In this prospective cohort study involving middle-aged and elderly Chinese adults, we were the first to identify that high CumAIP exposure is a significant risk factor for the progression of prediabetes and is detrimental to the regression of prediabetes. Moreover, compared to unmarried individuals, being married significantly reduces the risk of prediabetes progression associated with high CumAIP exposure.

With the rapid global aging population [32], atherosclerosis is becoming increasingly prevalent [33, 34]. Previous studies have shown that AIP, representing atherosclerosis, is closely associated with prediabetes and diabetes [16,17,18,19,20,21], and further follow-up studies have indicated a nonlinear relationship between AIP and prediabetes and diabetes [17, 22]. Additionally, recent research has emphasized the clinical application potential of AIP in glycemic metabolic diseases. Studies have shown that AIP can be directly used for risk assessment of diabetes and prediabetes [17,18,19,20, 22] and for evaluating cardiovascular and metabolic complications in diabetic patients [35,36,37,38]. Despite the substantial evidence highlighting the importance of AIP in adult glycemic metabolic diseases, a limitation of previous studies is that they assessed AIP at a single time point, lacking repeated measurements of AIP. This has hindered a more comprehensive understanding of how changes in AIP affect disease progression. Notably, in a recent cohort study, Yi et al. conducted a bold design to investigate the impact of AIP transition patterns on diabetes [39]. They categorized baseline and follow-up AIP into low and high groups based on specific cutoff values and examined the effects of four transition patterns (maintaining high, maintaining low, high to low, and low to high) on diabetes. The study found that maintaining high AIP, high to low AIP, and low to high AIP transition patterns were positively associated with diabetes occurrence. However, the finding that the high to low AIP transition pattern was identified as a risk factor for diabetes warrants further verification, as it seems counterintuitive. We believe this particular result reported by Yi et al. is primarily related to the cutoff values used for AIP, where minor changes around the cutoff values could significantly affect the AIP transition patterns and further impact the study results. In the current study, we adopted an approach similar to some previous studies [24, 40, 41], combining baseline and repeated measurements of AIP with follow-up duration to calculate the continuous variable CumAIP. Our results showed that high CumAIP exposure was associated with a higher risk of diabetes, providing more direct evidence that monitoring and maintaining appropriate AIP levels is crucial for diabetes prevention.

Most previous studies on the development of prediabetes have focused on the progression to diabetes [42,43,44]. However, it is also important to note that the regression of prediabetes deserves attention as it is closely associated with reduced risks of diabetes and related complications [2, 7,8,9]. The progression and regression rates of prediabetes largely depend on the criteria used, which remains a significant challenge in this field [2]. The ADA criteria are the most inclusive, while the International Expert Committee and WHO criteria are more restrictive [45]. According to ADA criteria, a recent meta-analysis of 103 prospective studies reported that 18% of prediabetic patients progressed to diabetes within five years [6, 8]. Furthermore, another meta-analysis in 2022, based on 35 randomized controlled trials (RCTs) involving 10,164 prediabetic adults, showed that 31% of participants regressed to NFG within 1.6 years [9]. In the current national survey data based on CHARLS, 15.21% of prediabetic patients progressed to diabetes, and 22.12% regressed to NFG during a median follow-up of three years. Regarding diabetes progression, our study’s data are similar to international analysis results. However, the regression rate to NFG in our analysis is slightly lower than in the meta-analysis data [9]. We believe this is due to several reasons: (1) Compared to RCTs [9], our study is observational and did not include interventions specific to the study population. Instead, our findings are more reflective of the real-world situation in China. (2) Our study primarily involved middle-aged and elderly individuals, who are generally older and may have poorer metabolic conditions [1]. (3) Our analysis did not include oral glucose tolerance test data, which could lead to the omission of some patients with impaired glucose tolerance [45]. In this study, we also investigated the association between CumAIP exposure and prediabetes regression. Multivariate Cox regression showed a negative correlation, which further RCS analyses confirmed to be linear. Notably, in the dose-response relationship plots, the RCS analyses with both knots 4 and 5 showed wide confidence intervals for the associations between CumAIP and regression to NFG. Most of the CumAIP-related HRs crossed the reference line both before and after adjustments, regardless of whether the levels of CumAIP were high or low. These findings suggest that the use of CumAIP in assessing regression to NFG involves some uncertainty, and the results are relatively unstable. Regarding the RCS analysis results of CumAIP and regression to NFG, we have the following considerations: (1) Compared with the diagnostic criteria for diabetes, the threshold for diagnosing prediabetes based on blood glucose parameters is relatively loose, which may lead to some subjects who actually have normal blood glucose metabolism to be inappropriately classified as prediabetic patiens; It is necessary to improve and unify the criteria for prediabetes diagnosis as soon as possible [45], and then verify the current research results according to the latest standards. (2) The exclusion of a larger number of subjects with missing blood measurement information in the current study may somewhat contribute to the relative lack of sample size leading to a decrease in test efficacy, and validation of the results in further large sample cohorts is needed.

After establishing the relationship between CumAIP exposure and the progression or reversal of prediabetes in middle-aged and elderly populations, we further investigated the differences in this association among various subgroups. The results of the study showed that no significant specific population dependencies in almost all subgroups, indicating that the current findings are relatively stable, which was further confirmed by sensitivity analyses. However, we did find some notable differences within the marital status subgroup. Specifically, compared to married individuals, those who were unmarried (including separated, divorced, widowed, or never married) had a relatively higher risk of CumAIP-related diabetes. Previous studies have shown that being unmarried or having a poor marital relationship significantly increases atherosclerotic burden [46, 47] and negatively impacts cardiovascular health [48]. Additionally, evidence from the United States and Korea suggests that being unmarried also significantly promotes glucose deterioration and adverse metabolic outcomes [49, 50]. These findings provide context for our results, indicating that being unmarried may influence diabetes progression through atherosclerosis and glucose metabolism. These results suggest that marital status should be considered in diabetes risk assessment, and further research into atherosclerosis-related diabetes outcomes based on marital status is warranted.

The high prevalence and rapid growth of prediabetes worldwide have imposed a significant burden on society [2, 3]. Despite numerous RCTs indicating the potential of pharmacological treatments for prediabetes [9], no drugs have been approved for prediabetes treatment by regulatory agencies. Current evidence and clinical policies favor lifestyle changes for prediabetes management [26, 51], underscoring the importance of effectively implementing diabetes prevention strategies. In the current study, we also investigated the relative impact of physical exercise on the evolution of prediabetes during the follow-up period. Although the final interaction analyses did not detect significant modulatory effects, results from stratified analyses suggest that moderate activity may help reduce the risk of diabetes associated with CumAIP, whereas intensive activity appears to hinder diabetes prevention. Additionally, our study found that any level of physical exercise promotes the regression of CumAIP-related prediabetes. Similar findings have been reported in several previously conducted RCTs [52,53,54,55]. Overall, physical exercise can improve the progression of various chronic diseases and has a beneficial effect on glycaemic control in prediabetic patients [56]. Although conclusive evidence is lacking on the impact of physical exercise on CumAIP-related glycemic metabolism, from a pathophysiological perspective, atherosclerosis results from the interaction between metabolic and inflammatory pathways. Exercise can beneficially modulate these pathways, including lipid, inflammatory, and glucose metabolism [56]. Therefore, findings of the current study can be interpreted as moderate physical exercise exerting a beneficial effect on atherosclerosis, which may, in turn, promote favourable changes in glycaemic metabolism. Of course, the impact of physical exercise is multifaceted, as it influences both atherosclerosis and glycaemic metabolism, creating a positive feedback loop that prevents the progression of various chronic diseases.

The significance of the current study lies in further clarifying the impact of changes in AIP over follow-up periods on the development of prediabetes, building on previous research. High exposure to CumAIP during the follow-up period may reflect an important factor indicating ongoing adverse metabolic conditions. Given the simplicity and convenience of calculating and obtaining CumAIP [16], along with its crucial role in assessing cardiovascular and cerebrovascular diseases risk [24, 40, 57] and its relevance to glycaemic metabolic disorders, CumAIP holds significant potential for clinical applications and prognostic evaluation. China currently bears one of the highest global burdens of atherosclerotic disease, making the use of simple tools to quantify cumulative atherosclerotic exposure particularly significant [58, 59]. We believe that incorporating CumAIP into clinical practice could be instrumental in reducing the burden of atherosclerotic diseases and could inform primary prevention strategies for the regression of prediabetes. It is important to note that calculating CumAIP is a straightforward task for clinicians, with the main challenge being the repeated measurement of AIP. This is due to the relatively low rate of annual physical examination (APE) among middle-aged and elderly populations in China, which ranges from approximately 35–65%, with significant regional and urban-rural disparities [60,61,62]. Despite free APE services being available to older adults in China, a considerable proportion of them do not utilize these services, highlighting the importance of identifying barriers to APE uptake. It is recommended that government agencies establish health consultation centers for middle-aged and elderly individuals in communities and rural areas, enhance health education, and raise awareness of the importance of APE. In addition to the difficulty in obtaining repeated AIP measurements, there are currently no clear recommendations for maintaining appropriate CumAIP levels, as related studies remain limited and further research is needed.

This study has several strengths, including its prospective design and dynamic assessment methods, which elucidate the association between atherosclerosis fluctuations and the development of prediabetes, thus enhancing its clinical significance. Additionally, the design involving repeated measurements of AIP adds a novel aspect to the study.

However, there are several limitations to consider when interpreting our results: (i) The diagnosis of prediabetes in the current analysis did not include oral glucose tolerance test data, potentially missing some patients with impaired glucose tolerance [45], which could lead to an underestimation of both the incidence of diabetes and the regression rate to NFG [45]. (ii) A substantial number of participants were excluded due to missing blood glucose data, which may introduce sampling bias. (iii) Considering the increasing trend of atherosclerosis in younger populations, our findings may also apply to younger adults, adolescents, and children [11, 63], but caution is needed when extrapolating our results based on middle-aged and elderly populations.

(iv) The study participants were Chinese, so caution is required when applying these findings to other racial or ethnic groups. (v) As with all observational studies, despite our efforts to account for relevant confounders, some residual confounding factors may still be present. Nonetheless, sensitivity analyses estimating the minimum E-value needed to account for unmeasured confounders indicated that our findings are relatively robust. (vi) Since this is a non-interventional study, we cannot infer the specific impact of lifestyle interventions on the progression and regression of prediabetes associated with cumulative CumAIP. (vii) Since the Wave 3 survey did not include information on lipid-lowering drug use, the impact of these drugs on the development of CumAIP-associated prediabetes during follow-up could not be assessed in this analysis. In addition, we only obtained information on the use of antihypertensive drugs, antidiabetic drugs, and physical exercise information during follow-up for no more than half of the subjects, which brought certain obstacles to the production of meaningful results in subgroup analysis and also further research is needed. (viii) The first wave of CHARLS national data was applied as the baseline information in the current study. Since prediabetes was diagnosed based on blood glucose measurement parameters and the questionnaire lacked detailed data on the history of prediabetes, the duration of prediabetes in the study population could not be confirmed in the current study for the time being, which is an important help for further interpretation of the results [9], and further research is needed. (ix) Estimates of CumAIP for the current study were calculated using AIP data from waves 1 and 3. Although this cumulative exposure assessment method has been widely used in many studies [24, 40, 41], it should be noted that the calculation of CumAIP includes measurement data at the time of the study outcome, which may influence the results and necessitates further prospective research.