Information on HMPO samples for this study was obtained from the Food Safety Risk Monitoring System of the Guangzhou Center for Disease Control and Prevention. HMPO consumption in Guangzhou is concentrated in peripheral districts. Sampling focused on the months of June to August, a period characterized by high temperatures and humidity. From 2010 to 2022, streets in peripheral districts were designated as sampling units, and street types were stratified based on their distance from the center of each district. Two central streets and one remote street were randomly selected as sampling sites in each peripheral area, resulting in 15 streets selected as sampling sites for HMPO. The distribution of sampling sites during the study period is presented in Figure 1. The sampling process involved two individuals, acting as consumers, procuring HMPO from small workshops to ensure the representativeness of the samples. A total of 590 HMPO samples were included in this study. After collection, the samples were sealed, refrigerated at 4°C, and analyzed within 48 h.

The detection of AFB₁ was conducted using high-performance liquid chromatography (HPLC) in accordance with the methodologies specified in Determination of Aflatoxins B₁, B₂, G₁, and G₂ in Foods (GB/T 5009.23-2006) (32) and Determination of Aflatoxins B and G Groups in Foods (GB 5009.22-2016) (33). The methodology outlined in GB 5009.22-2016 has been described in our previous study (14). The experimental procedures following GB/T 5009.23-2006 are as follows.

The deterministic exposure assessment model was employed to calculate the estimated dietary intake (EDI) of AFB₁ through HMPO. C represents the AFB₁ concentrations detected in HMPO samples; IR represents the daily intake of edible oil in the population, estimated from the consumption data in the Guangzhou Statistical Yearbook (36); BW represents the average body weight. According to the Report on Nutrition and Chronic Diseases of the Chinese Population, the average body weight was 66.2 kg for males and 57.3 kg for females (37). After applying the sex ratio of the Guangzhou population (38), the average body weight was determined to be 62.0 kg. The equation is as follows:

Risk characterization for genotoxic and carcinogenic compounds, such as AFB₁, is based on the margin of exposure (MOE), calculated by dividing 10% of the lower confidence limit of the benchmark dose (BMDL₁₀) of AFB₁ by the estimated dietary intake, as shown in Equation 2. The BMDL₁₀ for AFB₁ is 400 ng/kg bw/d (39). A smaller MOE indicates a greater risk of exposure to genotoxic and carcinogenic compounds. The European Food Safety Authority considers BMDL₁₀ to be derived from animal studies. Due to inherent uncertainties, the MOE ≥ 10,000 indicates a low public health concern and a low priority for risk management (40).

The study setting was Xiaolou Town, Zengcheng District, Guangzhou. Xiaolou is a traditional agricultural town with a population of ~30,000 residents (41). Figure 1 and Supplementary Table S2 demonstrate the presence of numerous HMPO workshops in the area, establishing it as a representative high-HMPO-consumption region in Guangzhou. The Xiaolou Town Health Center is the sole health institution designated to provide the National Basic Public Health Service Program in the town (42) and offers annual health check-ups to residents. The study included participants from 20 local administrative villages, ensuring broad representativeness. Health data for residents from 2010 to 2022 were retrieved from the medical examination system of Xiaolou Town Health Center, and study variables included gender, age and liver enzyme levels. Individuals with a history of hepatitis B were excluded from the study.

AFB₁ exposure is strongly associated with abnormal liver function, as indicated by elevated blood levels of liver enzymes, particularly aspartate aminotransferase (AST) and alanine aminotransferase (ALT) (43–45). These enzymes are abundant in liver cells and are released into the bloodstream upon liver cell damage, serving as bioindicators of abnormal liver function (46, 47). Liver function abnormality was defined according to the Reference Intervals for Common Clinical Biochemistry Tests (WS/T 404) as elevated AST or ALT levels, with AST > 40 U/L or ALT > 50 U/L for males and AST > 35 U/L or ALT > 40 U/L for females (48).

Interrupted Time Series Analysis (ITSA) is a statistical method used to evaluate the effectiveness of interventions through segmented regression models. Small Food Workshops Management Regulations have been in effect since October 1, 2015 (28). Considering the lag in the implementation effects and the annual nature of the data analyzed, the year 2016 was designated as the intervention breakpoint.

The continuous outcome variable was analyzed using a segmented linear regression model. The regression equation is as follows:

In Equation 3, Y_t represents the outcome variable, which in this study denotes the AFB₁ contamination in HMPO. T_t is a count variable for the year. X_t is an indicator variable for the intervention, with a value of 0 before the implementation of SFWMR and 1 after. (T_t−T) represents the difference between the year count variable and the intervention breakpoint. ε_t is the random error term. The values assigned to these variables are detailed in Supplementary Table S3. β₀ is the intercept. β₁ indicates the trend before the intervention. β₂ represents the level change after the intervention. β₃ represents the difference in the slope between pre- and post-intervention periods. (β₁+β₃) reflects the trend after the intervention. We fitted both an unadjusted model and a model adjusted for potential confounders, including average annual temperature, rainfall, sunlight, and peanut yield.

Considering that the prevalence of liver function abnormality in this study was >10%, the binary outcome variable was analyzed using the log-binomial regression model (49). The regression equation is as follows:

In Equation 4, Y_t represents whether the resident has liver function abnormality. The prevalence ratio (PR) was employed to quantify the intervention effect in the segmented log-binomial regression model. The PR was calculated by comparing the post-intervention fitted values with the counterfactual outcomes. Specifically, the prevalence ratio PR_t, at each time point during the intervention period was calculated, and the geometric mean of PR_t was used to obtain the overall PR for the intervention period (50). Furthermore, stratified analyses were conducted by gender and age ≥ 60 years, with age stratification following the definition of older adults as stated by the World Health Organization (51). The Durbin-Watson method was used to detect first-order autocorrelation in the ITSA model. In the presence of autocorrelation, the Newey-West method was applied to adjust the standard errors of the parameter estimates.

Mediation analysis was conducted to investigate the role of AFB₁ contamination in HMPO as a mediator of the relationship between the implementation of SFWMR and liver function abnormality in residents. The formula is as follows:

Where Y represents whether the resident has liver function abnormality. X indicates whether the resident was exposed to the implementation of SFWMR. M represents the annual AFB₁ contamination in HMPO. C refers to potential confounding factors, including gender and age. The direct effect (DE) and indirect effect (IE) were described using the prevalence ratio, with the calculation of PR detailed in Equation 9. Mediation effects were evaluated using 1,000 bootstrap samples.

Microsoft Excel 2021 was employed for data organization and cleaning. Categorical variables were summarized as frequency and percentage. Continuous variables with a normal distribution were summarized using mean and standard deviation, while those with a skewed distribution were reported as median and interquartile range. All statistical analyses were conducted in R (version 4.1.3). P < 0.05 (two-tailed) was considered statistically significant.

The AFB₁ contamination in HMPO were 1.29 (0.12, 6.58) μg/kg. Additionally, 67 samples exceeded the AFB₁ standard, resulting in an exceedance rate of 11.36%. The variation in AFB₁ contamination in HMPO between years was statistically significant (P = 0.019). After 2016, the AFB₁ contamination in HMPO decreased from 2.72 (0.49, 7.47) μg/kg to 0.98 (0.12, 6.30) μg/kg (P = 0.011) (Figure 2; Supplementary Table S4).

As shown in Figure 3, the average daily intake of AFB₁ through HMPO among Guangzhou residents from 2010 to 2022 ranged from 0.25 to 1.68 ng/kg bw/d. The MOE ranged from 238 to 1,600, which was lower than 10,000.

The study included 21,828 residents, comprising 11,558 females (52.95%) and 10,270 males (47.05%). A total of 15,501 (71.01%) participants were aged ≥ 60 years and 6,327 (28.99%) were aged < 60 years. The mean age of the population was 64.34 ± 13.43 years. Of these residents, 2,289 (10.49%) were identified with liver function abnormality (Supplementary Table S5). Prior to 2016, both females, and individuals aged ≥ 60 years exhibited a higher prevalence of liver function abnormality, with an increasing trend. After 2016, the prevalence of liver function abnormality decreased and stabilized across all groups (Figure 4).

According to the adjusted model, prior to the implementation of SFWMR, AFB₁ contamination in HMPO exhibited an increasing annual trend of 0.634 μg/kg (P = 0.013). In the first year following the implementation of SFWMR, AFB₁ levels in HMPO decreased by 2.865 μg/kg (P = 0.006). After the implementation, there was a continued annual decrease in the AFB₁ levels of 2.593 μg/kg (P = 0.034). Compared to the counterfactual scenario, the trend decreased by 3.227 μg/kg per year (P = 0.024) (Table 1; Figure 5).

In the entire population, the implementation of SFWMR was associated with a reduction in the prevalence of liver function abnormality compared to the counterfactual outcome (PR = 0.650, 95% CI: 0.469–0.902). When stratified by gender and age, significant associations were observed in both the female group (PR = 0.484, 95% CI: 0.310–0.755) and the age ≥ 60 years group (PR = 0.586, 95% CI: 0.395–0.868). However, no significant associations were found in the male group or among participants aged < 60 years (Figures 6, 7).

As shown in Table 2, the implementation of SFWMR had a total effect on liver function abnormality (PR = 0.970, 95% CI: 0.959–0.981). The analysis revealed an indirect effect of AFB₁ contamination in HMPO on the relationship between the implementation of SFWMR and the liver function abnormality (PR = 0.981, 95% CI: 0.969–0.993). AFB₁ contamination in HMPO fully mediated the relationship between the implementation of SFWMR and liver function abnormality.

Aflatoxin contaminants in food are widely recognized as a significant health concern (52). In this study, AFB₁ was detected in HMPO samples from Guangzhou from 2010 to 2022 by HPLC, with an exceedance rate of 11.36%. This rate is comparable to the national average of 11.11% (53) and lower than rates reported in other areas of Guangdong Province, including Huizhou (19.72%) (54), Heyuan (11.82%) (55), and the western region of Guangdong (18.7%) (15). This variability in regulatory enforcement could partly explain the disparities in exceedance rates between Guangzhou and other cities in Guangdong Province. While the SFWMR specifies the requirements for obtaining a food workshop registration certificate, the specific application processes, inspection criteria, and acceptance scopes are tailored by each prefecture-level city according to their local circumstances (56).

However, the exceedance rate in Guangzhou is higher than the rate reported in Tianjin (6.67%) (57). This discrepancy may be attributed to regional differences in climate and soil characteristics. Guangzhou's warm and humid subtropical climate provides optimal conditions for the growth of Aspergillus molds that produce toxins (58). Moreover, the predominantly clay-based soil in Guangzhou contrasts with the sandy soil prevalent in Tianjin (59). The high water retention capacity of clay facilitates the growth of Aspergillus species (60), thereby increasing the risk of peanut contamination before harvesting. Contaminated peanuts introduce AFB₁ into the product during the pressing process (61).

Dietary intake is the main route of AFB₁ exposure. Our study observed that the average daily intake of AFB₁ through HMPO ranged from 0.25 to 1.68 ng/kg bw/d for Guangzhou residents. The MOEs were all below 10,000, consistent with the findings of HE, indicating that the health risks associated with HMPO warrant attention (25). A deterministic exposure assessment was performed based on the general population in our study. Notably, age was found to be an important factor influencing the EDI of AFB₁, with children experiencing the highest exposure risk (14, 62). Given their higher exposure levels and the potential for synergistic hepatotoxic effects from chronic AFB₁ intake, coupled with their relatively immature immune systems, children are particularly vulnerable (63).

Based on the ITSA, we evaluated the time-variant impact of SFWMR on AFB₁ contamination in HMPO. In the counterfactual scenario, AFB₁ contamination in HMPO would have increased yearly. Aflatoxin contamination in food is predominantly due to improper handling of raw materials (64). A study in Kenya showed that the source of peanuts and the presence of defective nuts were the primary contributors to increased aflatoxin contamination in the cottage industry (65). Small workshops often purchase large quantities of peanut raw materials at one time. Their storage facilities lack temperature control and dehumidification systems, rendering peanuts highly susceptible to mold contamination. Additionally, these workshops often fail to effectively remove moldy peanuts and do not perform oil refining treatments (55). A research demonstrated that the refining process can significantly reduce AFB₁ content in peanut oil (66). Thus, incorporating the refining process into the standard production of HMPO is recommended.

After the implementation of SFWMR, there was a significant reduction in AFB₁ contamination in HMPO, and this downward trend persisted in subsequent years. This decrease can be attributed to the impact of SFWMR. For instance, article 14 of the regulations explicitly prohibits small food workshops from producing edible oils that do not meet food safety standards and defines penalties for non-compliance. Secondly, the regulations standardize the registration process of small food workshops by requiring on-site inspection of storage facilities, production environments, layout of functional areas, refining equipment, and product labels. Workshops must also provide qualified AFB₁ inspection reports for HMPO issued by the third-party testing institution during the inspection. Furthermore, the regulations establish the production license publicity system, which promotes the healthy development of small food workshops by leveraging market forces through consumer choice (28). Our study emphasized the positive impact of these comprehensive intervention measures in standardizing HMPO production in small food workshops. In Africa, small-scale operators and unorganized markets present key challenges to effective mycotoxin regulation (67). The practices implemented in Guangzhou may serve as a reference for these regions.

Aflatoxin contamination in HMPO in Guangdong urgently requires urgent attention, highlighting the need for enhanced public health management of consumers (25). The study evaluated the potential health impact of SFWMR in a region characterized by high HMPO consumption, focusing on liver function. Our findings indicated that the implementation of SFWMR was associated with a decrease in the prevalence of liver function abnormality. Moreover, mediation analysis revealed that the protective effect of SFWMR implementation on liver function operates through the reduction of AFB₁ contamination in HMPO. From the disease prevention perspective, the implementation of SFWMR reduced AFB₁ exposure and played the primary preventive role in improving liver function in the population. Evidence from animal experiments and epidemiologic studies demonstrates an association between exposure to AFB₁ and abnormal liver function. In rats, chronic exposure to AFB₁ resulted in hepatocellular damage and significantly elevated serum levels of liver enzymes and oxidative stress markers (46). Population-based studies in Saudi Arabia and Ghana similarly reported a significant association between AFB₁ exposure and liver function enzyme levels (45, 68). In addition, AFB₁ exposure acts synergistically with hepatitis B virus infection to cause liver dysfunction (25). Mechanistically, the hepatotoxicity of AFB₁ arises from its active metabolite, AFB₁-8,9-epoxide, which induces mitochondrial dysfunction, oxidative stress, and inflammatory responses, ultimately leading to apoptosis and necrosis of hepatocytes (17).

Additionally, subgroup analysis revealed that the implementation of SFWMR resulted in significant improvements in liver function for females and individuals aged ≥ 60 years. This improvement may be attributed to consumer habits. HMPO, which has a lower oil yield, is infrequently used by commercial food service establishments, but is preferred by females and older adults for home cooking. Males and younger individuals, particularly those working outdoors, tend to consume less HMPO. This preference aligns with Mills' findings, which indicate that females and older individuals are more likely to consume homemade meals (69). Consequently, after the implementation of SFWMR, AFB₁ intake among females and older adults significantly decreased, resulting in a more pronounced protective effect on liver function. AFB₁ exposure is not only related to food contamination but also to consumption. This highlights the need for greater attention to high-HMPO-consumption groups.

Small food workshops face supervision challenges due to their low entry thresholds, broad distribution, and mobility. Despite the enactment of the National Food Safety Standard for Maximum Levels of Mycotoxins in Food in China, supervising small food workshops remains problematic (15). These small workshops are crucial for supporting local agriculture and sideline products, addressing employment issues, and increasing residents' incomes. They also contribute significantly to preserving regional characteristics and dietary culture. The implementation of SFWMR is not a restriction on entrepreneurial freedom. On the contrary, it provides a framework for standardizing productions and encourages small food workshops to upgrade and renovate their facilities to obtain food production licenses. The implementation of SFWMR address legislative and regulatory gaps for small food workshops, enhancing food safety through legal thinking and methods (70).

The study analyzed the implementation effects of SFWMR from the perspective of food contamination and the health impacts on consumers. However, our study had several limitations. Firstly, although ALT and AST are crucial for liver function assessment and hepatocellular damage monitoring, their elevation may also indicate non-hepatic tissue damage, potentially leading to false positives (71). To enhance the specificity of liver disease diagnosis, other biomarkers and imaging tests can be used in tandem. Gamma-glutamyltransferase (GGT) and Alkaline phosphatase (ALP) exhibit distinct expression patterns compared to ALT and AST in liver disease (44). GGT, concentrated in the liver and biliary system, often rises in cholestatic diseases and liver injury. ALP, present in the hepatobiliary system, bones, placenta, and intestines, may indicate bone or liver disease (47). Additionally, imaging tests such as ultrasound, Computed Tomography scans, and Magnetic Resonance Imaging are important tools for assessing liver function. They provide visual information about the structure and morphology of the liver and help identify the presence and progression of liver diseases. The combined use of these biomarkers and imaging tests can help differentiate liver from non-liver diseases and improve diagnostic accuracy (72). Secondly, the liver function data were obtained from the health examination system, which may introduce selection bias. Finally, although the single-group ITSA method used in this study provides reliable inferences, the ecological study design limits the determination of causality. Future follow-up studies will address these limitations by incorporating additional biochemical and imaging tests, increasing the sample size and establishing a control group to more comprehensively evaluate the effectiveness of SFWMR implementation.