances. Fluctuations in population size and HIV-related awareness during the study period indirectly suggest that the policy adjustment increased pressures associated with commercial sex work. They also indicate that small social clusters within the FSW community may play a critical role in shaping individual behaviours. Qualitative studies have shown that in environments characterised by strong state regulation and pervasive social stigma toward commercial sex, FSWs tend to form tightly connected and relatively stable subgroups [29]. Evidence from behavioural research in other fields similarly demonstrates that such small groups are vital in maintaining stable behavioural patterns within populations [30]. Therefore, from the public health perspective of reducing unsafe sexual behaviour, future behavioural intervention studies should explore the causes of such behaviour among FSWs, taking into account group interactions. From a social governance perspective, regulatory policies may reduce syphilis transmission through structural mechanisms rather than direct behavioural change. Specifically, strengthened enforcement and regulatory measures increase the economic and social costs of commercial sex, thereby reducing the frequency of transactions and potentially contracting the size of the FSW population [6, 31]. In this context, even if individual-level behaviours remain unchanged, a reduction in the overall volume of commercial sexual activity may lead to fewer unsafe sexual encounters at the population level, thereby lowering opportunities for syphilis transmission. Previous studies indicate that following the implementation of intensified regulatory measures, public health departments recorded a substantial decline in the number of individuals active in community venues [32–34]. However, field-based investigations became markedly more difficult, possibly because of reductions in the number of FSWs or their displacement to more concealed settings, thereby reducing the visibility of this population [35, 36]. Moreover, regulatory policies create greater barriers to contact and raise transaction costs between clients and service providers. Reduced visibility of the target population and increased cost of community-based activities constrain opportunities for group-level interactions, potentially contracting the size of the population participating in these activities. If this population shrinks, then even in the absence of changes in individual transaction frequency or the probability of unsafe sex, the total volume of commercial sexual activity—and consequently the volume of unsafe sexual encounters—may decrease at the aggregate level. However, this is not the only pathway of influence; in an environment with strengthened regulation, risks may also be redistributed among remaining sex workers, potentially leading to increased structural vulnerability. Nevertheless, within a health promotion framework, regulatory policies may still be a potential strategy for reducing the societal and individual burden of HIV. This study integrated CAT and ITS to accommodate analytical needs for both short- and long-term surveillance data. CAT can capture immediate shifts in study variables after policy rollout, while ITS quantifies the long-term impacts of such interventions. These methods allow not only the examination of policy impact trajectories but also the quantification of changes over time. By including ITS analysis within a sociological event study framework, this approach evaluates shifts in pre- and post-intervention trends, applies defined temporal windows to mitigate temporal inertia, and addresses potential inferential bias that may arise when relying solely on short-term trend analyses. Moreover, the integration of natural experiment principles and event study designs commonly used in the social sciences strengthens causal inference and offers new methodological insights for public health policy evaluation and behavioural research. This interdisciplinary approach increases the analytical rigour of policy impact assessments and provides a theoretical foundation for improving collaboration between public health agencies and other administrative and social governance sectors. Finally, we used standardised surveillance data that were systematically and continuously collected from 2010 to 2020, permitting an accurate assessment of long-term trends in key variables within the target population. This study has several limitations. First, this natural experiment lacked an external control group, making its causal inference less robust than that of a randomised controlled trial; thus, the findings should be interpreted cautiously. Moreover, the analysis was limited to surveillance data from 2010 to 2020, restricting assessment of long-term policy sustainability, and unmeasured time-varying confounders may have influenced the results. Future studies should extend the surveillance period beyond 2020 and adopt multi-region or control-group designs to strengthen causal inference. Second, behavioural heterogeneity may exist within the FSW population owing to differences in age, work venues, and other factors, yet both the ITS design and the event study approach assume behavioural homogeneity, which may obscure important within-group variations. Future research should apply other methods to explore the differences in policy effects across various demographic and social characteristics, enabling a better understanding of within-group heterogeneity. Third, the sample size was relatively limited, which may have reduced statistical power and widened confidence intervals, thereby weakening the ability to detect true effects. In particular, the robustness test based on CARt was conducted over a short event window, which may further limit inference reliability. Therefore, non-significant findings should be interpreted with caution. Future studies should incorporate longer observation periods or alternative robustness testing strategies to improve inferential reliability. Fourth, ASPR was the only blood-based measure in this study. It served as an indirect reference for changes in overall unsafe sexual behaviour, but it should be interpreted with caution. This limits our ability to capture multidimensional and dynamic changes in behavioural patterns among FSWs. ASPR is influenced not only by individual behaviour but also by other factors, including testing coverage, access to treatment, and syphilis reinfection. Therefore, ASPR should not be considered a direct measure of recent unsafe sexual behaviour at the individual level. Observed declines or fluctuations may partly reflect these background factors rather than true behavioural change. Future studies should include additional STI indicators and comprehensive behavioural data to distinguish policy effects on sexual behaviour from those on disease transmission. Finally, our analysis focused exclusively on female sex workers. Including male sex workers—an equally high-risk yet substantially understudied population—would allow future studies to more fully examine how regulatory policies may differentially affect diverse subgroups. Conclusion Mandatory regulatory policies were not associated with measurable changes in unsafe sexual behaviours among FSWs, nor was there clear evidence that these policies inadvertently increased their sexual risk practices. From a social governance perspective, such policies may reduce the size of the sex worker population and raise the costs of commercial sex, thereby lowering the risk of syphilis transmission to the general public. Embedded in the health promotion framework and complementing existing behavioural public health interventions, regulatory policies delivered by non-health sectors can facilitate the development of a multi-level social prevention and control system. Importantly, as these policies do not drive up unsafe sexual behaviours, they protect both the general population and FSWs by cutting down overall high-risk sexual encounters. As such, regulatory policies constitute a valuable part of syphilis prevention strategies and work alongside behavioural interventions to sustain this multi-level prevention and control system. Supplementary Information Supplementary Material 1. Supplementary Material 2. Supplementary Material 3. Acknowledgements We are very grateful to the teachers of Baise City Center for Disease Control and Prevention for their participation in long-term face-to-face field investigations to collect and summarise the serological surveillance data of female sex workers. Many thanks to Professor Xu Liang for providing innovative ideas to modify the content of this study. We also acknowledge the financial support provided by all listed funding sources, which made this work possible. Abbreviations Human Immunodeficiency Virus Female Sex Worker Acquired Immune Deficiency Syndrome AIDS Knowledge Awareness Rate Condom Use Rate at Last Commercial Sexual Encounter Active Syphilis Positivity Rate Sexually Transmitted Infection Interrupted Time Series Event Study Method Cochran–Armitage test for trend Center(s) for Disease Control and prevention Return Abnormal Return Expected Return Cumulative Abnormal Return Kwiatkowski-Phillips-Schmidt-Shin Test Statistical Package for the Social Science Analysis Online Autoregressive Model Standard Error Confidence Interval Autocorrelation function Partial autocorrelation function Authors' contributions JC collected and summarised the preliminary data, assessed its authenticity, and provided recommendations for revising the research conclusions. LT gathered and organised the preliminary data and reviewed the research findings. WL conceived and designed the study, analysed and implemented innovative methods, authored the article, conducted data analysis, and refined the conclusions. XL conceived and designed the study, participated in writing the article and analysing data, and oversaw the study’s content revisions. YP critically evaluated the statistical methods used in the research. RZ and JL assisted with the review process and formatted the article. RZ and XL collaborated to conduct the literature review and extract relevant key points. All authors have reviewed and approved the final version of the manuscript submitted for publication. Funding This research was supported by Youjiang Medical University for Nationalities Postgraduate Innovation Program (YZCXJH202303) and Baise Science and Technology Program (202435002). The funders of the study had no role in study design, data collection, data analysis, or writing of the report. Data availability The datasets generated and/or analysed during the current study are not publicly available due to ethical and legal reasons but are available from the corresponding author on reasonable request. Declarations Ethics approval and consent to participate This study was conducted in accordance with the Declaration of Helsinki and complied with applicable laws, regulations, and internationally recognised ethical standards. The data were obtained from the China HIV/AIDS Comprehensive Response Information Management System, with original surveillance data collected by the Baise Center for Disease Control and Prevention in Guangxi, China, in accordance with the Law of the People’s Republic of China on the Prevention and Treatment of Infectious Diseases, national technical specifications and relevant laws and regulations on the protection of personal privacy, and were completed by the staff of the corresponding institutions. Data collection procedures, including standardised blood sample collection among FSWs, were conducted under a clear legal and public health mandate. This study represents a secondary analysis of routinely collected surveillance data and was approved by the Ethics Committee of Youjiang Medical University for Nationalities (Approval No: 2024071201). All data were fully anonymised prior to analysis, and no personally identifiable information was accessible to the researchers. Given the retrospective use of de-identified data, the requirement for informed consent was waived by the ethics committee. Consent for publication Not applicable. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Wangjin Liang and Jian Chen contributed equally to this work as co-first author. References Associated Data Supplementary Materials Supplementary Material 1. Supplementary Material 2. Supplementary Material 3. Data Availability Statement The datasets generated and/or analysed during the current study are not publicly available due to ethical and legal reasons but are available from the corresponding author on reasonable request.