This study protocol has been approved by the Ethics Committee of the Guangdong Provincial Hospital of Chinese Medicine (Ref No. BE2021-247-04) and the Ethics Committee of the Guangzhou 11th People’s Hospital (Ref. No. K2022-03). The study was registered with the Clinical Trial Registry (ClinicalTrials.gov, ID: NCT05315895) on November 14, 2022.

This cross-sectional study was the first survey from a population-based prospective cohort study conducted by the Guangdong Provincial Hospital of Chinese Medicine in collaboration with the Guangzhou 11th People’s Hospital.

From May 1, 2022, to December 31, 2023, a total of 26,472 subject were enrolled at the Health Management Center in Guangzhou 11th People’s Hospital. The study employed a dual-stage screening process encompassing (1) certified physicians and research personnel systematically performed standardized eligibility evaluations using predefined inclusion/exclusion criteria, followed by (2) face-to-face eligibility confirmation through structured clinical assessments, issuing formal written invitations to potential participants. This methodological framework was specifically designed to ensure study integrity and adherence to protocol-specified selection parameters. All participants with diabetes, prediabetes, or normal glycemic levels included in the study were diagnosed by medical professionals. All study subjects signed the informed consent form. The personal information of the participants will be strictly protected by some strict protective measures, including anonymization, authority, and using strong passwords for the data-accessing process.

The diagnosis for the diabetes group followed the ADA criteria (25), including hemoglobin A1C (HbA1C) ≥ 6.5%, fasting plasma glucose (FPG) ≥ 126 mg/dl (7.0 mmol/L), oral glucose tolerance test (OGTT) ≥ 200 mg/dl (11.1mmol/L), or random plasma glucose (PG) ≥ 200 mg/dl. The diagnosis for the prediabetes group was also performed with adherence to ADA guidelines (25) This process involved identifying individuals with FPG levels between 100–125 mg/dl (5.6-6.9 mmol/L), OGTT levels between 140–199 mg/dl (7.8-11.0 mmol/L), or an HbA1C level between 5.7%-6.4%. Conversely, the rest of the people were listed with normal glucose levels.

The eligible subjects met the following criteria: (1) age 18 years old or above; (2) completion of routine physical examination at the Guangzhou 11th People’s Hospital, Guangzhou, China. There were no gender restrictions for the participants. Exclusion criteria comprised: (a) severe cognitive impairment (medical record-documented diagnosis); (b) active psychiatric disorders; (c) communication barriers; (d) critical systemic illnesses requiring intensive care; (e) incomplete EQ-5D-5L assessments. All subjects had informed consent and signed an informed consent form.

Data were collected on-site through a preplanned data collection form (DCF), including electronic or paper questionnaires, an EQ-5D-5L instrument, and a physical examination. The questionnaires in DCF were validated by endocrinology experts and data manager, consisting of 10 items to measure the life behavior of individuals. It covered various aspects, including sociodemographic factors (education levels, marital status, annual income, and family history of diabetes), lifestyle habits (cigarette smoking and alcohol drinking), sleep patterns (bedtime, length of sleep), and hypertension comorbidity. The detailed classification was shown in Table 1 . Face-to-face interviews between the investigator and the subjects collected demographic information. The medical staff at the physical examination center collected the laboratory data.

The EQ-5D-5L instrument was employed as the quality of life assessment tool in this study, administered through self-completion by participants. This validated instrument consists of two components: the EQ-5D descriptive system and the EQ visual analog scale (EQ-VAS). The EQ-VAS provides a quantitative measure of self-rated health status through a 100-mm vertical scale, anchored by the descriptors “best imaginable health state” (100 points) and “worst imaginable health state” (0 points). Meanwhile, the EQ-5D-5L descriptive system evaluates health-related quality of life across five discrete dimensions: mobility, self-care, usual activities, pain/discomfort, and anxiety/depression. Responses across these dimensions were converted to an EQ-5D index value using the standardized Chinese population-based value set (26).

The investigators received comprehensive training to ensure thorough understanding of the study protocol and data collection procedures. Data entry procedures employed a dual approach: paper-based DCF utilized double data entry with discrepancy resolution, while electronic data capture systems incorporated automated validation checks. The consolidated database underwent systematic quality control measures, including comprehensive logic checks and independent verification by an independent data management team. These multi-layered validation processes were implemented to maximize data accuracy and reliability throughout the study.

The Sample sizes were calculated by using Gpower 3.1.2 (http://www.ats.ucla.edu/stat/gpower/) using a ANCOVA with fixed effects, considering main effects and interactions, with a desired power of 90%, at a 5% significance level, and an effect size of 0.0859. The effect size calculation derived from EQ-VAS scores (group means: 78, 80, 82) with pooled standard deviation (SD) of 19 across three experimental groups (17). The calculation indicated that a minimum of 1426 participants were required. Therefore, there were 18605 respondents in this survey, which met the requirements indicated by power analysis. Following propensity score matching (PSM), the matched sample size of 3130 still retain analytical adequacy.

All statistical analyses were performed using SPSS Statistics 18.0 (IBM Corporation, Chicago, IL, USA). Categorical variables were expressed as frequency percentages, while continuous variables were presented as mean ± standard deviation (SD) or median with interquartile range (IQR), based on data distribution characteristics. Missing data were systematically evaluated prior to matching and addressed through multiple imputation techniques for continuous variables and modal substitution for categorical variables. The study population was stratified into three glycemic status categories: normoglycemic (n=16,389, 88.1%), prediabetic (n=1,339, 7.2%), and diabetic (n=875, 4.7%). To address group imbalance while maintaining statistical power and maximizing analytical validity, an optimal propensity score matching (PSM) was constructed using nearest-neighbor matching (1:1:4 ratio for diabetes: prediabetes: normal glycemic group) with a caliper width of 0.01. This matching ratio was selected based on two key considerations: (1) The original sample size disparity necessitated retaining sufficient normal glycemic controls (largest group) to ensure adequate power for subgroup analyses; (2) Pilot matching tests demonstrated that 1:1:4 achieved optimal balance (standardized mean differences <0.1) while preserving 92% of diabetic cases, compared to 67% retention with 1:1:1 matching. The matching algorithm took several categorical variables as covariates, incorporated age, gender, educational attainment, body mass index (BMI), and annual income. Intergroup comparisons employed χ² tests or Fisher’s exact tests for categorical parameters, and one-way ANOVA with post hoc Bonferroni adjustment or Wilcoxon rank-sum tests for continuous measures, as appropriate. Multivariate linear regression models were subsequently developed, incorporating all variables demonstrating statistical significance (P<0.05) in univariate analyses. These models quantified the adjusted associations between glycemic status (independent variable) and HRQoL outcomes (EQ-VAS scores and EQ-Index), while controlling for identified confounders. Stratified analyses were conducted across predetermined subgroups: biological sex, age tertiles, BMI categories, and hypertension comorbidity status.

A total of 26,472 participants were enrolled, and 7,867 subjects not reporting EQ-5D-5L were excluded; then, the overall characteristics of the 18,605 subjects were outlined. Figure 1 depicts the participant selection process through a standardized flowchart. Within the study population, diabetes mellitus prevalence was 4.7% (n=875), demonstrating significant gender disparity (male: 77.7% vs female: 22.3%). Similarly, prediabetes affected 7.2% of participants (n=1,345), with comparable gender distribution patterns (male: 74.1% vs female: 25.9%), as detailed in Supplementary Table S1 ( Supplementary Appendix Table S1 ).

Post-matching comparative analysis demonstrated no statistically significant intergroup differences (P>0.05) in sociodemographic parameters (age, biological sex, BMI, educational attainment, marital status, and annual income) or sleep architecture characteristics (bedtime regularity and sleep duration). However, significant disparities emerged in behavioral and clinical profiles, particularly regarding alcohol consumption patterns, familial diabetes history, and hypertension comorbidity status (P<0.05). Table 1 delineates the demographic and clinical characteristics of the propensity-matched groups stratified by glycemic status (diabetes mellitus, prediabetes, and normal glycemic levels).

Comprehensive analysis of the EQ-5D-5L descriptive system demonstrated comparable dimensional impairment patterns across propensity score-matched glycemic strata (diabetes, prediabetes, and normoglycemia). The “Pain/discomfort” and “Anxiety/depression” dimensions emerged as predominant health concerns, followed sequentially by “Mobility”, “Self-care”, and “Usual activities”. Chi-square testing revealed no statistically significant intergroup differences in dimension distribution patterns, as detailed in Table 2 and visualized in Figure 1 .

Consistent with these findings, composite EQ-Index scores remained not statistically differ across study groups post-matching: prediabetes (0.972 ± 0.047), diabetes mellitus (0.968 ± 0.054), and normal glycemic controls (0.972 ± 0.043) (P>0.05). This dimensional concordance persisted despite observed differences in raw dimension prevalence rates. Conversely, significant intergroup disparities emerged in EQ-VAS measurement. The diabetes group exhibited markedly reduced EQ-VAS scores (79.11 ± 8.48) compared to normal glycemic levels counterparts (81.65 ± 19.64) (P<0.05). These results are visualized in Figure 2 .

Multivariate linear regression models incorporating adjustments for key confounders (family history of diabetes, alcohol consumption patterns, and hypertension comorbidity status) revealed significant associations between glycemic status and HRQoL outcomes ( Table 1 ). Specifically, the fully adjusted model demonstrated a 2.139-point reduction in EQ-VAS scores for the diabetes group compared to normal glycemic group (95% CI: -3.748 to -0.530; P=0.009) ( Table 3 , adjusted model). Conversely, no statistically significant differences in EQ-Index scores persisted after controlling for confounding variables, consistent with the preliminary intergroup comparisons ( Table 4 , adjusted model).

Stratified analyses revealed significant glycemic status-dependent disparities in EQ-VAS distributions across age and sex subgroups. Participants with diabetes and hypertension comorbidity demonstrated significantly reduced EQ-VAS scores (1.766-point deficit, 95% CI: -3.025 to -0.508; P=0.006) compared to normal glycemic levels with similar comorbidity profiles. BMI-stratified analyses showed progressive EQ-VAS reductions in diabetes subgroups: 1.654-point deficit (95% CI: -2.742 to -0.566; P=0.003) in overweight (BMI 24-28) and 2.953-point deficit (95% CI: -4.626 to -1.280; P=0.001) in obese (BMI≥28) categories relative to BMI-matched normal glycemic levels. EQ-Index analyses identified specific vulnerability patterns: geriatric diabetes patients (≥60 years) exhibited 0.013-point decrement (95% CI: -0.022 to -0.003; P=0.001) and female diabetes patients demonstrated 0.018-point reduction (95% CI: -0.030 to -0.006; P=0.003) compared to their respective normal glycemic subgroups ( Figure 3 ).

Stratified analyses of prediabetes subgroups revealed significant EQ-VAS impairments in older adults (≥60 years: 1.628-point deficit, 95% CI: -3.058 to -0.199; P=0.026) compared to age-matched normal glycemic controls. Female prediabetes subjects demonstrated a 1.950-point reduction (95% CI: -3.624 to -0.277; P=0.022) relative to sex-matched normal glycemic levels. Notably, prediabetes status showed no statistically significant association with EQ-Index variations across any analyzed demographic or clinical subgroups ( Figure 3 ).