Abstract
Background
The impact of discordance between remnant cholesterol (RC) and low-density lipoprotein cholesterol (LDL-c) on diabetes, diabetic kidney disease (DKD), diabetic retinopathy (DR) and cardiovascular disease (CVD) remains unclear. This study aims to explore the association between the discordance and these outcomes, using data from the National Health and Nutrition Examination Survey (NHANES) 1999.1–2020.3 and Clinical Medical College & Affiliated Hospital of Chengdu University.
Methods
We prespecified a ± 15 percentile-point cutoff for discordance between RC and LDL-c, defined as RC percentile minus LDL-c percentile. Using this rule, 11,826 NHANES participants (cohort 1) and 306 participants from the Clinical Medical College & Affiliated Hospital of Chengdu University (cohort 2) were categorized as low discordance ( ≤ − 15), concordant (− 15 to + 15), or high discordance ( ≥ + 15). Key variables were screened by the Boruta algorithm. Logistic regression analysis models, restricted cubic spline (RCS), and subgroup analyses were used to assess the associations of discordance with outcomes. Receiver operating characteristic (ROC) and three machine learning models were used to assess the predictive value of the discordance for outcomes.
Results
High discordance was significantly associated with increased risks of diabetes (OR -cohort 1: 2.371, 95% CI: 1.848–3.055; OR -cohort 2: 4.064, 95% CI: 1.750–10.020), DKD (OR: 2.593, 95% CI: 1.930–3.521), DR (OR: 2.205, 95% CI: 1.404–3.556), and CVD (OR -cohort 1: 2.299, 95% CI: 1.900-2.791; OR -cohort 2: 3.220, 95% CI: 1.266–8.175). In cohort 1, the RCS analysis showed linear relationships for these outcomes. Hypertension, HOMA-IR ≥ 3.1 and HOMA-β < 100 were identified as significant modifiers in subgroup analyses. In cohort 2, the RCS analysis showed linear relationships for diabetes and non-linear for CVD. All machine learning models demonstrated great predictive value of the discordance for diabetes and CVD.
Conclusion
The discordance is a significant predictor of diabetes, diabetic microvascular diseases, and cardiovascular disease.
Key message
The discordance between RC and LDL-c is significantly associated with diabetes, diabetic microvascular diseases, and cardiovascular disease.
In NHANES, the RCS analysis showed linear relationships for diabetes, DKD, DR and CVD. In clinical cohort, the RCS analysis showed linear relationships for diabetes and non-linear for CVD.
The association of discordance with diabetes, DKD and CVD was more prevalent in individuals with hypertension and HOMA-IR ≥3.1, while individuals with hypertension were more sensitive to DR and individuals with HOMA-β <100 were more sensitive to diabetes.
The discordance showed the certain predictive ability for all outcomes.
Supplementary Information
The online version contains supplementary material available at 10.1186/s13098-025-02039-2.
Keywords: Remnant cholesterol, Low-density lipoprotein cholesterol, Discordance, Diabetes, Diabetic microvascular diseases, Cardiovascular diseases, Machine learning, Risk stratification.
Introduction
Diabetes mellitus (DM) is characterized by chronic hyperglycemia, caused by insufficient or defective insulin secretion [1]. Recent research indicates that the global prevalence of diabetes among adults has surpassed 800 million and will reach 1.3 billion by 2050 [2]. This rise will be accompanied by a significant increase in severe complications, particularly microvascular diseases such as diabetic kidney disease (DKD) and diabetic retinopathy (DR) [3–5]. Furthermore, diabetes is a major risk factor for cardiovascular disease (CVD), contributing to a substantial proportion of the global burden of heart disease, stroke, and other cardiovascular events [6, 7]. Given the profound impact of diabetes on individuals and healthcare systems, it is essential to investigate the factors influencing cardiovascular and microvascular diseases.
Among the various factors influencing the progression of diabetes, dyslipidemia plays an important role [8]. Traditional lipid markers, particularly total cholesterol (TC) and low-density lipoprotein cholesterol (LDL-c), have been utilized to assess cardiovascular risk and guide therapeutic decisions [9]. LDL-c has been a primary target of lipid-lowering therapies aimed at reducing cardiovascular events [10]. However, emerging evidence has highlighted the significance of remnant cholesterol (RC)—the cholesterol component of triglyceride-rich lipoproteins as an independent predictor of cardiovascular risk [11, 12]. Elevated RC levels are implicated in endothelial dysfunction, inflammation, and atherogenesis [13, 14]. Since one of the important causes of diabetes is the disorder of glucose and lipid metabolism, diabetes should be considered as a primary consideration [15]. Notably, the discordance between RC and LDL-c (RC percentile minus LDL-c percentile), has emerged as a potential marker for accurate risk stratification, particularly in diabetes populations [16]. Additionally, discordance may capture residual inflammatory risk mediated by triglyceride rich lipoproteins (TRL) and reflected by RC, which is not conveyed by LDL-c alone [17, 18]. LDL-c primarily indexes the cholesterol burden within LDL particles, so discordance could identify phenotypes at higher risk even when LDL-c levels meet guideline targets [12].
However, current studies have focued on individual lipid markers, neglecting the potential interplay between RC and LDL-c and their combined influence on diabetes progression [19–22]. It is unclear that the discordance serves as an accurate predictor of diabetes, diabetic cardiovascular and microvascular diseases. The objective of this study is to investigate the association between the discordance and diabetes, diabetic cardiovascular and microvascular diseases, using data from the National Health and Nutrition Examination Survey (NHANES) and Clinical Medical College & Affiliated Hospital of Chengdu University. By testing whether discordance improves risk stratification over LDL-c alone, our findings could inform patient profiling and therapeutic intensification in diabetes care.
Methods and materials
Study design and population
The NHANES is a crucial research program that aims to evaluate the health and nutritional status of both adults and children residing in the United States. The survey used a stratified, multistage probability design to recruit a representative sample of the U.S. population. Data for this research were taken from seven cycles of NHANES (1999.1–2020.3) with a total of 11,826 participants (Fig. 1A). Exclusion criteria included the following: (1) participants < 18 years of age; (2) participants with missing LDL-c, HDL-c and TC data; (3) participants with missing outcomes or covariates. All participants informed written consent was given at recruitment, the NCHS Research Ethics Review Board agreed to the study methodology, and no external ethics approval was required to perform the study. The detailed experimental design and data can be accessed at NHANES (https://www.cdc.gov/nchs/nhanes).
Fig. 1.
Flowchart of this study. Participants from (A) NHANES; (B) Participants from Clinical Medical College & Affiliated Hospital of Chengdu University
To validate the robustness and generalizability of our findings, the correlation between discordance and these outcomes prevalence was calculated using the data from Clinical Medical College & Affiliated Hospital of Chengdu University. The same inclusion and exclusion criteria were applied to ensure comparability with the NHANES participants (Fig. 1B). Ethical approval for the validation study was obtained from the Ethics Committee of Clinical Medical College & Affiliated Hospital of Chengdu University.
Assessment of discordance
We assessed discordance between RC (RC is equal to TC minus high-density lipoprotein cholesterol (HDL-c) minus LDL-c) and LDL-c percentiles by calculating the difference in their respective percentile ranks using the cumulative distribution function for each cohort. Discordance was defined as a difference greater than 15 percentile units (RC percentile minus LDL-c percentile). Based on previous studies, participants were categorized into three groups: the discordantly low group (RC percentile minus LDL-c percentile ≤ − 15 percentile units), the concordant group (RC percentile minus LDL-c percentile within ± 15 percentile units), and the discordantly high group (RC percentile minus LDL-c percentile ≥ + 15 percentile units) [23]. This discordance reflects the relative distance between two correlated markers, taking into account their distribution in percentiles. This approach has been widely applied in population-based studies, including those involving biomarkers such as cholesterol and estimated glomerular filtration rate [24–27].
Definition of diabetes, diabetic cardiovascular and microvascular diseases
Individuals meeting one or more of the following criteria were considered to have diabetes: (1) fasting blood glucose (FBG) ≥ 7.0 mmol/L (laboratory measurement); (2) Hemoglobin A1c (HbA1c) ≥ 6.5% (laboratory measurement); (4) use of diabetes medication or insulin (standardized questionnaire); and (5) self-reported doctor diagnosis of diabetes (standardized questionnaire) [28]. On this basis, DKD was defined using objective laboratory criteria: urinary albumin-to-creatinine ratio (UACR) >30 mg/g (spot urine) or estimated glomerular filtration rate (eGFR) < 60 mL/min/1.73 m² calculated by the Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) Eq. 2 [9], and DR status was ascertained by self-report in participants with diabetes, recorded via the standardized medical conditions questionnaire [30].
CVD was diagnosed using self-reported physician diagnoses obtained during an individual interview using a standardized medical condition questionnaire. The participants were asked, “Have you been told by a doctor or other health professional that you have congestive heart failure/coronary heart disease/angina pectoris/myocardial infarction/stroke”. A person was regarded as having CVD if he or she replied “yes” to any of the above questions [31]. The result was transformed to a binary variable.
Assessment of covariates
Demographics, examination, laboratory and questionnaire data of participants were collected. Demographic data included age, gender, race ethnicity, education level and marital status. Tobacco use was classified into two categories: never smoking (defined as having smoked fewer than 100 cigarettes in a lifetime) and smoking (defined as having smoked at least 100 cigarettes in a lifetime). Alcohol consumption was similarly categorized as never drinking (defined as consuming fewer than 12 alcoholic beverages per year) and alcohol consumption (defined as consuming at least 12 alcoholic beverages per year). Body mass index (BMI) was calculated by dividing the weight in kilograms by the square of the height in meters. The CKD-EPI equation was utilized to compute the eGFR. Hypertension was defined as measured systolic blood pressure (SBP) ≥ 140 mmHg or measured diastolic blood pressure (DBP) ≥ 90 mmHg [32]. TC and FBG were determined by enzymatic assay. HDL-c was measured directly and triglyceride was measured by colorimetric assay. LDL-c is calculated based on [LDL-c]=[TC]-[HDL-c]-[triglyceride/5]. HbA1c was determined by High-Performance Liquid Chromatography (HPLC) method. Further details on covariate measurements are available on the NHANES website (https://www.cdc.gov/nchs/nhanes/index.htm). In the clinical dataset, covariates were collected where available, but some covariates from the NHANES analysis were not available in the clinical dataset due to routine data collection practices.
Statistical analysis
Sample weights, clustering, and stratification were incorporated into all analyses to account for the complex sampling design of NHANES, as required for its data analysis. Continuous variables were summarized as means with standard deviations (SD), while categorical variables were expressed as frequencies and percentages. Logistic regression analysis were used to assess the odds ratio (OR) and 95% confidence interval (CI) of discordance with diabetes, DKD, DR and CVD. Three models were constructed to adjust for possible confounding factors. Model 1 was unadjusted. Model 2 was adjusted for age, gender, race ethnicity, Education Level, Marital Status, BMI, smoking status and alcohol status. Model 3 was further adjusted for FBG, triglycerides, and DBP. To explain the dose-response relationships (linear or nonlinear) between the discordance with diabetes, DKD, DR and CVD, restricted cubic spline (RCS) was conducted for nonlinear assessment. In addition, participants were categorized into subgroups based on age (< 60 or ≥ 60), gender (male or female), BMI (< 25, ≥ 25 to < 30, ≥30), triglyceride levels (< 1.7 or ≥ 1.7), HOMA-IR (< 3.1 or ≥ 3.1), HOMA-β (< 100 or ≥ 100), smoking status (yes, no), alcohol consumption (yes, no), and hypertension (yes, no). Subsequently, subgroup analyses were conducted based on these categorizations. Effect modification was assessed using multiplicative interaction terms between discordance and each prespecified subgroup in the fully adjusted model. To address multiplicity across interaction tests for the primary outcome, we controlled the false discovery rate at 5% using the Benjamini–Hochberg procedure; adjusted q-values are reported alongside interaction p-values [33]. Finally, we evaluated the discrimination ability and accuracy of the fully adjusted model using receiver operating characteristic (ROC) curves, area under the curve (AUC), and calibration curves.
In the clinical dataset, we only report diabetes and CVD due to data limitations. We used the Boruta algorithm, a supervised categorical feature selection method, to pinpoint all relevant features (Fig. 2). Next, the results of Boruta’s algorithm combined with actual clinical significance were used to screen the variables to be included in the multivariate logistic regression. Model 1 was unadjusted; Model 2 was adjusted for age, gender, BMI, height, smoking status and alcohol status; Model 3 was further adjusted for hypertension, eGFR, FBG and DBP. Because the Friedewald equation can be inaccurate at higher triglyceride concentrations, we performed two prespecified sensitivity analyses [34]. First, we restricted the analytic sample to participants with triglyceride < 400 mg/dL (4.52 mmol/L) and retained the original exposure definition. Second, we re-estimated LDL-C using the Sampson equation in the full sample, then recomputed RC and discordance and re-categorized discordance using the same ± 15 percentile thresholds as in the primary analysis. For both analyses we fit the same fully adjusted logistic models as in the primary analysis.
Fig. 2.
Feature selection process based on Boruta algorithm. Participants from Clinical Medical College & Affiliated Hospital of Chengdu University (A)Diabetes; (B) Cardiovascular disease
Using RCS to explore nonlinear correlations between the discordance with diabetes and CVD. ROC analysis was used to evaluate the predictive performance of discordance. Three supervised machine learning (ML) algorithms—multivariate logistic regression, Random Forest, and XGBoost—were used to assess the predictive contribution of discordance. We prespecified severe class imbalance as minority-class prevalence < 20%. Outcome prevalence in the clinical cohort was approximately balanced (diabetes 24.5% [75/306]; CVD 24.2% [74/306]). Therefore, no re-sampling or class weighting was applied. The models included variables such as age, gender, height, BMI, DBP, smoking, drinking, hypertension, eGFR, and discordance. Model performance was evaluated using metrics such as the AUC and accuracy.
Data analysis was completed by software Stata (version 17.0) and R (version 4.5.0). P < 0.05 on both sides was considered statistically significant.
Results
Baseline characteristics of study participants
The screening process for the study population is presented in Fig. 1. We included 11,826 participants from NHANES and 306 participants from were Clinical Medical College & Affiliated Hospital of Chengdu University in the analysis. Table 1 and Table 2 shows the baseline characteristics of the study population by concordant/discordant categories between RC and LDL-c. Individuals in the high discordant group were older, had higher BMI and SBP (P < 0.05), and had higher triglyceride levels but lower HDL-c levels than those in the concordant and low discordant groups. In addition, there were significant differences in biochemical markers between the groups, with participants in the high discordant group having significantly higher levels of FBG, HbA1c, fasting insulin compared to individuals in the concordant and low discordant groups (P < 0.05). In the clinical dataset, individuals in the high discordant group had higher BMI and SBP (P < 0.05), and had higher RC and triglyceride levels but lower HDL-c, LDL-c and TC levels than those in the concordant and low discordant groups (P < 0.001).
Table 1.
Baseline characteristics of the study population from NHANES
| Low discordance | Concordant | High discordance | P-value | |
|---|---|---|---|---|
| Participants, n | 4139 | 3548 | 4139 | |
| Age (years) | 49.60 (17.07) | 50.49 (17.68) | 52.43 (18.61) | < 0.001 |
| Height (cm) | 166.95 (10.02) | 166.68 (10.33) | 166.88 (10.42) | 0.501 |
| Weight (kg) | 75.01 (20.02) | 75.70 (20.96) | 79.04 (21.38) | < 0.001 |
| BMI (kg/m 2 ) | 26.79 (6.34) | 27.11 (6.62) | 28.25 (6.75) | < 0.001 |
| Waist Circumference (cm) | 92.26 (17.23) | 93.96 (18.70) | 97.83 (19.22) | < 0.001 |
| Gender (%) | 0.172 | |||
| Male | 1943 (46.9) | 1649 (46.5) | 2007 (48.5) | |
| Female | 2196 (53.1) | 1899 (53.5) | 2132 (51.5) | |
| Race Ethnicity (%) | < 0.001 | |||
| Mexican American | 708 (17.1) | 790 (22.3) | 1292 (31.2) | |
| Other Hispanic | 427 (10.3) | 413 (11.6) | 487 (11.8) | |
| Non-Hispanic White | 1958 (47.3) | 1689 (47.6) | 1780 (43.0) | |
| Non-Hispanic Black | 930 (22.5) | 543 (15.3) | 451 (10.9) | |
| Other Race | 116 (2.8) | 113 (3.2) | 129 (3.1) | |
| Education Level (%) | < 0.001 | |||
| Less than 9th grade | 1134 (27.4) | 1187 (33.5) | 1321 (31.9) | |
| 9-11th grade | 668 (16.1) | 616 (17.4) | 737 (17.8) | |
| High school graduate | 731 (17.7) | 593 (16.7) | 740 (17.9) | |
| Some college | 898 (21.7) | 648 (18.3) | 804 (19.4) | |
| College graduate or above | 708 (17.1) | 504 (14.2) | 537 (13.0) | |
| Marital Status (%) | 0.004 | |||
| Married | 2113 (51.1) | 1812 (51.1) | 2179 (52.6) | |
| Widowed | 295 (7.1) | 270 (7.6) | 357 (8.6) | |
| Divorced | 427 (10.3) | 310 (8.7) | 393 (9.5) | |
| Separated | 138 (3.3) | 127 (3.6) | 146 (3.5) | |
| Never married | 851 (20.6) | 759 (21.4) | 743 (18.0) | |
| Living with partner | 315 (7.6) | 270 (7.6) | 321 (7.8) | |
| Smoking Status (%) | < 0.001 | |||
| Yes | 1720 (41.6) | 1638 (46.2) | 2074 (50.1) | |
| No | 2419 (58.4) | 1910 (53.8) | 2065 (49.9) | |
| Alcohol Status (%) | 0.511 | |||
| Yes | 2670 (64.5) | 2326 (65.6) | 2845 (68.7) | |
| No | 1469 (35.5) | 1222 (34.4) | 1294 (31.3) | |
| Hypertension (%) | 0.197 | |||
| Yes | 738 (17.8) | 615 (17.3) | 781 (18.9) | |
| No | 3401 (82.2) | 2933 (82.7) | 3358 (81.1) | |
| Diabetes (%) | < 0.001 | |||
| Yes | 235 (5.7) | 334 (9.4) | 744 (18.0) | |
| No | 3904 (94.3) | 3214 (90.6) | 3395 (82.0) | |
| Diabetic kidney disease (%) | < 0.001 | |||
| Yes | 76 (1.8) | 104 (2.9) | 292 (7.1) | |
| No | 4063 (98.2) | 3444 (97.1) | 3847 (92.9) | |
| Diabetic retinopathy (%) | < 0.001 | |||
| Yes | 29 (0.7) | 42 (1.2) | 96 (2.3) | |
| No | 4110 (99.3) | 3506 (98.8) | 4043 (97.7) | |
| CVD (%) | < 0.001 | |||
| Yes | 184 (4.4) | 233 (6.6) | 484 (11.7) | |
| No | 3955 (95.6) | 3315 (93.4) | 3655 (88.3) | |
| Congestive heart failure (%) | 0.046 | |||
| Yes | 433 (10.5) | 438 (12.3) | 444 (10.7) | |
| No | 3706 (89.5) | 3110 (87.7) | 3695 (89.3) | |
| Coronary heart disease (%) | < 0.001 | |||
| Yes | 983 (23.7) | 1069 (30.1) | 1237 (29.9) | |
| No | 3156 (76.3) | 2479 (69.9) | 2902 (70.1) | |
| Angina (%) | 0.002 | |||
| Yes | 423 (10.2) | 442 (12.5) | 511 (12.3) | |
| No | 3716 (89.8) | 3106 (87.5) | 3628 (87.7) | |
| Heart attack (%) | < 0.001 | |||
| Yes | 548 (13.2) | 609 (17.2) | 709 (17.1) | |
| No | 3591 (86.8) | 2939 (82.8) | 3430 (82.9) | |
| Stroke (%) | < 0.001 | |||
| Yes | 247 (6.0) | 276 (7.8) | 358 (8.6) | |
| No | 3892 (94.0) | 3272 (92.2) | 3781 (91.4) | |
| TC (mg/dL) | 205.11 (38.27) | 191.08 (46.32) | 170.57 (32.14) | < 0.001 |
| LDL-C (mg/dL) | 131.95 (32.55) | 115.09 (36.46) | 90.80 (23.03) | < 0.001 |
| HDL-C (mg/dL) | 57.11 (14.90) | 52.35 (14.49) | 48.40 (14.18) | < 0.001 |
| Remnant_cholesterol (mg/dL) | 16.05 (6.07) | 23.64 (12.61) | 31.36 (14.42) | < 0.001 |
| FBG (mmol/L) | 5.48 (1.21) | 5.61 (1.54) | 5.95 (1.93) | < 0.001 |
| HbA1c (%) | 5.43 (0.69) | 5.49 (0.89) | 5.63 (1.02) | < 0.001 |
| Fasting Insulin (uU/mL) | 11.03 (8.28) | 13.02 (9.65) | 16.31 (14.98) | < 0.001 |
| Urine albumin (mg/L) | 35.97 (269.39) | 42.13 (480.38) | 53.54 (389.96) | 0.109 |
| CR (g/L) | 1.39 (0.84) | 1.36 (0.82) | 1.35 (0.79) | 0.061 |
| Triglycerides (mmol/L) | 1.63 (1.49) | 2.08 (1.47) | 2.30 (1.32) | < 0.001 |
| ALT (U/L) | 28.17 (26.05) | 30.40 (30.18) | 29.68 (24.05) | 0.001 |
| BUN (mmol/L) | 7.23 (5.43) | 7.36 (5.73) | 7.22 (5.62) | 0.482 |
| Serum Creatinine (mg/dL) | 0.70 (0.30) | 0.69 (0.42) | 0.73 (0.41) | < 0.001 |
| eGFR (mL/min/1.73 m²) | 61.95 (48.81) | 57.62 (49.75) | 58.47 (48.41) | < 0.001 |
| HOMA-IR | 3.25 (7.34) | 3.78 (6.78) | 4.88 (8.20) | < 0.001 |
| HOMA-B | 122.92 (88.43) | 138.78 (94.67) | 155.27 (119.89) | < 0.001 |
| SBP (mmHg) | 118.93 (20.11) | 118.64 (19.64) | 119.75 (19.71) | 0.037 |
| DBP (mmHg) | 70.20 (15.53) | 69.14 (16.04) | 68.36 (16.15) | < 0.001 |
Continuous numerical variables are expressed as mean (standard deviation) and categorical variables are expressed as numbers (percentages)
NHANES, National Health and Nutrition Examination Survey; BMI, Body mass index; TC, Total Cholesterol; LDL-c, Low density lipoprotein cholesterol; HDL-C, High density lipoprotein cholesterol; RC, Remnant cholesterol; FBG, Fasting blood glucose; HbA1c, Hemoglobin A1c; CR, Creatinine; ALT, Alanine Aminotransferase; BUN, Blood Urea Nitrogen; eGFR, estimated Glomerular Filtration Rate; HOMA-IR, Homeostasis Model Assessment of Insulin Resistance; HOMA-β, Homeostasis Model Assessment of β-cell Function; SBP, Systolic Blood Pressure; DBP, Diastolic Blood Pressure; CVD, Cardiovascular Disease
Table 2.
Baseline characteristics of the study population from clinical medical college & affiliated hospital of Chengdu university
| Low discordance | Concordant | High discordance | P-value | |
|---|---|---|---|---|
| Participants, n | 107 | 92 | 107 | |
| Age (years) | 45.21 (12.01) | 48.64 (14.57) | 52.33 (14.35) | 0.001 |
| Height (cm) | 163.60 (8.35) | 161.93 (8.11) | 164.21 (6.80) | 0.108 |
| Weight (kg) | 66.23 (16.35) | 66.40 (14.89) | 68.22 (14.02) | 0.574 |
| BMI (kg/m 2 ) | 24.53 (4.57) | 25.18 (4.47) | 25.20 (4.27) | 0.469 |
| Gender (%) | 0.001 | |||
| Male | 47 (43.9) | 38 (41.3) | 69 (64.5) | |
| Female | 60 (56.1) | 54 (58.7) | 38 (35.5) | |
| Smoking Status (%) | 0.215 | |||
| Yes | 32 (29.9) | 19 (20.7) | 33 (30.8) | |
| No | 75 (70.1) | 73 (79.3) | 74 (69.2) | |
| Alcohol Status (%) | 0.491 | |||
| Yes | 43 (40.2) | 32 (34.8) | 46 (43.0) | |
| No | 64 (59.8) | 60 (65.2) | 61 (57.0) | |
| Hypertension (%) | 0.003 | |||
| Yes | 31 (29.0) | 34 (37.0) | 55 (51.4) | |
| No | 76 (71.0) | 58 (63.0) | 52 (48.6) | |
| Diabetes (%) | < 0.001 | |||
| Yes | 12 (11.2) | 17 (18.5) | 46 (43.0) | |
| No | 95 (88.8) | 75 (81.5) | 61 (57.0) | |
| CVD (%) | < 0.001 | |||
| Yes | 11 (10.3) | 18 (19.6) | 45 (42.1) | |
| No | 96 (89.7) | 74 (80.4) | 62 (57.9) | |
| TC (mg/dL) | 197.94 (39.15) | 183.86 (47.97) | 172.09 (47.70) | < 0.001 |
| LDL-c (mg/dL) | 120.84 (31.10) | 103.02 (31.59) | 78.00 (24.67) | < 0.001 |
| HDL-c (mg/dL) | 56.10 (13.45) | 48.87 (10.68) | 39.63 (12.52) | < 0.001 |
| RC(mg/dL) | 21.00 (8.63) | 21.00 (8.63) | 54.46 (50.21) | < 0.001 |
| FBG (mmol/L) | 6.49 (3.06) | 7.74 (4.21) | 7.19 (3.76) | 0.057 |
| Creatinine (mg/dL) | 0.74 (0.22) | 0.75 (0.24) | 0.87 (0.35) | 0.001 |
| Triglycerides (mmol/L) | 122.33 (54.19) | 179.66 (208.48) | 294.19 (297.82) | < 0.001 |
| eGFR (mL/min/1.73 m²) | 106.90 (15.52) | 104.05 (17.99) | 97.94 (24.53) | 0.004 |
| SBP (mmHg) | 118.97 (13.87) | 122.63 (13.83) | 124.06 (14.82) | 0.028 |
| DBP (mmHg) | 72.21 (10.99) | 73.41 (11.72) | 73.18 (10.92) | 0.715 |
Continuous numerical variables are expressed as mean (standard deviation) and categorical variables are expressed as numbers (percentages)
BMI, Body mass index; TC, Total Cholesterol; LDL-c, Low density lipoprotein cholesterol; HDL-C, High density lipoprotein cholesterol; RC, Remnant cholesterol; FBG, Fasting blood glucose; eGFR, estimated Glomerular Filtration Rate; SBP, Systolic Blood Pressure; DBP, Diastolic Blood Pressure; CVD, Cardiovascular Disease
Associations of the discordance with diabetes, diabetic microvascular diseases, and cardiovascular disease
Table 3 illustrates the associations of the discordance with diabetes, diabetic microvascular diseases, and CVD. Compared to the low discordance group, participants in the high discordance group exhibited significantly elevated risks for diabetes (OR: 2.371, 95% CI: 1.848–3.055, P < 0.001), DKD (OR: 2.593, 95% CI: 1.930–3.521, P < 0.001), DR (OR: 2.205, 95% CI: 1.404–3.556, P < 0.001), and CVD (OR: 2.299, 95% CI: 1.900–2.791, P < 0.001) in fully adjusted models. Besides, there was some significant difference shown here between discordance and individual cardiovascular disease (Table S1). Using the dataset from Clinical Medical College & Affiliated Hospital of Chengdu University confirmed diabetes (OR: 4.064, 95% CI: 1.750–10.020, P = 0.002) and CVD (OR: 3.220, 95% CI: 1.266–8.175, P = 0.017) findings (Table 3).
Table 3.
Logistic regression analysis for the associations between the discordance with diabetes, diabetic microvascular diseases, and cardiovascular disease
| Outcomes | Low discordance | Concordant | High discordance | P for trend | |||
|---|---|---|---|---|---|---|---|
| HR (95% CI) | P-value | HR (95% CI) | P-value | HR (95% CI) | P-value | ||
| Participants from NHANES | |||||||
| Diabetes | |||||||
| Model 1 | 1 [Reference] | 1.726 (1.453, 2.055) | < 0.001 | 3.641 (3.127, 4.253) | < 0.001 | < 0.001 | |
| Model 2 | 1 [Reference] | 1.732 (1.444, 2.080) | < 0.001 | 3.205 (2.722, 3.784) | < 0.001 | < 0.001 | |
| Model 3 | 1 [Reference] | 1.612 (1.229, 2.119) | < 0.001 | 2.371 (1.848, 3.055) | < 0.001 | < 0.001 | |
| DKD | |||||||
| Model 1 | 1 [Reference] | 1.614 (1.199, 2.184) | 0.002 | 4.058 (3.159, 5.277) | < 0.001 | < 0.001 | |
| Model 2 | 1 [Reference] | 1.567 (1.155, 2.136) | 0.004 | 3.249 (2.500, 4.272) | < 0.001 | < 0.001 | |
| Model 3 | 1 [Reference] | 1.390 (0.989, 3.521) | < 0.001 | 2.593 (1.930, 3.521) | < 0.001 | < 0.001 | |
| DR | |||||||
| Model 1 | 1 [Reference] | 1.698 (1.060, 2.756) | 0.029 | 3.365 (2.247, 5.197) | < 0.001 | < 0.001 | |
| Model 2 | 1 [Reference] | 1.666 (1.036, 2.714) | 0.037 | 2.830 (1.869, 4.412) | < 0.001 | < 0.001 | |
| Model 3 | 1 [Reference] | 1.520 (0.917, 2.552) | 0.107 | 2.205 (1.404, 3.556) | < 0.001 | < 0.001 | |
| CVD | |||||||
| Model 1 | 1 [Reference] | 1.511 (1.239, 1.844) | < 0.001 | 2.846 (2.393, 3.400) | < 0.001 | < 0.001 | |
| Model 2 | 1 [Reference] | 1.525 (1.239, 1.879) | < 0.001 | 2.508 (2.082, 3.032) | < 0.001 | < 0.001 | |
| Model 3 | 1 [Reference] | 1.469 (1.191, 1.814) | < 0.001 | 2.299 (1.900, 2.791) | < 0.001 | < 0.001 | |
| Participants from Clinical Medical College & Affiliated Hospital of Chengdu University | |||||||
| Diabetes | |||||||
| Model 1 | 1 [Reference] | 0.557 (0.246, 1.230) | 0.151 | 3.327 (1.762, 6.513) | < 0.001 | < 0.001 | |
| Model 2 | 1 [Reference] | 0.794 (0.301, 2.078) | 0.637 | 4.355 (1.931, 10.421) | 0.001 | < 0.001 | |
| Model 3 | 1 [Reference] | 0.680 (0.243, 1.863) | 0.453 | 4.064 (1.750, 10.020) | 0.002 | < 0.001 | |
| CVD | |||||||
| Model 1 | 1 [Reference] | 0.471 (0.204, 1.045) | 0.068 | 2.984 (1.591, 5.779) | < 0.001 | < 0.001 | |
| Model 2 | 1 [Reference] | 0.834 (0.300, 2.308) | 0.726 | 3.603 (1.501, 9.220) | 0.005 | < 0.001 | |
| Model 3 | 1 [Reference] | 0.750 (0.248, 2.235) | 0.605 | 3.220 (1.266, 8.175) | 0.017 | 0.002 | |
DKD, diabetic kidney disease; DR, diabetic retinopathy; CVD, Cardiovascular Disease; OR: odds ratio; CI: confidence interval
Participants from NHANES: Model 1 was unadjusted; Model 2 was adjusted for age, gender, ethnicity, education Level, marital status, BMI, smoking status and alcohol status; Model 3 was further adjusted for FBG, triglycerides, and DBP
Participants from Clinical Medical College & Affiliated Hospital of Chengdu University: Model 1 was unadjusted; Model 2 was adjusted for age, gender, BMI, height, smoking status and alcohol status; Model 3 was further adjusted for hypertension, eGFR, FBG and DBP
Sensitivity analyses
Restriction to triglyceride < 400 mg/dL yielded effect estimates that were almost in the direction of the primary analysis. Using the Sampson equation to re-estimate LDL-C and recompute RC–LDL-C discordance produced broadly similar results. Additional details regarding the sensitivity analyses are provided in Table S2. Overall, the pattern and direction of effects were consistent with the primary findings, indicating robustness to the LDL-C estimation method.
Restricted cubic spline analysis exploring the relationship between the discordance with diabetes, diabetic microvascular diseases, and cardiovascular disease
The RCS analysis (Fig. 3A-D) revealed significant associations between the discordance and the risks of diabetes, diabetic microvascular diseases, and CVD. In NHANES cohort, for diabetes, DKD, DR, and CVD, the relationships were linear (P-overall < 0.001, P-non-linear > 0.05), with risks increasing more sharply at higher levels of discordance. Figure S1 shows the relationship between the discordance with individual CVD. In Clinical Medical College & Affiliated Hospital of Chengdu University dataset, the RCS analysis showed linear relationships for diabetes and non-linear for CVD (Fig. 3E and F).
Fig. 3.
Restricted cubic spline analysis (A)-(D): Participants from NHANES; (A) Diabetes; (B) Diabetic kidney disease; (C) Diabetic retinopathy; (D) Cardiovascular disease (E)-(F): Participants from Clinical Medical College & Affiliated Hospital of Chengdu University; (E) Diabetes; (F) Cardiovascular disease
Subgroup analysis
We performed a subgroup analysis to evaluate the impact of discordance on outcome measures. Figures S2-S5 present the results. A stronger association between elevated discordance and diabetes was observed among individuals with BMI ≥ 30 (OR 2.424, 95% CI 1.715–3.426, P < 0.001, P for interaction = 0.026), hypertension (OR 2.163, 95% CI 1.440–3.248, P < 0.001, P for interaction < 0.001), HOMA-IR ≥ 3.1 (OR 2.164, 95% CI 1.556–3.011, P < 0.001, P for interaction < 0.001), and HOMA-β < 100 (OR 3.041, 95% CI 2.110–4.382, P < 0.001, P for interaction < 0.001). For DKD, individuals with hypertension and HOMA-IR ≥ 3.1 in the high discordance group demonstrated increased susceptibility to DKD (P for interaction < 0.05). Furthermore, a significant interaction between discordance and DR was identified in individuals with a history of alcohol consumption and hypertension (P for interaction < 0.05). Notably, significant interactions between discordance and all subgroups were observed in relation to CVD (P for interaction < 0.05). Additional details regarding the subgroup analysis are provided in Figures S2–S5. After Benjamini–Hochberg FDR adjustment at 5%, CVD displayed clear effect modification by age, gender, hypertension, smoking, alcohol use, BMI, triglycerides, HOMA-IR, and HOMA-β (q < 0.05). For DKD, significant interactions were seen for gender, hypertension, alcohol use, BMI, triglycerides, and HOMA-IR (q < 0.05), whereas age, smoking, and HOMA-β did not reach significance (q < 0.05). In DM, interactions survived correction for gender, hypertension, BMI, triglycerides, HOMA-IR, and HOMA-β (q < 0.05), but age, smoking, and alcohol use were not significant. For DR, only hypertension remained significant after adjustment (q < 0.05). Stratum-specific adjusted effect estimates are reported in Table S3.
Predictive ability of the discordance for diabetes, diabetic microvascular diseases, and cardiovascular disease
The ROC analysis in Figure S6 highlights the predictive performance of the discordance for diabetes, diabetic microvascular diseases, and CVD. Among the outcomes, the discordance showed the strongest predictive ability for DKD (AUC = 0.688), followed by diabetes (AUC = 0.669), and DR (AUC = 0.662), indicating its reliable discrimination for these conditions. For CVD, the predictive performance was fair (AUC = 0.642).
In Clinical Medical College & Affiliated Hospital of Chengdu University dataset, the discordance showed the great predictive ability for diabetes (AUC = 0.734) and CVD (AUC = 0.742) (Figure S6). Table 4 shows the performance of multivariate logistic regression, random forest, and XGBoost models in predicting diabetes and CVD. XGBoost exhibited the highest AUC for diabetes (0.886 [95% CI: 0.845–0.927]) and random forest for CVD (0.898 [95% CI: 0.856–0.940]), with all models demonstrating strong accuracy and specificity. Additionally, all ML models identified the discordance as one of the most important predictors, indicating its robustness and potential clinical utility across different modeling approaches (Figure S7-S8).
Table 4.
Machine learning for predicting diabetes and cardiovascular disease
| Model | AUC (95% CI) | Accuracy | Sensitivity | Specificity |
|---|---|---|---|---|
| Diabetes | ||||
| Multivariate logistic regression | 0.845 (0.790, 0.901) | 0.843 (0.798, 0.882) | 0.627 | 0.913 |
| Random Forest | 0.876 (0.828, 0.925) | 0.837 (0.790, 0.876) | 0.493 | 0.948 |
| XGBoost | 0.886 (0.845, 0.927) | 0.853 (0.808, 0.891) | 0.627 | 0.926 |
| Cardiovascular disease | ||||
| Multivariate logistic regression | 0.896 (0.848, 0.944) | 0.869 (0.826, 0.905) | 0.703 | 0.922 |
| Random Forest | 0.898 (0.856, 0.940) | 0.853 (0.808, 0.891) | 0.622 | 0.927 |
| XGBoost | 0.896 (0.852, 0.941) | 0.859 (0.815, 0.896) | 0.662 | 0.922 |
Participants from Clinical Medical College & Affiliated Hospital of Chengdu University
Features: age, gender, height, BMI, DBP, smoking, drinking, hypertension, eGFR, discordance
Discussion
Our primary research results demonstrated the following: (1) The discordance between RC and LDL-c is significantly associated with diabetes, diabetic microvascular diseases, and CVD. (2) Diabetes, DKD, DR, and CVD were linearly associated with discordance in NHANES cohort, whereas in the clinical cohort diabetes was linear but CVD showed non-linear. (3) The association of discordance with diabetes, DKD and CVD was more prevalent in individuals with hypertension and HOMA-IR ≥ 3.1, while individuals with hypertension were more sensitive to DR and individuals with HOMA-β < 100 were more sensitive to diabetes. (4) The discordance showed the certain predictive ability for all outcomes.
The discordance between RC and LDL-c reflects an imbalance in lipid metabolism, which may outperform single markers because it captures a mismatch between cholesterol carriage and triglyceride-rich remnant burden, thereby integrating pro-inflammatory, endothelial dysfunction, and remnant-lipoprotein atherogenicity beyond LDL-c levels alone [35–37]. This makes discordance a potentially useful clinical marker derivable from routine lipid panels. At the same time, our study shows that this discordance is significantly associated with diabetes, DKD, DR, and CVD. Elevated RC as part of triglyceride-rich lipoproteins and their remnants promotes vascular inflammation including monocyte and macrophage activation, endothelial dysfunction with reduced nitric oxide bioavailability, and oxidative stress [20, 25, 38]. Remnant particles can penetrate and be retained in the arterial wall which fuels plaque growth even when LDL-c is at target [19]. In the pancreas, cholesterol and remnant exposure may impair β-cell survival and insulin secretion which aggravates insulin resistance and glucose dysregulation [20, 39]. These pathways differ from the predominant LDL-c role in cholesterol deposition and they help explain why high discordance with RC disproportionately elevated relative to LDL-c tracks risk across metabolic and vascular beds. Several recent studies have shown that RC is an independent predictor of diabetes, and that discordance between RC and LDL-c lead to a higher risk of developing diabetes [16, 20]. A nationwide Korean population-based study shows that higher RC is independently associated with the incidence of chronic kidney disease (CKD) in newly diagnosed patients with type 2 diabetes (T2D), even when routine lipid levels are well controlled [39]. At the same time, several recent meta-analyses have shown that RC is directly associated with risk factors for CKD, with a significant negative association observed between RC and estimated kidney function, with a 24% increased risk of progression to end-stage renal disease for every 1 standard deviation increase in T2D-associated CKD patients [40]. In addition, a cross-sectional study suggests that RC may be a risk factor for DR in patients with T2D [21]. The discordance between RC and LDL-c has been gradually applied to predict the risk of CVD [25]. Besides, a study involving two cohorts showed that the discordance between RC and LDL-c, representing the intraindividual discrepancy, is significantly associated with stroke onset among Chinese adults [23].
The observed discordance indicates a metabolic environment dominated by RC, which may disrupt normal glucose regulation. The linear relationships observed in diabetes, DKD, DR and CVD in NHANES suggest that a small difference between RC and LDL-c can lead to a gradual cumulative increase in disease risk, while a larger difference further amplifying this adverse effect through mechanisms such as inflammatory response, endothelial dysfunction and lipid deposition. Animal experiments have shown that the high cholesterol environment inhibits the survival of pancreatic beta cells, resulting in the inhibition of insulin secretion [41]. RC is a component of triglyceride-rich lipoproteins, which are more abundant, larger, and carry more cholesterol than LDL-c particles [12, 42]. Therefore, RC may indirectly exacerbate insulin resistance and disease progression by aggravating β-cell dysfunction and glucose metabolism disorders. This is consistent with our subgroup findings. Insulin resistance for example HOMA-IR ≥ 3.1 promotes hepatic VLDL overproduction and RC elevation which indicates a more atherogenic and lipotoxic milieu [16, 43]. Beta-cell dysfunction with HOMA-β < 100 limits compensatory insulin secretion which makes glycemic control more vulnerable to RC linked metabolic stress [41]. Clinically, these groups may merit tighter blood pressure control, triglyceride focused lifestyle measures including dietary modification, weight loss, and glycemic optimization, and closer surveillance when discordance is high. In diabetic microvascular diseases, RC may aggravate capillary damage through its role in endothelial activation and leukocyte adhesion, mechanisms that promote the progression of DKD and DR [39, 44]. Chen et al. reported that higher RC levels are associated with wider retinal arterioles and venules and higher fractal dimension, which may promote DR Through dilation of the retinal venules [21]. The linear relationship between discordance and DKD suggests a dose-dependent impact, potentially mediated by sustained inflammatory and oxidative stress pathways [45]. For CVD, unlike LDL-c, which primarily drives atherosclerosis through cholesterol deposition in arterial walls, RC exerts its effects through inflammatory and pro-atherogenic pathways [46, 47]. Hypertension may amplify discordance related risk through baseline endothelial dysfunction, increased arterial stiffness, and heightened inflammatory signaling, which makes hypertensive people more sensitive to abnormal lipid metabolism [48]. Our study shows that elevated levels of discordance are strongly associated with an increased risk of cardiovascular events, and that high blood pressure amplifies this risk by exacerbating arteriosclerosis and inducing endothelial inflammatory responses. Differences in RCS shape likely reflect data and precision rather than analytic inconsistency. Compared with NHANES, the clinical cohort was smaller with fewer CVD events and with events concentrated at higher exposure alongside sparse data at the low exposure tail. Flexible curves are least stable where data are sparse, so confidence intervals widen and apparent curvature can be magnified at extremes, which plausibly explains the nonlinear CVD pattern while diabetes remained approximately linear.
In addition to traditional biomarkers, ML models are increasingly being used to predict outcomes in clinical settings. A recent systematic review and meta-analysis demonstrated the growing potential of ML algorithms in predicting disease outcomes, especially in areas like CVD and diabetes [49, 50]. In our setting, discordance can act as a high value feature because it may capture nonlinear interactions with hypertension, insulin resistance, and kidney function. This study found that XGBoost and random forest models, in particular, exhibited high predictive accuracy, with XGBoost achieving an AUC of 0.886 for diabetes and random forest achieving an AUC of 0.898 for CVD. These findings suggest that ML models can outperform traditional methods in terms of predictive power. Our present analyses focus on association and risk estimation, so the machine learning component should be viewed as exploratory. Future work will evaluate calibration, quantify clinical utility using decision curves, and define thresholds that are validated externally before clinical deployment. Because discordance is computable from routine lipid panels, it is feasible for opportunistic screening in primary care or cardiometabolic clinics. When discordance is high, clinicians can intensify lifestyle change and consider triglyceride and remnant cholesterol lowering strategies alongside LDL-c management. Further research could focus on integrating these models with biomarkers to refine predictions and provide more clinically relevant insights, ultimately improving clinical decision-making and patient outcomes.
Strengths and limitations
This study has several advantages. Firstly, this study builds upon and extends prior research by addressing a critical gap: the discordance between RC and LDL-c as predictors of diabetes and its cardiovascular and microvascular diseases. Unlike previous studies that predominantly focused on the impact of a single lipid indicator, this research explores the interplay and discordance between these two markers, offering a novel perspective. The study’s approach provides a more comprehensive understanding of metabolic and vascular risk by capturing the imbalance between pro-inflammatory and cholesterol-driven pathways. Secondly, we applied of the Boruta algorithm to select the inclusion factors for multifactorial logistic regression on the basis of a large sample of data, which improved the confidence and accuracy of the study. In addition, we used machine learning algorithms to assess the predictive contribution of discordance. Thirdly, we conducted external validation using an independent clinical dataset, which enhanced the robustness and generalizability of our findings across different populations and healthcare settings.
Despite these advancements, the study has some limitations relative to prior research. First, its observational design (NHANES cross-sectional and a retrospective hospital cohort) limits causal inference and may introduce selection bias. Second, outcome ascertainment relied on standardized self-report for CVD and DR in NHANES, which may under-ascertain events compared with adjudicated records or graded fundus photography. Third, we lacked consistent data on medication exposures (e.g., statins and other lipid-lowering agents, antihypertensives, antidiabetic therapies) and diabetes duration across cycles, leaving potential residual confounding. Fourth, while our metrics captured comprehensive risk relationships, the reliance on single baseline measures of RC and LDL-c is consistent with earlier studies and remains a limitation, potentially underestimating intra-individual variability over time. Finally, the external validation cohort was small and hospital-based, which may limit generalizability despite applying the same eligibility criteria. We therefore view it as a consistency check rather than definitive replication. Future longitudinal studies with repeated lipid measurements, richer medication data, adjudicated or imaging-based endpoints, and larger, population-based validation cohorts are warranted to confirm temporality and refine risk stratification based on the discordance.
Conclusion
In summary, this study identified RC and LDL-c discordance as a significant predictor for diabetes, diabetic cardiovascular and microvascular diseases. These findings underscore the importance of lipid for risk stratification and the need for further research to validate these results in diverse populations.
Supplementary Information
Acknowledgements
We thank the NCHS for providing access to the NHANES data and the participants for their valuable contributions. Special thanks to the research staff and data analysts for ensuring the quality of the data used in this study.
Author contributions
YW and NL, Data curation-Equal, Formal analysis-Equal, Funding acquisition-Equal, Methodology-Equal, Project administration-Equal, Resources-Equal, Supervision-Equal, Writing – original draft Equal, Writing – review & editing-Equal; XS, GX, JW, and MW, Data curation-Equal, Formal analysis-Equal, Writing – original draft-Equal, Writing – review & editing-Equal; XN, XH, HL, JM, WW, JL, JZ, NL, KW, SL and TY, Formal analysis-Equal, Writing – original draft-Equal, Writing – review & editing-Equal; YW and ZS, Funding acquisition-Equal, Project administration-Equal, Writing – review & editing Equal.
Funding
All authors were supported by the Natural Science Foundation of Sichuan Province (NO.2025ZNSFSC0743, NO.2024NSFSC1618, NO.2024YFFK0287); the Key Research and Development Program of Chengdu Economic Development Zone New Economy and Science and Technology Bureau (NO.2025LQRD0014, NO.2024LQRD0048, NO.2025LQRD0024); Academic Degree and Postgraduate Education Reform Project of Sichuan Province (NO.CDJGY2024006); Chengdu Medical Research Project (2022291, 2023618); Chengdu University Research Initiation Programme (2081923030); The President’s Fund of Nanfang Hospital, Southern Medical University (2024B012); Achievements of the provincial college Student Entrepreneurship Training Program of Chengdu University (S202511079022X).
Data availability
The datasets generated or analyzed during the study are available from the corresponding author on reasonable request.
Declarations
Ethics approval and consent to participate
The NHANES has been approved by the National Center for Health Statistics Ethics Review Board, and all participants were provided informed written consent at enrollment. Ethical approval for the validation study was obtained from the Ethics Committee of Clinical Medical College & Affiliated Hospital of Chengdu University.
Consent for publication
Not applicable.
Competing interests
The authors declare no competing interests.
Footnotes
Publisher’s note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Yao Wang and Na Li contributed equally to the work.
Contributor Information
Yuqing Wu, Email: seekorseek@126.com.
Zheng Shi, Email: shizheng@cdu.edu.cn.
References
- 1.Diagnosis and Classification of Diabetes. Standards of care in Diabetes-2025. Diabetes Care. 2025;48:S27–49. 10.2337/dc25-S002. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Worldwide trends in diabetes prevalence and treatment. From 1990 to 2022: a pooled analysis of 1108 population-representative studies with 141 million participants. Lancet. 2024;404:2077–93. 10.1016/s0140-6736(24)02317-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Rayego-Mateos S, Morgado-Pascual JL, Opazo-Ríos L, Guerrero-Hue M, García-Caballero C, Vázquez-Carballo C, et al. Pathogenic pathways and therapeutic approaches targeting inflammation in diabetic nephropathy. Int J Mol Sci. 2020. 10.3390/ijms21113798. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Wong TY, Sun J, Kawasaki R, Ruamviboonsuk P, Gupta N, Lansingh VC, et al. Guidelines on diabetic eye care: the international council of ophthalmology recommendations for screening, follow-up, referral, and treatment based on resource settings. Ophthalmology. 2018;125:1608–22. 10.1016/j.ophtha.2018.04.007. [DOI] [PubMed] [Google Scholar]
- 5.Cole JB, Florez JC. Genetics of diabetes mellitus and diabetes complications. Nat Rev Nephrol. 2020;16:377–90. 10.1038/s41581-020-0278-5. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Marx N, Federici M, Schütt K, Müller-Wieland D, Ajjan RA, Antunes MJ, et al. 2023 ESC guidelines for the management of cardiovascular disease in patients with diabetes. Eur Heart J. 2023;44:4043–140. 10.1093/eurheartj/ehad192. [DOI] [PubMed] [Google Scholar]
- 7.Bebu I, Schade D, Braffett B, Kosiborod M, Lopes-Virella M, Soliman EZ, et al. Risk factors for first and subsequent CVD events in type 1 diabetes: the DCCT/EDIC study. Diabetes Care. 2020;43:867–74. 10.2337/dc19-2292. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Yousri NA, Suhre K, Yassin E, Al-Shakaki A, Robay A, Elshafei M, et al. Metabolic and metabo-clinical signatures of type 2 Diabetes, obesity, retinopathy, and dyslipidemia. Diabetes. 2022;71:184–205. 10.2337/db21-0490. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Arnett DK, Blumenthal RS, Albert MA, Buroker AB, Goldberger ZD, Hahn EJ, et al. 2019 ACC/AHA guideline on the primary prevention of cardiovascular disease: a report of the American college of Cardiology/American heart association task force on clinical practice guidelines. Circulation. 2019;140:e596-646. 10.1161/cir.0000000000000678. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Baigent C, Blackwell L, Emberson J, Holland LE, Reith C, Bhala N, et al. Efficacy and safety of more intensive Lowering of LDL cholesterol: a meta-analysis of data from 170,000 participants in 26 randomised trials. Lancet. 2010;376:1670–81. 10.1016/s0140-6736(10)61350-5. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Stürzebecher PE, Katzmann JL, Laufs U. What is “remnant cholesterol”? Eur Heart J. 2023;44:1446–8. 10.1093/eurheartj/ehac783. [DOI] [PubMed] [Google Scholar]
- 12.Castañer O, Pintó X, Subirana I, Amor AJ, Ros E, Hernáez Á, et al. Remnant Cholesterol, not LDL cholesterol, is associated with incident cardiovascular disease. J Am Coll Cardiol. 2020;76:2712–24. 10.1016/j.jacc.2020.10.008. [DOI] [PubMed] [Google Scholar]
- 13.Yang PT, Tang L, Yang SQ, Shi QL, Wang YQ, Qin YX, et al. Remnant cholesterol trajectory and subclinical arteriosclerosis: a 10-year longitudinal study of Chinese adults. Sci Rep. 2024;14:9037. 10.1038/s41598-024-59173-6. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Burnett JR, Hooper AJ, Hegele RA. Remnant Cholesterol and atherosclerotic cardiovascular disease risk. J Am Coll Cardiol. 2020;76:2736–9. 10.1016/j.jacc.2020.10.029. [DOI] [PubMed] [Google Scholar]
- 15.Liu H, Hu W, Li X, Hu F, Xi Y, Su Z, et al. Do perfluoroalkyl substances aggravate the occurrence of obesity-associated glucolipid metabolic disease? Environ Res. 2021;202:111724. 10.1016/j.envres.2021.111724. [DOI] [PubMed] [Google Scholar]
- 16.Hu X, Liu Q, Guo X, Wang W, Yu B, Liang B, et al. The role of remnant cholesterol beyond low-density lipoprotein cholesterol in diabetes mellitus. Cardiovasc Diabetol. 2022;21:117. 10.1186/s12933-022-01554-0. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Zhou Y, Madsen JM, Özbek BT, Køber L, Bang LE, Lønborg JT, et al. The role of remnant cholesterol in patients with ST-segment elevation myocardial infarction. Eur J Prev Cardiol. 2024;31:1227–37. 10.1093/eurjpc/zwae102. [DOI] [PubMed] [Google Scholar]
- 18.Wadström BN, Wulff AB, Pedersen KM, Nordestgaard BG. Small remnants versus large triglyceride-rich lipoproteins in risk of atherosclerotic cardiovascular disease. Clin Chem. 2025;71:463–73. 10.1093/clinchem/hvae222. [DOI] [PubMed] [Google Scholar]
- 19.Wadström BN, Pedersen KM, Wulff AB, Nordestgaard BG. Elevated remnant cholesterol and atherosclerotic cardiovascular disease in diabetes: a population-based prospective cohort study. Diabetologia. 2023;66:2238–49. 10.1007/s00125-023-06016-0. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Huh JH, Roh E, Lee SJ, Ihm SH, Han KD, Kang JG. Remnant cholesterol is an independent predictor of type 2 diabetes: a nationwide population-based cohort study. Diabetes Care. 2023;46:305–12. 10.2337/dc22-1550. [DOI] [PubMed] [Google Scholar]
- 21.Meng X, Xu H, Liang Y, Liang M, Song W, Zhou B, et al. Enhanced CT-based radiomics model to predict natural killer cell infiltration and clinical prognosis in non-small cell lung cancer. Front Immunol. 2023;14:1334886. 10.3389/fimmu.2023.1334886. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Jansson Sigfrids F, Dahlström EH, Forsblom C, Sandholm N, Harjutsalo V, Taskinen MR, et al. Remnant cholesterol predicts progression of diabetic nephropathy and retinopathy in type 1 diabetes. J Intern Med. 2021;290:632–45. 10.1111/joim.13298. [DOI] [PubMed] [Google Scholar]
- 23.Cui C, Li P, Qi Y, Song J, Han T, Shang X, et al. Intraindividual discordance between remnant cholesterol and low-density lipoprotein cholesterol associated with incident stroke: results from 2 national cohorts. J Am Heart Assoc. 2024;13:e035764. 10.1161/jaha.124.035764. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Wang J, Liu Y, Jin R, Zhao X, Wu Z, Han Z, et al. Intraindividual difference in estimated GFR by creatinine and cystatin C, cognitive trajectories and motoric cognitive risk syndrome. Nephrol Dial Transplant. 2024;39:860–72. 10.1093/ndt/gfad234. [DOI] [PubMed] [Google Scholar]
- 25.Quispe R, Martin SS, Michos ED, Lamba I, Blumenthal RS, Saeed A, et al. Remnant cholesterol predicts cardiovascular disease beyond LDL and ApoB: a primary prevention study. Eur Heart J. 2021;42:4324–32. 10.1093/eurheartj/ehab432. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Wu Z, Wang J, Zhang H, Pan H, Li Z, Liu Y, et al. Longitudinal association of remnant cholesterol with joint arteriosclerosis and atherosclerosis progression beyond LDL cholesterol. BMC Med. 2023;21:42. 10.1186/s12916-023-02733-w. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Wu S, Su X, Zuo Y, Chen S, Tian X, Xu Q, et al. Discordance between remnant cholesterol and low-density lipoprotein cholesterol predicts arterial stiffness progression. Hellenic J Cardiol. 2023;74:24–31. 10.1016/j.hjc.2023.05.008. [DOI] [PubMed] [Google Scholar]
- 28.Wan Z, Guo J, Pan A, Chen C, Liu L, Liu G. Association of serum 25-Hydroxyvitamin D concentrations with all-cause and cause-specific mortality among individuals with diabetes. Diabetes Care. 2021;44:350–7. 10.2337/dc20-1485. [DOI] [PubMed] [Google Scholar]
- 29.Afkarian M, Zelnick LR, Hall YN, Heagerty PJ, Tuttle K, Weiss NS, et al. Clinical manifestations of kidney disease among US adults with diabetes, 1988–2014. JAMA. 2016;316:602–10. 10.1001/jama.2016.10924. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.Zhang Y, Liu X, Zhang X, Li L, Li Q, Geng H, et al. Association between serum heavy metal levels and diabetic retinopathy in NHANES 2011–2020. Sci Rep. 2024;14:1268. 10.1038/s41598-024-51749-6. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.Dang K, Wang X, Hu J, Zhang Y, Cheng L, Qi X, et al. The association between triglyceride-glucose index and its combination with obesity indicators and cardiovascular disease: NHANES 2003–2018. Cardiovasc Diabetol. 2024;23:8. 10.1186/s12933-023-02115-9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.Unger T, Borghi C, Charchar F, Khan NA, Poulter NR, Prabhakaran D, et al. 2020 international society of hypertension global hypertension practice guidelines. Hypertension. 2020;75:1334–57. 10.1161/hypertensionaha.120.15026. [DOI] [PubMed] [Google Scholar]
- 33.Shuken SR, McNerney MW. Costs and benefits of popular P-value correction methods in three models of quantitative omic experiments. Anal Chem. 2023;95:2732–40. 10.1021/acs.analchem.2c03719. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34.Sajja A, Park J, Sathiyakumar V, Varghese B, Pallazola VA, Marvel FA, et al. Comparison of methods to estimate low-density lipoprotein cholesterol in patients with high triglyceride levels. JAMA Netw Open. 2021;4:e2128817. 10.1001/jamanetworkopen.2021.28817. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35.Kraaijenhof JM, Kerkvliet MJ, Nurmohamed NS, Grefhorst A, Kroon J, Wareham NJ, et al. The role of systemic inflammation in remnant cholesterol associated cardiovascular risk: insights from the EPIC-Norfolk study. Eur J Prev Cardiol. 2025. 10.1093/eurjpc/zwaf037. [DOI] [PubMed] [Google Scholar]
- 36.Han M, Huang K, Shen C, Hu H, Liu F, Li J, et al. Discordant high remnant cholesterol with LDL-C increases the risk of stroke: a Chinese prospective cohort study. Stroke. 2024;55:2066–74. 10.1161/strokeaha.124.046811. [DOI] [PubMed] [Google Scholar]
- 37.Talebi S, Zeraattalab-Motlagh S, Rahimlou M, Naeini F, Ranjbar M, Talebi A, et al. The association between total protein, animal protein, and animal protein sources with risk of inflammatory bowel diseases: a systematic review and meta-analysis of cohort studies. Adv Nutr. 2023;14:752–61. 10.1016/j.advnut.2023.05.008. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 38.Morvaridi M, Rayyani E, Jaafari M, Khiabani A, Rahimlou M. The effect of green coffee extract supplementation on cardio metabolic risk factors: a systematic review and meta-analysis of randomized controlled trials. J Diabetes Metab Disord. 2020;19:645–60. 10.1007/s40200-020-00536-x. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 39.Jang SY, Kang M, Song E, Jang A, Choi KM, Baik SH, et al. Remnant cholesterol is an independent risk factor for the incidence of chronic kidney disease in newly-diagnosed type 2 diabetes: a nationwide population-based study. Diabetes Res Clin Pract. 2024;210:111639. 10.1016/j.diabres.2024.111639. [DOI] [PubMed] [Google Scholar]
- 40.Karakasis P, Patoulias D, Rizzo M, Fragakis N, Mantzoros CS. Association between remnant cholesterol and chronic kidney disease: systematic review and meta-analysis. Diabetes Obes Metab. 2025;27:2573–83. 10.1111/dom.16258. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 41.Chen ZY, Liu SN, Li CN, Sun SJ, Liu Q, Lei L, et al. Atorvastatin helps preserve pancreatic β cell function in obese C57BL/6 J mice and the effect is related to increased pancreas proliferation and amelioration of endoplasmic-reticulum stress. Lipids Health Dis. 2014;13:98. 10.1186/1476-511x-13-98. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 42.Sokooti S, Flores-Guerrero JL, Heerspink HJL, Connelly MA, Bakker SJL, Dullaart RPF. Triglyceride-rich lipoprotein and LDL particle subfractions and their association with incident type 2 diabetes: the PREVEND study. Cardiovasc Diabetol. 2021;20:156. 10.1186/s12933-021-01348-w. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 43.Mousavi SN, Rayyani E, Heshmati J, Tavasolian R, Rahimlou M. Effects of Ramadan and non-ramadan intermittent fasting on gut microbiome. Front Nutr. 2022;9:860575. 10.3389/fnut.2022.860575. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 44.Yu D, Wang Z, Zhang X, Qu B, Cai Y, Ma S, et al. Remnant cholesterol and cardiovascular mortality in patients with type 2 diabetes and incident diabetic nephropathy. J Clin Endocrinol Metab. 2021;106:3546–54. 10.1210/clinem/dgab533. [DOI] [PubMed] [Google Scholar]
- 45.Wang AY, Lam CW, Wang M, Woo J, Chan IH, Lui SF, et al. Circulating soluble vascular cell adhesion molecule 1: relationships with residual renal function, cardiac hypertrophy, and outcome of peritoneal dialysis patients. Am J Kidney Dis. 2005;45:715–29. 10.1053/j.ajkd.2004.12.012. [DOI] [PubMed] [Google Scholar]
- 46.Xu D, Xie L, Cheng C, Xue F, Sun C. Triglyceride-rich lipoproteins and cardiovascular diseases. Front Endocrinol (Lausanne). 2024;15:1409653. 10.3389/fendo.2024.1409653. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 47.Varbo A, Benn M, Nordestgaard BG. Remnant cholesterol as a cause of ischemic heart disease: evidence, definition, measurement, atherogenicity, high risk patients, and present and future treatment. Pharmacol Ther. 2014;141:358–67. 10.1016/j.pharmthera.2013.11.008. [DOI] [PubMed] [Google Scholar]
- 48.Gallo G, Volpe M, Savoia C. Endothelial dysfunction in hypertension: current concepts and clinical implications. Front Med (Lausanne). 2021;8:798958. 10.3389/fmed.2021.798958. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 49.Liu T, Krentz A, Lu L, Curcin V. Machine learning based prediction models for cardiovascular disease risk using electronic health records data: systematic review and meta-analysis. Eur Heart J Digit Health. 2025;6:7–22. 10.1093/ehjdh/ztae080. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 50.Zhang Z, Yang L, Han W, Wu Y, Zhang L, Gao C, et al. Machine learning prediction models for gestational diabetes mellitus: meta-analysis. J Med Internet Res. 2022;24:e26634. 10.2196/26634. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
The datasets generated or analyzed during the study are available from the corresponding author on reasonable request.