Abstract

Background and objective:

While the creatinine-to-body weight ratio (Cre/BW) has emerged as a promising biomarker for muscle mass assessment, its relationship with prediabetes remains unclear. This study aimed to investigate the association between Cre/BW ratio and incident prediabetes in Chinese adults.

Methods:

We conducted a large-scale retrospective cohort study involving 173,476 participants from health check-up programs across 11 Chinese cities. Cox proportional hazards models were employed to evaluate the association between baseline Cre/BW ratio and incident prediabetes. To address potential non-linear relationships, we applied Cox proportional hazards regression with cubic spline functions and smooth curve fitting, using a recursive algorithm to calculate inflection points. Multiple imputation was used for missing data, and comprehensive sensitivity analyses were performed to assess result robustness.

Results:

During a median follow-up of 3.0 years, 18,506 participants (10.67%) developed prediabetes. A lower Cre/BW ratio was associated with an increased risk of prediabetes (adjusted HR = 0.869, 95%CI: 0.806-0.973). Exploratory threshold effect analysis suggested a potential inflection point at 0.96(95% CI 0.90-1.01)μmol/L/kg, below which the association might be stronger (HR = 0.407, 95%CI: 0.328-0.506). The association remained stable in sensitivity analyses excluding participants with smoking history, drinking history, or family history of diabetes. Subgroup analyses revealed more pronounced associations among individuals aged 30–40 years (HR = 0.614, 95%CI: 0.532-0.708), females (HR = 0.726, 95%CI: 0.640-0.824), and those with normal blood pressure (systolic blood pressure <140 mmHg, HR = 0.816, 95%CI: 0.752-0.886).

Conclusion:

A lower Cre/BW ratio is associated with an increased risk of prediabetes in Chinese adults. Exploratory threshold effect analysis suggested a potential inflection point at 0.96(95% CI 0.90-1.01) μmol/L/kg. These findings suggest that the Cre/BW ratio could serve as a simple, cost-effective tool for prediabetes risk stratification in clinical practice.

Background

Diabetes mellitus (DM) has become a major global public health challenge, and its prevalence has increased sharply over recent decades (1, 2). The International Diabetes Federation (IDF) estimated that, among adults aged 20–79 years, diabetes affected 10.5% of the population (537 million people) in 2021 and is projected to rise to 12.2% (783 million) by 2045 (3). In China, the world’s most populous nation, the diabetes burden is particularly severe, with recent national surveys indicating a prevalence of 11.2% for diabetes and an alarming 35.7% for prediabetes among adults (4).

Prediabetes, characterized by impaired fasting glucose (IFG) and/or impaired glucose tolerance (IGT), represents a critical transitional state between normal glucose metabolism and diabetes according to the American Diabetes Association (ADA) criteria (5, 6). Individuals with prediabetes face a 5-10% annual risk of progressing to type 2 diabetes mellitus (T2DM) (7). Therefore, early identification of high-risk individuals with prediabetes and implementation of effective interventions are crucial for diabetes prevention.

Recent research has increasingly focused on the link between muscle mass and metabolic disturbances (8, 9). Skeletal muscle, the body’s largest insulin-responsive tissue, is central to glucose homeostasis (10). Accumulating evidence indicates that reduced muscle mass is associated with higher risks of insulin resistance, impaired glucose tolerance, and T2DM (11, 12). However, traditional methods of assessing muscle mass, such as dual-energy X-ray absorptiometry (DXA) and bioelectrical impedance analysis (BIA), are limited in their application to large-scale population studies due to cost and logistical constraints.

The creatinine-to-body weight ratio (Cre/BW) has recently emerged as a simple and cost-effective surrogate marker for muscle mass (13). Serum creatinine, primarily produced by skeletal muscle, correlates positively with muscle mass (14). The Cre/BW ratio not only reflects muscle mass but also considers the weight factor, potentially offering a more accurate representation of the metabolic role of skeletal muscle (15).

Prior studies have examined associations between the Cre/BW ratio and a range of metabolic conditions. Hashimoto et al. reported that the Cre/BW ratio was inversely related to incident T2DM (13). In addition, another study showed that a lower Cre/BW ratio was linked to a higher risk of non-alcoholic fatty liver disease (16). However, research on the relationship between the Cre/BW ratio and prediabetes remains limited, particularly in large-scale studies among the general Chinese population. Recent longitudinal studies and population-based cohorts over the past few years have increasingly validated the utility of the Cre/BW ratio as a reliable, cost-effective surrogate for skeletal muscle mass in predicting cardiometabolic outcomes, including incident type 2 diabetes and insulin resistance (13, 17).

We hypothesized that a lower Cre/BW ratio might be inversely associated with the risk of prediabetes. This study aimed to investigate the relationship between this potential muscle mass surrogate and prediabetes susceptibility, and to explore its utility as a simple marker for risk stratification. These findings might contribute to future strategies for prediabetes screening and prevention in China and beyond.

MethodsStudy design

This study was designed as a retrospective cohort analysis to evaluate the association between the baseline Cre/BW ratio and the risk of developing prediabetes over follow-up. The baseline Cre/BW ratio was treated as the exposure, and incident prediabetes during follow-up was defined as the outcome (time-to-event: 0 = no prediabetes, 1 = prediabetes).

Data source

The study drew on health examination data from the Rich Healthcare Group database, which is publicly available via the DATADRYAD repository (https://datadryad.org/stash/dataset/doi:10.5061/dryad.ft8750v). This database encompasses standardized health examination information collected between 2010 and 2016. The Dryad repository provides non-commercial access to this dataset for academic researchers and permits its adaptation and the creation of derivative works, provided that the original source and authors are properly acknowledged (18).

Study population

This multi-center study enrolled consecutive participants from 32 sites across 11 Chinese cities (Beijing, Changzhou, Chengdu, Guangzhou, Nanjing, Hefei, Nantong, Shenzhen, Shanghai, Suzhou, and Wuhan). The study data, derived from the Rich Healthcare Group’s electronic medical records database (2010–2016), was anonymized using non-traceable identification codes. The research protocol adhered to the Declaration of Helsinki guidelines and received approval from the Rich Healthcare Group’s clinical research ethics committee. Considering the retrospective study design and the use of de-identified data, the Institutional Review Board waived the need to obtain informed consent (18, 19).

From an initial cohort of 685,277 participants who underwent multiple health examinations, we implemented a stepwise exclusion procedure. Participants were excluded according to the following criteria: (1) missing or indeterminate data on demographics, anthropometrics, laboratory parameters (fasting plasma glucose (FPG), Cre/BW ratio), or outcome status; (2) implausible or extreme values (e.g., extreme body mass index (BMI) or Cre/BW ratio outliers); (3) insufficient follow-up duration (<2 years); and (4) presence of diabetes or prediabetes at baseline (FPG ≥5.6 mmol/L), or progression to diabetes during follow-up. The final analytical cohort consisted of 173,476 participants. A detailed participant selection process and specific exclusion counts are presented in Figure 1.We have reinforced the justification for the exclusion criteria to clarify it is strictly for defining the incident cohort.

Flowchart of study participants. The systematic participant selection process is illustrated in Figure 1. The initial cohort consisted of 685,277 individuals who underwent at least two health examinations. After applying multiple predefined exclusion criteria, a final cohort of 173,476 eligible participants was established for subsequent analyses.

VariablesAssessment of creatinine-to-body weight ratio and outcome measures

The Cre/BW ratio was calculated as serum creatinine (μmol/L) divided by body weight (kg) and analyzed as a continuous variable at baseline. Analyses were performed for Cre/BW as a continuous variable (per 1-μmol/L/kg increment) and as a categorical variable (quartiles).”. The primary endpoint was incident prediabetes, defined according to the American Diabetes Association’s 2018 criteria as impaired fasting glucose with FPG values ranging from 5.6 to 6.9 mmol/L (20). Participants were monitored for up to 5 years, with follow-up concluding at prediabetes diagnosis, final clinical visit, or December 31, 2016, whichever occurred first.

Covariates

Covariate selection was informed by prior research and clinical expertise (13, 17, 21). The analyzed covariates comprised two categories: (1) Continuous variables: age, anthropometric measures (height), blood pressure parameters (systolic blood pressure (SBP) and diastolic blood pressure (DBP)), metabolic indices (FPG, blood urea nitrogen (BUN)), lipid profile (total cholesterol (TC), triglyceride(TG), high-density lipoprotein cholesterol (HDL-c), low-density lipoprotein cholesterol(LDL-c), and liver function markers (alanine aminotransferase(ALT), aspartate aminotransferase(AST)). (2) Categorical variables: demographic factors (gender), lifestyle characteristics (smoking and drinking status), and hereditary factors (family history of diabetes).

Data collection and measurements

Trained healthcare personnel performed thorough baseline evaluations in accordance with standardized procedures. Anthropometric measurements were obtained with participants in light clothing without footwear. Height was measured to 0.1 cm precision using a stadiometer, while weight was recorded to 0.1 kg using calibrated electronic scales. Body mass index was computed as weight(kg)/height(m²). Blood pressure measurements were performed using standard mercury sphygmomanometers after participants maintained a seated position for 5 minutes.

Lifestyle factors were assessed through structured questionnaires at baseline. Smoking and alcohol consumption status were each categorized into three groups: current, former, or never users. Demographic information and medical history were similarly documented through standardized questionnaires.

Biochemical analyses were conducted on overnight fasting (≥10 hours) blood samples using a Beckman 5800 autoanalyzer (18). The assessed parameters included HDL-c, BUN, TC, AST, TG, Scr, FPG, LDL-c, and ALT.

Missing data processing

The dataset exhibited different levels of missingness across variables. Missing data were minimal for physiological parameters: blood pressure measurements (n=15, <0.01%), total cholesterol (n=2,680, 1.5%), triglycerides (n=2,685, 1.6%), alanine aminotransferase (n=976, 0.6%), and blood urea nitrogen (n=9,504, 5.5%). More substantial missing data were observed for lipid parameters (HDL-c: n=74,455, 42.9%; LDL-c: n=73,970, 42.6%), liver function (AST: n=100,494, 57.9%), and lifestyle factors (smoking and drinking status: both n=124,935, 72.0%). To minimize potential bias and improve data efficiency, we applied multiple imputation by chained equations (MICE) (22). The imputation model incorporated age, sex, height, SBP, DBP, FPG, TC, TG, HDL-c, LDL-c, ALT, AST, BUN, smoking status, drinking status, and family history of diabetes. Missingness was evaluated under the assumption of missing at random (MAR) (23). Although the MAR assumption is strictly unverifiable from the observed data, we included all analysis variables and relevant auxiliary predictors in the imputation model to maximize the plausibility of this assumption.

Statistical analysis

Participants were categorized into quartiles according to the Cre/BW ratio. In descriptive statistics, continuous variables with a normal distribution were reported as mean ± standard deviation, whereas non-normally distributed variables were described using the median (IQR). Categorical data were presented as counts and proportions. Comparisons among Cre/BW quartile groups used one-way ANOVA for normally distributed variables, the Kruskal–Wallis H test for skewed variables, and the χ² test for categorical variables. For time-to-event outcomes, Kaplan–Meier methods were applied, and prediabetes-free survival across Cre/BW quartiles was assessed with the log-rank test.

We assessed potential collinearity among covariates using variance inflation factors (VIF) (24), computed as 1/(1−R2), where R2was obtained from a series of linear regression models in which each variable was alternately treated as the outcome and regressed on all remaining variables. Variables with VIF >5 were removed from the multivariable regression analyses to mitigate collinearity (Supplementary Table 1).

The association between Cre/BW ratio and incident prediabetes was evaluated using Cox proportional hazards models, with results expressed as hazard ratios (HRs) and 95% confidence intervals (CIs). We implemented three sequential models to assess the association: Model I (unadjusted), Model II (adjusted for demographic and clinical factors), and Model III (fully adjusted for all potential confounders). The specific covariates adjusted in each model are detailed in the footnotes of Table 1. Covariate selection was guided by previous literature (13, 17, 21), and collinearity assessment. Total cholesterol was excluded from the multivariate analysis due to demonstrated collinearity (Supplementary Table 1). The proportional hazards assumptions were validated using Schoenfeld residuals and log-minus-log plots.

ExposureModel I(HR,95%CI, P)Model II(HR,95%CI, P)Model III(HR,95%CI, P)Cre/BW ratio (per 1 μmol/L/kg increase)0.721 (0.673, 0.772) <0.00010.602 (0.559, 0.648) <0.00010.869 (0.806, 0.937) 0.0003Cre/BW ratio quartile  Q1Ref.Ref.Ref.  Q20.825 (0.793, 0.858) <0.00010.823 (0.791, 0.857) <0.00010.892 (0.857, 0.929) <0.0001  Q30.827 (0.795, 0.860) <0.00010.802 (0.770, 0.836) <0.00010.911 (0.874, 0.950) <0.0001  Q40.822 (0.789, 0.856) <0.00010.752 (0.720, 0.785) <0.00010.924 (0.884, 0.966) 0.0005P for trend<0.0001<0.00010.0010

Relationship between Cre/BW ratio and the incident prediabetes in different models.

Model I: we did not adjust other covariates.

Model II: we adjust age, gender, height, SBP, DBP, family history of diabetes, smoking and drinking status.

Model III: we adjust age, gender, height, SBP, DBP, FPG, BUN, TG, HDL-c, LDL-c, ALT, AST, family history of diabetes, smoking and drinking status.

HR, Hazard ratios; CI, confidence; Ref, reference; Cre/BW ratio, creatinine to body weight ratio.

To address potential non-linear relationships between Cre/BW ratio and prediabetes risk, we extended our analysis beyond traditional Cox proportional hazards modeling. We implemented cubic spline functions and penalized spline smoothing techniques to characterize the non-linear associations. Upon confirming non-linearity, we employed a recursive algorithm (an iterative grid-search procedure based on maximum likelihood estimation) (25) to identify the inflection point and subsequently fitted two-piecewise Cox proportional hazards models. The optimal model selection was determined through log-likelihood ratio testing (26).

We conducted stratified Cox proportional hazards analyses across key demographic and clinical parameters. Age was categorized into six groups (<30, 30-39, 40-49, 50-59, 60-69, ≥70 years), while clinical parameters were dichotomized using established thresholds (27, 28): SBP (<140, ≥140 mmHg), DBP (<90, ≥90 mmHg), and TG (<1.7, ≥1.7 mmol/L). Each stratum-specific analysis adjusted for all covariates except the stratification variable itself. Interaction effects were assessed via likelihood ratio tests by comparing models that included interaction terms with corresponding models that did not (29, 30). All subgroup analyses and interaction tests were treated as exploratory, and no adjustments for multiple comparisons (e.g., Bonferroni correction) were applied. Therefore, subgroup-specific estimates should be interpreted with caution.

We conducted comprehensive sensitivity analyses to validate our findings. The Cre/BW ratio was analyzed both as a continuous variable and categorically (quartiles) with trend testing to assess result consistency and potential non-linear relationships. Given the established associations between prediabetes risk and family history, smoking, and alcohol consumption (31–34), we performed additional analyses excluding participants with these risk factors. Due to substantial missing data (~70%) for smoking and drinking status, these variables were omitted from the multivariate model to prevent potential adjustment bias. To evaluate the robustness of our findings against unmeasured confounding, we calculated E-values (35). This methodological approach provided further insights into the reliability of our findings.

Statistical analyses were conducted using R (The R Foundation; http://www.R-project.org) and EmpowerStats (X&Y Solutions, Inc; http://www.empowerstats.com). We reported exact two-sided P-values alongside 95% confidence intervals (CIs) to evaluate the strength of the evidence, avoiding the reliance on a rigid dichotomous P-value threshold.

ResultsCharacteristics of participants

The study cohort (N = 173,476) had a mean age of 41.08 ± 12.09 years, with male predominance (53.19%). The mean Cre/BW ratio was 1.10 ± 0.22 μmol/L/kg, with participants stratified into quartiles (Q1-Q4): <0.94, 0.94-1.08, 1.09-1.24, and ≥1.24 μmol/L/kg. During median follow-up (3.00 years), 18,506 participants (10.67%) developed prediabetes. Compared with Q1, participants in Q4 had higher levels of BUN and Scr, and higher proportions of males, current smokers, and current drinkers. Conversely, the Q4 group had lower values for metabolic parameters (BMI, blood pressure, FPG, TG, TC, ALT, and AST) and a lower proportion of individuals with a family history of diabetes (Table 2). It should be noted that due to the exceptionally large sample size of our cohort, statistically significant P-values (<0.001) were observed across almost all baseline variables; however, these statistical differences may not necessarily reflect clinically meaningful variations.

Cre/BW ratio quartileQ1(<0.94)Q2(0.94-1.08)Q3(1.08-1.24)Q4(≥1.24)P-valueParticipants43369433444339043373Age(years)41.2 ± 11.241.0 ± 11.540.9 ± 12.141.2 ± 13.5<0.001Height(cm)166.5 ± 8.6166.6 ± 8.5166.6 ± 8.2166.3 ± 7.9<0.001Weight(kg)70.0 ± 13.465.0 ± 11.662.5 ± 10.658.9 ± 9.3<0.001BMI (kg/m2)25.1 ± 3.423.3 ± 2.922.4 ± 2.721.2 ± 2.6<0.001SBP (mmHg)119.9 ± 16.2117.5 ± 15.7117.0 ± 15.4116.7 ± 15.7<0.001DBP (mmHg)74.8 ± 11.173.5 ± 10.673.1 ± 10.372.6 ± 10.1<0.001FPG (mmol/L)4.8 ± 0.54.8 ± 0.54.7 ± 0.54.7 ± 0.5<0.001TC (mmol/L)4.7 ± 0.94.7 ± 0.94.7 ± 0.94.6 ± 0.9<0.001TG (mmol/L)1.1 (0.8-1.7)1.0 (0.7-1.6)1.0 (0.7-1.5)1.0 (0.7-1.4)<0.001HDL-c (mmol/L)1.4 ± 0.31.4 ± 0.31.4 ± 0.31.4 ± 0.3<0.001LDL-c (mmol/L)2.7 ± 0.72.7 ± 0.72.7 ± 0.72.7 ± 0.7<0.001ALT (U/L)19.0 (13.0-31.0)18.0 (12.5-27.8)17.2 (12.5-25.9)16.5 (12.2-23.3)<0.001AST (U/L)22.6 (17.9-29.1)22.0 (17.6-27.9)21.9 (17.6-27.0)21.7 (17.6-26.7)<0.001BUN (mmol/L)4.4 ± 1.14.5 ± 1.14.7 ± 1.14.9 ± 1.2<0.001Scr (μmol/L)58.3 ± 11.365.9 ± 11.972.1 ± 12.481.5 ± 13.3<0.001Cre/BW ratio0.8 ± 0.11.0 ± 0.01.2 ± 0.01.4 ± 0.1<0.001Gender<0.001  Male16757 (38.6%)21234 (49.0%)24818 (57.2%)29468 (67.9%)  Female26612 (61.4%)22110 (51.0%)18572 (42.8%)13905 (32.1%)Smoking status<0.001  Never smoker36627 (84.5%)35257 (81.3%)34386 (79.2%)33187 (76.5%)  Ever smoker1212 (2.8%)1488 (3.4%)1635 (3.8%)1816 (4.2%)  Current smoker5530 (12.8%)6599 (15.2%)7369 (17.0%)8370 (19.3%)Drinking status<0.001  Never drinker38358 (88.4%)37514 (86.5%)36831 (84.9%)36103 (83.2%)  Ever drinker4436 (10.2%)5162 (11.9%)5805 (13.4%)6409 (14.8%)  Current drinker575 (1.3%)668 (1.5%)754 (1.7%)861 (2.0%)Family history of diabetes<0.001  No42266 (97.5%)42408 (97.8%)42555 (98.1%)42734 (98.5%)  Yes1103 (2.5%)936 (2.2%)835 (1.9%)639 (1.5%)

The Baseline characteristics of participants.

Values are n (%), mean ± SD or medians (IQR).

BMI, body mass index; FPG, fasting plasma glucose; DBP, diastolic blood pressure; TC, total cholesterol; SBP, systolic blood pressure; TG, triglyceride; ALT, alanine aminotransferase; LDL-c, low-density lipid cholesterol; AST, aspartate aminotransferase; HDL-c, high-density lipoprotein cholesterol; BUN, blood urea nitrogen; Scr, serum creatinine; Cre/BW ratio, creatinine to body weight ratio.

Given the exceptionally large sample size of this study (N = 173,476), P-values are highly significant (<0.001) for nearly all comparisons. This phenomenon reflects high statistical power, meaning even clinically negligible differences may reach statistical significance.

The Cre/BW ratio exhibited an approximately normal distribution (range: 0.414-1.810 μmol/L/kg; mean: 1.098 μmol/L/kg; Figure 2). Notably, participants who progressed to prediabetes showed lower Cre/BW ratios compared to those who remained normoglycemic (Figure 3).

Distribution of Cre/BW ratio. Figure 2 demonstrates the statistical distribution of Cre/BW ratio levels among study participants. The ratio exhibited a normal distribution pattern, with values ranging from 0.414 to 1.810 μmol/L/kg, and a mean value of 1.098 μmol/L/kg.

Comparison of Cre/BW ratio between groups with different prediabetes outcomes. Figure 3 illustrates the comparison of Cre/BW ratios between individuals who progressed to prediabetes and those who did not during the study period. The visualization demonstrates that participants who developed prediabetes showed lower Cre/BW ratios compared to those who maintained normal glycemic status.

The incidence of prediabetes

During the median 3.0-year follow-up, 18,506 participants developed prediabetes (10.67%, 95% CI: 10.52-10.81%), yielding an overall incidence rate of 3.40 per 100 person-years. A significant inverse relationship emerged between Cre/BW ratio quartiles and prediabetes risk (P<0.0001 for trend), with incidence rates declining from 3.94 (Q1) to 3.12 (Q4) per 100 person-years. The corresponding cumulative incidence decreased progressively across quartiles: 12.38% (Q1), 10.29% (Q2), 10.30% (Q3), and 9.71% (Q4) (Table 3).

Cre/BW ratioParticipants(n)Diabetes events(n)Incidence rate (95% CI) (%)Cumulative incidence (Per 100 person-year)Total1734761850610.67 (10.52-10.81)3.40Q1(<0.94)43369536812.38 (12.07-12.69)3.94Q2(0.94-1.08)43344445810.29 (10.00-10.57)3.26Q3(1.08-1.24)43390446810.30 (10.01-10.58)3.26Q4(≥1.24)4337342129.71 (9.43-9.99)3.12P for trend <0.001

Incidence rate of prediabetes.

Cre/BW ratio, creatinine to body weight ratio.

Age-stratified analysis revealed consistently higher prediabetes incidence in men compared to women across all decades, with both genders showing age-dependent risk elevation (Figure 4).

Age- and sex-stratified analysis of prediabetes incidence. Figure 4 presents a comprehensive analysis of prediabetes incidence rates across different age groups, stratified by sex. The data demonstrates that men exhibited higher incidence rates of prediabetes than women across all age groups. Moreover, both men and women showed an increase in prediabetes incidence rates with advancing age.

Univariate analysis results based on Cox proportional-hazards regression

Univariate analysis identified multiple factors associated with prediabetes risk (Table 4). Positive associations were observed for anthropometric measures (height, weight, BMI), blood pressure (SBP, DBP), metabolic parameters (FPG, liver enzymes, lipids except HDL-c), renal function markers (BUN, Scr), and lifestyle factors (smoking, alcohol consumption). Conversely, female gender, HDL-c and Cre/BW ratio showed inverse associations. Family history of diabetes demonstrated no significant relationship with prediabetes risk.

VariableStatisticsHR (95%CI)P valueAge(years)41.081 ± 12.0881.033 (1.032, 1.035)<0.00001Gender  Male92277 (53.193%)Ref.  Female81199 (46.807%)0.639 (0.620, 0.658)<0.00001Height(cm)166.478 ± 8.3181.007 (1.005, 1.009)<0.00001