All data for this study were sourced from a multicenter health examination cohort of the Rich Healthcare Group in China. The multicenter retrospective cohort study, composed of a general examination population, aimed to explore risk factors influencing the onset of diabetes. The available dataset was organized and uploaded to the Dryad database by Professor Chen and his team for public sharing (31). According to the Dryad Data Sharing Terms and the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License, researchers are permitted to use the datasets publicly shared in the Dryad database for secondary creation, following the proper citation of the data source (31).

The Rich Healthcare Group multicenter health examination cohort was established by Professor Chen and his team. It includes adults (685,277) who had at least two health examinations at Rich Healthcare Group facilities across 32 areas in 11 Chinese cities from 2010 to 2016. Initially, Chen and colleagues aimed to explore the impact of the obesity measurement parameter, body mass index (BMI), on diabetes onset (32). Based on the research objectives, they excluded subjects with: (i) a clear diabetes diagnosis at baseline or diabetic status could not be determined during subsequent visits (6,630 subjects); (ii) less than a 2-year interval between two visits (324,233 subjects); (iii) a BMI less than 15 kg/m² or higher than 55 kg/m² at baseline (152 subjects); (iv) missing age, gender, and fasting plasma glucose (FPG) data at baseline (135,317 subjects). Ultimately, 211,833 subjects were included in Chen et al.’s study.

The current study, using the shared dataset (31) provided by Chen and colleagues, aimed to further explore the association between the IR surrogate MetS-IR and the progression or regression of prediabetes. Based on the inclusion and exclusion criteria established by Chen et al., we further excluded subjects with: (i) non-prediabetes status at baseline (185,815 subjects); (ii) missing MetS-IR data at baseline (10,594 subjects); (iii) missing FPG measurements during follow-up (3 subjects). Finally, our study included 15,421 subjects for analysis.

As a secondary analysis, the current study’s protocol was submitted to and approved by the Ethics Committee of Jiangxi Provincial People’s Hospital. Considering that this research is a secondary creation based on completed studies and the dataset is anonymized, the Committee waived the requirement for informed consent from each study participant.

During each visit to the examination centers, each participant, assisted by medical staff, recorded their diabetes history, family history of diabetes, personal lifestyle information (drinking/smoking status), demographic information (gender and age), and basic physical measurement parameters (blood pressure, height, and weight). Blood pressure was measured using a sphygmomanometer, recording systolic blood pressure (SBP) and diastolic blood pressure (DBP). Height and weight were measured without shoes and in light clothing, recorded to 0.001m and 0.1kg precision, respectively, and used to calculate each individual’s BMI.

Biochemical blood measurements were conducted on venous blood drawn after at least 10 hours of fasting. In standard laboratories, professional technicians measured total cholesterol (TC), TG, HDL-C, low-density lipoprotein cholesterol (LDL-C), FPG, creatinine (Cr), blood urea nitrogen (BUN), alanine aminotransferase (ALT), and aspartate aminotransferase (AST) using an automatic analyzer (Beckman 5800).

Calculation of MetS-IR: Ln  [ ( 2   ×   FPG  ( mg / dL ) )   +   TG  ( mg / dL ) ]   ×   BMI   ( kg / m 2 ) / ( Ln  [ HDL-C   ( mg / dL ) ] ) (27).

The study outcomes were primarily categorized into three groups: regression of prediabetes, maintenance of prediabetes status, and progression to diabetes. Based on the American Diabetes Association’s criteria for prediabetes/diabetes using FPG, we defined the following study outcomes (33): (i) regression of prediabetes defined as an FPG measurement less than 5.6 mmol/L during follow-up; (ii) maintenance of the prediabetic state defined as FPG levels between 5.6 and 6.9 mmol/L during follow-up. (iii) progression to diabetes defined as an FPG measurement above 7.0 mmol/L during follow-up or self-reported diagnosis of diabetes by other healthcare professionals.

Participants were categorized according to the study outcomes and their baseline characteristics were compared using univariate analysis of variance/Chi-square test/Kruskal-Wallis H test. Baseline information results were presented as median (interquartile range)/mean (standard deviation)/frequency (%).

For the three-category study outcomes, we considered participants who continued to maintain prediabetes status during follow-up as the non-event group and used a one-to-one method to split the data into binary datasets with prediabetes regression or progression as outcomes (34, 35). The association between MetS-IR and the progression or regression of prediabetes was analyzed using Cox regression models. Prior to model establishment, we assessed the collinearity of MetS-IR with other covariates using prediabetes progression or regression as outcome variables ( Supplementary Tables 1 , 2 ), as well as the Schoenfeld residuals plots of MetS-IR over time ( Supplementary Figures 1 , 2 ) (36, 37). After confirming that the Cox models met the proportional hazards assumption and identifying weight, BMI, and TC as collinear variables, we constructed four differently adjusted Cox regression models according to the STROBE statement: The unadjusted model was first constructed, analyzing the association between MetS-IR and the progression or regression of prediabetes without adjusting for any variables; then, Model I adjusted for demographic and lifestyle factors (age, gender, height, family history of diabetes, smoking status, and drinking status); Model II further considered blood pressure, blood glucose, and blood lipid-related factors (SBP, DBP, FPG, TG, HDL-C, and LDL-C) based on Model I; Model III, the final model, additionally considered liver and kidney function impacts (ALT, BUN, Cr). On the basis of adjusting covariates in Model III, we also plotted dose-response curves of MetS-IR with prediabetes regression or progression using 4-knot restricted cubic splines (RCS) regression model. When nonlinear associations were detected, potential inflection point values were calculated using recursive algorithms, and segmented Cox regression was used to assess HR values for prediabetes progression or regression before and after the inflection point.

Additionally, several sensitivity analyses were conducted: (i) To further verify the validity of the association between MetS-IR and prediabetes progression or regression, we treated the continuous variable MetS-IR as a categorical variable based on its quintiles and conducted similar analyses. (ii) According to the World Health Organization (WHO) standards for prediabetes/diabetes based on FPG (38), we analyzed the association between MetS-IR and prediabetes progression or regression in a new study population. (iii) We further validated the aforementioned associations by applying models with diabetes progression and return to normal fasting glucose (NFG) as competing risks. (iv) Considering the indirect influence of family history of diabetes (39), we selected a study population without a family history of diabetes and repeated the above analyses. (v) To verify the appropriateness of the dose-response curves between MetS-IR and prediabetes progression or regression, we continued using 3- and 5-knot additional analyses to test the stability of RCS results.

We also conducted stratification to further explore whether the association between MetS-IR and prediabetes progression or regression existed in potential specific populations, based on age (according to WHO age division standards) (40), gender, and BMI (according to the Chinese Obesity Working Group BMI division standards) (41), and used likelihood ratio tests to detect whether this association was significantly different between stratifications.

All analyses were performed using R language version 4.2.1 and Empower(R) version 2.20 statistical software. A two-sided p-value of less than 0.05 was considered statistically significant.

Figure 1 illustrates the selection process of the study population. A total of 15,421 eligible prediabetic subjects were included in this study, with an average age of 50 years and a male-to-female ratio of 1.85:1. During the average follow-up period of 2.96 years, 6,481 individuals (42.03%) returned to NFG, 2,424 (15.72%) progressed to diabetes, and 6,481 (42.03%) continued to maintain their prediabetic status. The cumulative incidence function calculated the cumulative risk of prediabetes progression or regression, showing that the cumulative incidence rates of prediabetes progression or regression at the third, fourth, fifth, and sixth years were 9.86%/30.11%, 17.46%/48.22%, 24.86%/62.51%, and 29.02%/70.72%, respectively. Figure 2 displays the cumulative incidence curves of prediabetes progression or regression, indicating a higher adjustability for regression to NFG compared to progression to diabetes.

We reviewed study characteristics of participants at baseline according to study outcomes at follow-up ( Table 1 ). Reports indicated that female participants and those who quit smoking and drinking were more likely to return to NFG in the future. Additionally, subjects who returned to NFG typically had lower age, weight, BMI, SBP, DBP, FPG, TC, TG, LDL-C, ALT, AST, BUN, Cr levels at baseline. In addition, it is worth noting that among the subjects, those who reverted to NFG and those who later progressed to diabetes had the lowest and highest baseline MetS-IR, respectively ( Figure 3 ); This result suggested that high MetS-IR may be a factor promoting the occurrence of diabetes, while low MetS-IR may be an important factor in promoting NFG recovery.

The estimated association between MetS-IR (as a continuous variable) and the progression or regression of prediabetes from Cox regression models was displayed in Table 2 . From four models with different adjustment levels, we found a significant negative correlation between continuous MetS-IR and prediabetes regression, and a positive correlation with prediabetes progression. According to the final model (Model III), each unit increase in MetS-IR resulted in a 26% increased risk of diabetes in prediabetic patients [HR: 1.26, 1.19-1.33], while a decrease in MetS-IR was protective for prediabetic patients (increased probability of returning to NFG), quantified as a 10% protective effect for each unit decrease in MetS-IR [HR: 0.90, 1.19-1.33].

Figures 4 and 5 show the dose-response curves of MetS-IR with the progression or regression of prediabetes. After multivariate adjustment based on Model III, we found a nonlinear correlation between MetS-IR and prediabetes regression (P for non-linearity < 0.001), and a linear correlation with prediabetes progression (P for non-linearity = 0.346). It was observed in the dose-response curve ( Figure 5 ) that the recovery rate to NFG was higher when MetS-IR was lower, significantly dropping when MetS-IR reached a certain value (between 35-40) (P for log likelihood ratio test < 0.001). Using recursive algorithms and segmented regression, the inflection point for MetS-IR related to prediabetes regression was determined to be 37.22 ( Table 3 ), with HRs of 0.95 (0.93-0.96) before and 0.98 (0.97-0.99) after the inflection point.

Several sensitivity analyses were conducted, showing consistent results. First, treating MetS-IR as a categorical variable yielded consistent results ( Table 2 ). Further analyses were repeated with changed inclusion criteria (WHO) and excluding participants with a family history of diabetes, showing no significant change in results ( Supplementary Table 3 ). The Fine-Gray method considering competing risks was used, again yielding consistent results ( Supplementary Table 3 ); finally, to verify the appropriateness of the 4-knot RCS in the current study, additional analyses using 3-knot and 5-knot RCS were performed, with no substantial changes in results ( Supplementary Figures 3 – 6 ).

As shown in Table 4 , within subgroups stratified by gender, age, and BMI, we observed a significant effect of the association between MetS-IR and prediabetes progression only in the age subgroup (P interaction < 0.05). In the age subgroup, compared to participants aged ≥45, those aged <45 had a stronger positive correlation between MetS-IR and prediabetes progression (HR: 1.42 vs 1.19, P-interaction < 0.0001).

This study is noteworthy due to the following strengths: (i) Based on the multicenter health examination cohort of Rich Healthcare Group, this study benefits from a large sample size, enabling extensive sensitivity and subgroup analyses, enhancing the robustness of the results. (ii) This analysis is the first to report the association between MetS-IR and the regression of prediabetes, offering valuable insights for prediabetes intervention based on MetS-IR thresholds. Moreover, no previous studies have explored the relationship between MetS-IR and diabetes in a prediabetic population.

However, some inherent limitations must be acknowledged: (i) Participants were primarily from the Chinese population, necessitating further research to ascertain the extent to which these findings are generalizable to other ethnic groups. (ii) Due to the lack of oral glucose tolerance test data and HbA1c measurement data in the original dataset, the diagnosis of outcomes in this study relied solely on FPG, which may lead to some bias in the incidence rates of outcomes; considering the diagnostic shortcomings, current research evidence recommends promotion mainly in patients with impaired fasting glucose. It’s important to note that most prediabetes-related studies diagnose based on FPG (2), as large-scale epidemiological surveys typically involve large sample sizes, with only a small proportion undergoing oral glucose tolerance testing. (iii) This study is a secondary creation based on the multicenter health examination cohort of Rich Healthcare Group, and the covariates included in the regression models were based on available variables in the dataset, inevitably leading to residual confounding from variables not included in the dataset (57). (iv) Although the follow-up period of the current study is substantial (2-7 years), the average follow-up duration remains relatively short; longer follow-up studies are needed to further clarify the impact of MetS-IR on long-term blood glucose changes in prediabetic patients. (v) The current study lacks MetS-IR data during follow-up, necessitating future research to assess the impact of increases/decreases in MetS-IR on the progression/regression of prediabetes. (vi) The current study did not identify people who used antidiabetic drugs for indications other than diabetes, which may have partially affected the results.