Nonlinear association of residual cholesterol to high-density lipoprotein cholesterol ratio with diabetes mellitus: a retrospective cohort study

Abstract

Diabetes mellitus is a major public health challenge in the world, and the role of lipid metabolism disorder in its pathogenesis has attracted much attention. The ratio of residual cholesterol to high-density lipoprotein cholesterol (RC/HDL-C), as a comprehensive index of atherosclerosis and anti-atherosclerotic lipid load, has shown predictive value in cardiovascular diseases, but its relationship with diabetes mellitus has not been clear. Therefore, we conducted a retrospective cohort study to investigate the relationship between RC/HDL-C and the risk of diabetes mellitus. This study was based on open data from the Murakami Memorial Hospital Health Screening Cohort in Japan. A total of 15,216 subjects without diabetes mellitus at baseline were included. The association between RC/HDL-C and diabetes mellitus risk was assessed by Cox proportional risk regression modeling. Restricted cubic spline (RCS) and smoothed curve fitting were used to explore the nonlinear association of RC/HDL-C with diabetes mellitus. Sensitivity analyses (excluding people of advanced age, obesity and hypertension) and subgroup analyses were performed to verify the robustness of the results. During median follow-up, 340 (2.23%) new cases of diabetes mellitus developed. After correcting for confounders, each 1-unit increase in RC/HDL-C was associated with a 5.21-fold increase in the risk of diabetes mellitus (HR 5.21, 95% CI 2.59–10.52; P < 0.001). The RCS model revealed a nonlinear association between RC/HDL-C and diabetes mellitus risk with a threshold point of 0.41. When RC/HDL-C ≤ 0.41, RC/HDL-C was significantly and positively associated with the risk of diabetes mellitus (HR 50.6, 95% CI 6.4–403.4; P < 0.001). In contrast, when RC/HDL-C > 0.41, RC/HDL-C was not associated with an increased risk of diabetes mellitus (HR 2.5, 95% CI 1.0–6.5; P = 0.053). Sensitivity and subgroup analyses showed that the positive association between RC/HDL-C and risk of diabetes mellitus was stable and consistent in the general population. This study revealed a nonlinear association between RC/HDL-C and the risk of developing diabetes mellitus through large-scale cohort data. This finding not only provides a new biomarker for early risk prediction of diabetes mellitus, but also deepens the understanding of lipid metabolic imbalance in the pathogenesis of diabetes mellitus. This study provides new ideas for early risk stratification and individualized lipid management in diabetes mellitus.

Introduction

Diabetes mellitus is a major global public health challenge, and its prevalence continues to rise, significantly increasing the risk of complications such as cardiovascular disease, nephropathy and retinopathy¹. Although obesity, insulin resistance, and genetic factors are widely recognized as core risk factors for diabetes mellitus, the role of disorders of lipid metabolism in the development of diabetes mellitus is still of great interest². In recent years, residual cholesterol (RC), i.e., triglyceride-rich lipoprotein remnants (e.g., Very Low-Density Lipoprotein remnants, celiac remnants), has become a hot topic in cardiovascular metabolism because of its pro-atherogenic properties³. Studies have shown that RC not only exacerbates insulin resistance by inducing endothelial dysfunction and chronic inflammation, but may also directly impair pancreatic beta cell function⁴. Meanwhile, high-density lipoprotein cholesterol (HDL-C) is regarded as a metabolically protective indicator due to its anti-inflammatory, antioxidant, and cholesterol reverse transporter functions⁵.

Currently, most studies have focused on the linear association between a single lipid marker and diabetes mellitus, but have neglected the dynamic balance between different lipid fractions and their synergistic effects^6,7. The residual cholesterol to HDL-C ratio (RC/HDL-C) has demonstrated potential in cardiovascular disease risk stratification as a composite of atherogenic and antiatherogenic lipid loads^8,9, but its role in the development of diabetes mellitus has not yet been clarified. In addition, most of the available evidence is based on linear assumptions, and the associations between biomarkers and outcomes in metabolic diseases often show nonlinear or threshold effects (e.g., U- or J-shaped curves), and traditional linear models may underestimate or misclassify the true pattern of associations^10,11. Based on this, the present study was the first to explore the nonlinear association between RC/HDL-C ratio and the risk of diabetes mellitus development using large-scale cohort data, and to reveal the potential threshold effect using the Restricted Cubic Spline (RCS) model. This study provides a novel biomarker for early risk prediction of diabetes mellitus and a scientific basis for the development of individualized lipid management strategies.

Methods

Data sources

Dryad is an internationally recognized open data repository that promotes open access to scientific data and reproducible research, and supports transparency and verifiability of scholarly results¹². We can download the required data from the Dryad Digital Repository website (https://Datadryad.org). According to the terms of service of the Dryad database, users can use the data in the database for secondary analysis without infringing the rights of the authors. In this study, we used the raw data uploaded and shared by Prof. Okamura’s team (https://datadryad.org/stash/dataset/doi:10.5061%2Fdryad.8q0p192)¹³.

Study participants

The initial study investigated common risk factors for the development of diabetes mellitus and fatty liver by analyzing data from the general population who underwent health screening at the Murakami Memorial Hospital Health Screening Center in Japan. Based on this study, the present study utilized this publicly available dataset to further evaluate the predictive value of baseline RC/HDL-C for the risk of developing diabetes mellitus in the future, thereby providing more accurate health advice for diabetes mellitus prevention as well as lipid management. The initial study was approved by the Takashi Murakami Memorial Hospital Ethics Committee and informed consent was obtained from each participant. The present study was a secondary analysis of the data, and the study protocol was approved by the Ethics Committee of Yangzhou University Hospital. The entire study complied with the principles set forth in the Declaration of Helsinki. In addition, this study followed the guidelines of the Statement for the Strengthening of Reporting of Observational Studies in Epidemiology (STROBE).

The initial study recruited 20,944 Japanese individuals who participated in physical examinations between 2004 and 2015 and completed at least 2 examinations. Individuals were excluded if they met any of the following criteria:(1) alcoholic fatty liver; (2) viral hepatitis (hepatitis B antigen and hepatitis C antibody detected at baseline); (3) use of any medication at baseline; (4) diabetes mellitus or fasting blood glucose (FPG) ≥ 6.1 mmol/L at baseline; (5) habitual alcohol consumption (6) covariate deletion. In this study, we further excluded participants with HDL-C deletion and RC/HDL-C outliers (more than the mean ± 3 times standard deviation). Finally, a total of 15,216 subjects were included in this study. The detailed inclusion exclusion criteria are shown in Fig. 1.

Fig. 1.

Flowchart for screening research participants.

Data acquisition

In the initial study, information on participants’ demographic characteristics, lifestyle (exercise habits, smoking and drinking status), past medical history, and medication use was collected by a professional caregiver using a standardized questionnaire. Height, weight, waist circumference (WC) and blood pressure were measured by standardized methods in a quiet environment. Venous blood was collected from all participants after fasting for at least 10 h, and total cholesterol (TC), triglyceride (TG), very low-density lipoprotein (VLDL), HDL-C, FPG, hemoglobin A1c (HbA1c), aspartate aminotransferase (AST), gamma-glutamyltransferase (GGT), and alanine aminotransferase (ALT) were measured on a fully automated biochemical analyzer. Fatty liver is diagnosed by ultrasound by experienced specialists. Diagnostic criteria include: depth attenuation, contrast between liver and kidney echoes, liver brightness and vascular blurring.

Calculation of RC/HDL-C

Non-HDL-C (mg/dL) = TC (mg/dL)—HDL-C (mg/dL)¹⁴.

LDL-C(mg/dL) = 90% Non-HDL-C(mg/dL)—10%TG(mg/dL)¹⁵.

RC = Non-HDL-C(mg/dL)—LDL-C(mg/dL)¹⁶.

Definition of diabetes mellitus

Participants who met HbA1c not less than 6.5% or FPG not less than 7 mmol/l or self-reported during follow-up.

Statistical analysis

This study divided participants into four groups based on RC/HDL-C quartiles. Continuous variables were expressed using mean ± standard deviation (SD) or median. Differences between groups were compared using two-sample t-tests or Kruskal–Wallis H-tests. Categorical variables were expressed using frequencies (n, %). Differences between groups were compared using the chi-square test or Fisher exact test. Cumulative event rates were compared using the Kaplan–Meier method. The log-rank test was used to analyze the risk ratio (HR) for the development of diabetes mellitus between the RC/HDL-C groups.

Multivariate Cox proportional risk regression models were used to investigate the association between RC/HDL-C and diabetes mellitus risk. Covariate selection was based on their known association with diabetes mellitus and clinical relevance. Prior to constructing the multivariate model, we assessed multicollinearity among all continuous covariates by calculating variance inflation factors (VIF). VIF values greater than 5 were considered indicative of significant collinearity. We observed potential high multicollinearity between waist circumference (VIF = 5.2) and other variables. Therefore, to ensure the stability and interpretability of the Cox regression model, only BMI was included as the covariate representing obesity in the primary analysis.We used four different models, including models unadjusted for covariates, Non-adjusted model (no adjust for other covariates), Mini-adjusted mode (adjusted for age and sex), Moderate-adjusted mode (adjusted for age, sex, BMI, fatty liver, and hypertension), and Fully-adjusted mode (adjusted for age, sex, BMI, fatty liver, hypertension, HbA1c, ALT, AST, GGT, physical activity, smoking, and drinking status). Next, we excluded individuals aged ≥ 60 years or with BMI ≥ 25 kg/m² or hypertension for further sensitivity analysis. In addition, we calculated E-values to assess the potential impact of unknown confounders on the association between RC/HDL-C and diabetes mellitus risk.

We used Cox proportional risk regression models combined with cubic spline functions and smooth curve fitting to explore the nonlinear relationship between RC/HDL-C and diabetes mellitus. Segmented Cox proportional risk regression models were constructed using a recursive algorithm to identify inflection points and analyze data on either side of the inflection point. Next, the best explanatory model was determined by a log-likelihood ratio test. Finally, subgroup analyses were performed using stratified Cox proportional risk models for different groups of the population (age, sex, BMI, hypertension, smoking status and drinking status, and fatty liver). Likelihood ratio tests were used to confirm interactions between the different subgroups. P-value ≤ 0.05 were considered statistically significant. All analyses were performed using R software version 3.6.1 (http://www.r-project.org, R Foundation) and Empower Stats (R) (www.Empower Stats.com, X&Y Solutions, Inc., Boston, MA) version 6.0.

Results

Baseline characteristics of participants

A total of 15,216 participants were enrolled in this study, of which 44.09% were male, with a mean age of 43.67 ± 8.89 years. A total of 340 individuals were eventually diagnosed with diabetes mellitus during the follow-up period. The mean ± standard deviation of the RC/HDL-C ratio for all participants was 0.351 ± 0.181. The median baseline RC/HDL-C ratio was 0.30 (interquartile range: 0.22 to 0.43). All participants were categorized into 4 groups based on quartiles of RC/HDL-C (Q1 ≤ 0.22; 0.22 < Q2 ≤ 0.30; 0.30 < Q3 ≤ 0.43; and Q4 > 0.43), which showed that age, BMI, FBG, TG, TC, AST, and ALT increased significantly with increasing RC/HDL-C (all P-value < 0.001). Baseline characteristics of all participants are shown in Table 1.

Table 1.

The baseline characteristics of participants.

RC/HDL	Q1 (0.06–0.22)	Q2 (0.22–0.30)	Q3 (0.30–0.43)	Q4 (0.43–0.90)	P-value
Participants	3804	3804	3804	3804
Age, year	41.03 ± 8.32	43.21 ± 8.68	44.79 ± 9.17	45.68 ± 8.69	< 0.001
Gender					< 0.001
Female	2845 (74.79%)	2148 (56.47%)	1390 (36.54%)	635 (16.69%)
Male	959 (25.21%)	1656 (43.53%)	2414 (63.46%)	3169 (83.31%)
BMI, kg/m2	20.15 ± 2.23	21.30 ± 2.57	22.60 ± 2.90	24.22 ± 3.06	< 0.001
Waist, cm	70.27 ± 6.81	73.77 ± 7.69	78.07 ± 8.12	83.13 ± 7.97	< 0.001
Weight, kg	53.18 ± 8.38	57.38 ± 9.57	62.44 ± 10.58	68.77 ± 11.10	< 0.001
ALT, U/L	15.20 ± 7.73	16.88 ± 8.33	20.10 ± 11.65	26.88 ± 21.24	< 0.001
AST, U/L	16.89 ± 6.71	17.53 ± 6.06	18.34 ± 7.01	20.55 ± 12.49	< 0.001
GGT, U/L	14.98 ± 11.89	16.83 ± 12.64	21.29 ± 19.21	27.22 ± 22.71	< 0.001
HDL-C, mmol/L	1.91 ± 0.35	1.56 ± 0.25	1.34 ± 0.20	1.08 ± 0.18	< 0.001
TC, mmol/L	4.71 ± 0.75	4.96 ± 0.77	5.22 ± 0.79	5.57 ± 0.86	< 0.001
TG, mmol/L	0.46 ± 0.20	0.64 ± 0.25	0.89 ± 0.33	1.50 ± 0.64	< 0.001
LDL, mmol/L	2.47 ± 0.49	2.99 ± 0.50	3.40 ± 0.56	3.89 ± 0.68	< 0.001
RC, mmol/L	0.33 ± 0.06	0.40 ± 0.06	0.48 ± 0.07	0.60 ± 0.10	< 0.001
HbA1c, %	5.13 ± 0.30	5.14 ± 0.31	5.19 ± 0.33	5.22 ± 0.34	< 0.001
FPG, mmol/L	4.99 ± 0.40	5.09 ± 0.40	5.21 ± 0.39	5.34 ± 0.38	< 0.001
SBP, mmHg	108.68 ± 13.26	111.96 ± 14.19	116.18 ± 14.62	120.59 ± 14.88	< 0.001
DBP, mmHg	67.11 ± 9.44	69.77 ± 9.85	72.81 ± 10.08	76.20 ± 10.25	< 0.001
Fatty liver					< 0.001
No	3717 (97.71%)	3563 (93.66%)	3111 (81.78%)	2241 (58.91%)
Yes	87 (2.29%)	241 (6.34%)	693 (18.22%)	1563 (41.09%)
Habit of exercise					< 0.001
No	3081 (80.99%)	3093 (81.31%)	3135 (82.41%)	3235 (85.04%)
Yes	723 (19.01%)	711 (18.69%)	669 (17.59%)	569 (14.96%)
Drinking status					< 0.001
Non	3043 (79.99%)	2980 (78.34%)	2815 (74.00%)	2792 (73.40%)
Light	379 (9.96%)	395 (10.38%)	504 (13.25%)	441 (11.59%)
Moderate	295 (7.75%)	305 (8.02%)	345 (9.07%)	390 (10.25%)
Heavy	87 (2.29%)	124 (3.26%)	140 (3.68%)	181 (4.76%)
Smoking status					< 0.001
Never	2897 (76.16%)	2524 (66.35%)	1999 (52.55%)	1544 (40.59%)
Past	505 (13.28%)	629 (16.54%)	848 (22.29%)	913 (24.00%)
Current	402 (10.57%)	651 (17.11%)	957 (25.16%)	1347 (35.41%)
Hypertension					< 0.001
No	3718 (97.74%)	3642 (95.74%)	3548 (93.27%)	3380 (88.85%)
Yes	86 (2.26%)	162 (4.26%)	256 (6.73%)	424 (11.15%)
Diabetes mellitus					< 0.001
No	3781 (99.40%)	3770 (99.11%)	3721 (97.82%)	3604 (94.74%)
Yes	23 (0.60%)	34 (0.89%)	83 (2.18%)	200 (5.26%)

Incidence of diabetes mellitus

During the follow-up period, the overall prevalence of diabetes mellitus among 15,216 participants was 2.23%, with an average follow-up duration of 6.05 ± 3.78 years. The prevalence rates for the four groups of participants were Q1: 0.75%; Q2: 1.10%; Q3: 2.69%; and Q4: 6.49%, P < 0.001 (Fig. 2). In addition, the cumulative incidence rate in the total population was 2234.49/100,000 person-years. The cumulative incidence rates for different RC/HDL-C subgroups were 745.785/100,000 person-years, 1102.46/100,000 person-years, 2691.31/100,000 person-years, and 6485.08/100,000 person-years, respectively. The prevalence and cumulative incidence of diabetes mellitus were higher in the high RC/HDL-C group compared with the low RC/HDL-C group (P < 0.001) (Table 2).

Fig. 2.

Kaplan–Meier event-free survival curve. Kaplan–Meier analysis of incident diabetes mellitus based on RC/HDL-C quartiles (log-rank, P < 0.0001).

Table 2.

Incidence rate of diabetes mellitus.

RC/HDL	Participants (n)	Diabetes mellitus events (n)	Cumulative incidence (%)	Per 100,000 person-year
Total	15,216	340	2.23	2234.49
Q1	3084	23	0.75	745.78
Q2	3084	34	1.1	1102.46
Q3	3084	83	2.69	2691.31
Q4	3084	200	6.49	6485.08
P for trend		< 0.001	< 0.001	< 0.001

Relationship between RC/HDL-C and diabetes mellitus

We used four different Cox proportional risk regression models to assess the relationship between RC/HDL-C and diabetes mellitus (Table 3). In the unadjusted model, the risk of diabetes mellitus increased more than 50-fold for each 1-unit increase in RC/HDL-C, with an HR of 54.52 (95% CI 31.77–93.58; P < 0.001). In the Mini-adjusted mode djusted for sex and age, the HR was 42.14 (95% CI 23.03–77.09; P < 0.001). Further adjusted for BMI, fatty liver and hypertension, HbA1c, ALT, AST, GGT, physical activity, smoking and drinking status, the association between RC/HDL-C and diabetes mellitus remained statistically significant (HR 5.21, 95% CI 2.59–10.52; P < 0.001). Furthermore, we converted RC/HDL-C to quartiles and the association between RC/HDL-C and diabetes mellitus risk remained significant. In the fully adjusted model, the risk of diabetes mellitus in the highest quartile of RC/HDL-C levels was 148% higher than in the lowest quartile. Trend analysis revealed a significant upward trend in diabetes mellitus risk with increasing RC/HDL-C levels (P < 0.001, Table 3).

Table 3.

Relationship between RC/HDL-C and incident diabetes mellitus in different models.

Exposure	Non-adjusted model (HR.,95% CI, P)	Mini-adjusted mode (HR.,95% CI, P)	Moderate-adjusted mode (HR.,95% CI, P)	Fully-adjusted mode (HR.,95% CI, P)
RC/HDL-C	54.52 (31.77, 93.58) < 0.001	42.14 (23.03, 77.09) < 0.001	7.049 (3.53, 14.05) < 0.001	5.21 (2.59, 10.52) < 0.001
RC/HDL-C (quartile)
Q1	Reference	Reference	Reference	Reference
Q2	1.26 (0.74, 2.13) 0.401	1.20 (0.71, 2.04) 0.501	1.09 (0.64, 1.87) 0.731	1.10 (0.65, 1.87) 0.726
Q3	2.95 (1.86, 4.68) < 0.001	2.65 (1.65, 4.26) < 0.001	1.86 (1.14, 3.01) 0.012	1.82 (1.12, 2.96) 0.015
Q4	6.88 (4.47, 10.59) < 0.001	5.91 (3.73, 9.37) < 0.001	2.71 (1.67, 4.41) < 0.001	2.48 (1.52, 4.05) < 0.001
P for trend	< 0.001	< 0.001	< 0.001	< 0.001

Non-adjusted model: no adjust for other covariates; Mini-adjusted mode: adjusted for age and sex; Moderate-adjusted mode: adjusted for age, sex, BMI, fatty liver, and hypertension; Fully-adjusted mode: adjusted for age, sex, BMI, fatty liver, hypertension, HbA1c, ALT, AST, GGT, physical activity, smoking, and drinking status. Abbreviations: HR, hazard ratios; CI, confidence interval.

Sensitivity analysis

To assess the robustness of the previous results, we performed sensitivity analyses using different methods. First, we analyzed individuals with BMI < 25 kg/m². After adjusting for confounding covariates, a positive association between RC/HDL-C and risk of diabetes mellitus remained, with an HR of 7.28 (95% CI 2.87–18.52; P < 0.001) (Table 4, Model 1). Next, after excluding individuals over 60 years of age, the analysis showed that RC/HDL-C remained positively associated with diabetes mellitus risk (HR 5.07, 95% CI 2.44–10.53) (Table 4, Model 2). Furthermore, among individuals without hypertension, participants with high RC/HDL-C had a significantly higher risk of diabetes mellitus (HR 5.22, 95% CI 2.59–10.52; P < 0.001). Finally, the calculated E-value was 9.89, which shows a higher level of statistical significance, indicating that unrecognized confounders were negligible.

Table 4.

Relationship between RC/HDL-C and diabetes mellitus in different sensitivity analyses.

Exposure	Model 1	Model 2	Model 3
	(HR.,95% CI, P)	(HR.,95% CI, P)	(HR.,95% CI, P)
RC/HDL-C	7.28 (2.87, 18.52) < 0.001	5.07 (2.44, 10.53) < 0.001	5.22 (2.59, 10.52) < 0.001
RC/HDL-C (quartile)
Q1	Reference	Reference	Reference
Q2	1.17 (0.67, 2.03) 0.580	1.13 (0.63, 2.01) 0.686	1.10 (0.62, 1.92) 0.751
Q3	1.83 (1.10, 3.01) 0.020	1.76 (1.02, 3.02) 0.041	1.86 (1.12, 3.10) 0.016
Q4	2.53 (1.52, 4.23) < 0.001	2.56 (1.47, 4.44) < 0.001	2.72 (1.63, 4.55) < 0.001
P for trend	< 0.001	< 0.001	< 0.001

Model 1 was sensitivity analysis in participants with BMI < 25 kg/m². We adjusted age, sex, fatty liver, hypertension, HbA1c, ALT, AST, GGT, physical activity, smoking, and drinking status; Model 2 was sensitivity analysis in participants aged < 60 years. We adjusted sex, BMI, fatty liver, hypertension, HbA1c, ALT, AST, GGT, physical activity, smoking, and drinking status. Model 3 was sensitivity analysis in participants without hypertension. We adjusted age, sex, BMI, fatty liver, HbA1c, ALT, AST, GGT, physical activity, smoking, and drinking status. Abbreviations: HR, hazard ratios; CI, confidence.

Nonlinear association between RC/HDL-C and diabetes mellitus

A nonlinear association between RC/HDL-C and diabetes mellitus was found by combining restricted cubic spline function and smoothed curve-fitting analyses (Table 5). The two-stage Cox proportional risk regression model found an inflection point of 0.41 for RC/HDL-C (log-likelihood ratio test P < 0.001) (Fig. 3). When RC/HDL-C ≤ 0.41, RC/HDL-C was positively associated with the risk of diabetes mellitus (HR 50.6, 95% CI 6.4–403.4; P < 0.001). In contrast, when RC/HDL-C > 0.41, RC/HDL-C was not associated with the risk of developing diabetes mellitus (HR 2.5, 95% CI 1.0–6.5; P = 0.053).

Table 5.

The result of the two-piecewise Cox proportional hazards regression model.

Incident diabetes mellitus	HR (95%CI)	P-value
Fitting model by standard Cox proportional hazards regression	5.2 (2.6, 10.5)	< 0.001
Fitting model by two-piecewise Cox proportional hazards regression
Inflection points of RC/HDL-C	0.41
≤ 0.41	50.6 (6.4, 403.4)	< 0.001
> 0.41	2.5 (1.0, 6.5)	0.053
P for log likelihood ratio test		0.019

We adjusted age, sex, BMI, fatty liver, hypertension, HbA1c, ALT, AST, GGT, physical activity, smoking, and drinking status. Abbreviations: HR, hazard ratios; CI, confidence.

Fig. 3.

The nonlinear relationship between RC/HDL-C and incident diabetes mellitus. Adjusted for age, sex, BMI, fatty liver, hypertension, HbA1c, ALT, AST, GGT, physical activity, smoking, and drinking status. Abbreviations: HR, hazard ratios; CI, confidence interval.

Subgroup analyses and interaction tests

We assessed the stability of the relationship between RC/HDL-C and diabetes mellitus by subgroup analyses and identified potential differences between different populations (Table 6). Notably, the association between RC/HDL-C and diabetes mellitus risk was significantly stronger in the hypertension subgroup (P < 0.05). Furthermore, interaction tests revealed P-values greater than 0.05 in all subgroups except hypertension, indicating that the association between RC/HDL-C and diabetes risk remains stable and consistent across the general population.

Table 6.

Subgroups analysis for the associations between RC/HDL-C and diabetes mellitus.

Subgroup	HR (95%CI)	P-value	P for interaction
Gender			0.2786
Male	20.60 (4.46, 95.11)	< 0.001
Female	3.82 (1.76, 8.30)	< 0.001
Age			0.6233
< 60	5.07 (2.44, 10.53)	< 0.001
≥ 60	8.08 (0.52, 125.85	0.135
BMI			0.1591
< 25	7.28 (2.87, 18.52)	< 0.001
≥ 25	3.58 (1.23, 10.45)	0.019
Hypertension			0.0348
No	5.22 (2.59, 10.52)	< 0.001
Yes	6.44 (3.03, 13.72)	< 0.001
Habit of exercise			0.9551
No	4.99 (2.35, 10.62)	< 0.001
Yes	6.26 (0.93, 42.22)	0.059
Drinking status			0.9307
Non	4.91 (2.10, 11.45)	< 0.001
Light	8.46 (1.07, 66.61)	0.042
Moderate	5.18 (0.48, 55.86)	0.175
Heavy	5.22 (2.59, 10.52)	< 0.001
Smoking status			0.8548
Never	2.65 (0.76, 9.16)	0.125
Past	7.87 (1.76, 35.24)	0.007
Current	8.57 (2.94, 24.94)	< 0.001
Fatty liver			0.0903
No	13.65 (4.68, 39.79)	< 0.001
Yes	3.78 (1.54, 9.27)	0.003

Discussion

In this study involving Japanese adults, we observed a significant positive association between RC/HDL-C and diabetes mellitus risk. Notably, this association remained statistically significant even after adjusting for confounding variables. For each unit increase in RC/HDL-C, the risk of developing diabetes mellitus was more than fourfold. Participants in the group with the highest RC/HDL-C quartile were 148% more likely to develop diabetes mellitus compared with the group with the lowest RC/HDL-C quartile. In addition, RCS analysis revealed a nonlinear relationship between RC/HDL-C and diabetes mellitus with an inflection point of 0.41. Subgroup analyses showed that the correlation between RC/HDL-C and diabetes mellitus risk was stable and consistent in the general population.

In recent years, several cross-sectional and cohort studies have progressively revealed the potential value of RC/HDL-C in metabolic diseases. A cross-sectional study of the National Health and Nutrition Examination Survey (NHANES) database showed that RC/HDL-C was significantly associated with the prevalence of hyperuricemia after excluding confounding factors, with a 98% increase in the prevalence of hyperuricemia for every one-unit increase in RC/HDL-C¹⁷. In addition, higher RC/HDL-C ratios may increase the risk of NAFLD. Li et al. showed that elevated RC/HDL-C was significantly and positively correlated with insulin resistance in the organism, suggesting that it may serve as an early marker of glucose metabolism disruption¹⁸. A prospective cohort study of pregnant Korean women showed that the risk of diabetes mellitus was elevated by 37% for every 1 standard deviation increase in RC/HDL-C (HR 21.78; 95% CI 3.55–33.73; P < 0.001), and that the association was independent of traditional indicators of obesity¹⁹. Additionally, Li et al. found in a cohort study of Chinese adults that RC was positively associated with future diabetes risk (HR 1.13; 95% CI 1.06–1.27; P = 0.025). Sensitivity and stratification analyses indicated that the positive association between RC and diabetes risk remained consistent across different populations²⁰. However, previous studies were mostly limited to linear assumptions, which may overlook the dynamic threshold effect of metabolic markers. The present cohort study provides a critical addition to the existing evidence by revealing for the first time a nonlinear association between RC/HDL-C and diabetes mellitus risk (inflection point 0.41) through restricted cubic spline modeling and confirming that its predictive efficacy remains significant even in a normal BMI population. Together, these results suggest that RC/HDL-C is not only a sensitive indicator of atherosclerosis, but also may be involved in diabetes mellitus pathology through lipotoxicity, chronic inflammation and other mechanisms, and that dynamic monitoring of its ratio may provide a new dimension for metabolic risk stratification²¹.

Dysregulated RC (lipoprotein remnants rich in triglycerides) and HDL-C within the body can disrupt glucose homeostasis through multiple pathways. RC readily accumulate in insulin-sensitive tissues such as muscle, liver, and pancreas. They directly interfere with insulin signaling pathways like IRS-1/PI3K/Akt by generating toxic lipid metabolites such as ceramides, thereby inducing insulin resistance²². Second, RCs activate inflammasomes in endothelial cells and macrophages, promoting the release of multiple inflammatory cytokines (such as IL-1β and TNF-α) and triggering chronic low-grade inflammation, which further exacerbates insulin resistance²³. On the other hand, HDL-C exerts protective effects under normal conditions through its cholesterol reverse transport function and anti-inflammatory/antioxidant properties²⁴. Elevated RC/HDL-C is usually accompanied by metabolic abnormalities such as visceral fat accumulation and hypertriglyceridemia, which create a vicious cycle of insulin resistance^25,26.

RC/HDL-C exhibits a nonlinear positive correlation with diabetes risk, confirming the bidirectional pathological mechanism of lipid metabolism imbalance. When the RC/HDL-C ratio is low (≤ 0.41), the lipotoxicity and proinflammatory effects of RC may be in an ascendant phase, while HDL function remains relatively intact, sufficient to exert its protective compensatory role. At this stage, an elevated RC/HDL-C ratio sensitively reflects increased pathogenic factors alongside a relative insufficiency of protective factors, thereby exhibiting a strong positive correlation with diabetes risk. However, when the ratio exceeds a certain threshold (> 0.41), the body’s compensatory mechanisms may become overwhelmed. On one hand, sustained RC loading may lead to saturation of lipotoxic effects, meaning target organ damage approaches a plateau where further increases in RC yield diminishing marginal effects on metabolic impairment. High RC concentrations may saturate hepatic clearance receptors (e.g., LRP1), plateauing clearance efficiency²⁷. Insulin-sensitive tissues (e.g., skeletal muscle, liver) and pancreatic β-cells may possess upper limits for lipid uptake and storage capacity. When RC load exceeds this threshold, lipid content in these tissues likely approaches maximum carrying capacity. Beyond this threshold, further increases in RC may fail to enhance inflammatory signaling, resulting in a flattened response curve. More critically, under high RC conditions, HDL particles themselves may undergo functional remodeling—transforming from protective particles into dysfunctional or even pathogenic particles. Studies indicate that under conditions of hypertriglyceridemia and inflammation, HDL undergoes proteomic and lipidomic alterations. Its cholesterol reverse transport capacity and anti-inflammatory properties significantly diminish, potentially acquiring pro-inflammatory characteristics²⁸. This nonlinear feature challenges the assumptions of traditional linear models and highlights the complexity of the dynamic balance of metabolic disease biomarkers. Compared with previous studies focusing on a single lipid marker, the RC/HDL-C ratio reflects the dynamic balance between atherogenic and protective lipids^29,30. In addition, the predictive efficacy of RC/HDL-C in diabetes mellitus risk stratification may be superior to that of a single metric because it captures the dual pathologic processes of abnormal triglyceride metabolism (elevated RC) and anti-inflammatory dysfunction (reduced HDL-C).

The identification of threshold effects provides a quantitative basis for individualized intervention. Individuals with an RC/HDL-C ratio below 0.41 may benefit from lifestyle interventions such as the Mediterranean diet and physical activity, or pharmacologic targeted therapies (e.g., a combination of oral omega-3 fatty acid supplements and PPARγ agonists)^31,32. Subgroup analyses suggest that this ratio should be of particular interest in older adults, women, and people with fatty liver disease, which is associated with physiologic characteristics such as reduced lipoprotein lipase activity and fluctuating estrogen levels, which are often present in these populations^33,34. Except for the hypertension subgroup, the P-values for interaction tests in other subgroups were all greater than 0.05, indicating that the positive association between RC/HDL-C and diabetes risk exhibits overall stability and consistency across the entire population. Data from this study demonstrate that the predictive role of RC/HDL-C for diabetes risk is more pronounced in hypertensive individuals. Hypertension and elevated RC/HDL-C ratios may jointly promote diabetes onset and progression through a shared pathway of “synergistically exacerbating vascular endothelial dysfunction.” Furthermore, significant associations remained in sensitivity analyses in the normal BMI population, suggesting that RC/HDL-C could be a valid addition to traditional obesity indicators.

The present study has the following limitations: first, the observational design made it difficult to completely exclude residual confounding, although E-value analyses suggested a limited effect of unmeasured confounding. Second, the heterogeneity of RC components (e.g., VLDL versus celiac residue) was not subdivided, and the pathogenic mechanisms of particles of different sizes may differ. Further, the age structure of the cohort population was relatively young (mean 43.67 years), which may underestimate the long-term risk in older age groups. Finally, the current study focused on the Japanese population, which limits the generalizability of the findings to different ethnic groups. Future studies should validate the association between RC/HDL-C and diabetes risk across different ethnic populations and elucidate the specific pathways through which RC/HDL-C influences β-cell function.

Conclusion

In this study, we revealed for the first time the nonlinear association between RC/HDL-C and the risk of developing diabetes mellitus through large-scale cohort data. This finding not only provides a novel biomarker for early risk prediction of diabetes mellitus, but also deepens the understanding of lipid metabolism imbalance in the pathological mechanism of diabetes mellitus, which is an important insight for the optimization of clinical lipid management strategies.

Author contributions

G.Y contributed to the study concept and design, researched and interpreted the data, and drafted the manuscript. W. S analyzed the data and reviewed the manuscript. All authors read and approved the final manuscript.

Funding

This work was supported by kangda college Research and Development Fund of Nanjing Medical University (Grant No. KD2024KYJJ225), Huai 'an Natural Science Fund Project (Grant No. HABL202262) and Huai 'an Natural Science Fund Project (Grant No. HABL2023096).

Data availability

The raw data can be downloaded from the DATADRYAD database (www.Datadryad.org). Dryad Digital Repository. https://datadryad.org/stash/dataset/doi:10.5061%2Fdryad.8q0p192.

Competing interests

The authors declare no competing interests.