Remnant cholesterol and body mass index jointly associate with cardiometabolic Multimorbidity with mediation analysis in CHARLS a prospective cohort study

Abstract

Remnant cholesterol (RC) has emerged as a novel lipid parameter reflecting the residual risk of atherogenic dyslipidemia beyond low-density lipoprotein cholesterol (LDL-C). Previous studies have demonstrated that elevated RC and increased body mass index (BMI) are each independently associated with cardiometabolic diseases such as diabetes and cardiovascular disorders. However, whether RC interacts with BMI and how their combined influence contributes to the development of cardiometabolic multimorbidity (CMM)—the coexistence of multiple cardiometabolic conditions—remain unclear. Clarifying these relationships may provide new insights into the metabolic mechanisms underlying multimorbidity and improve early risk stratification. This longitudinal cohort study used data from 6,646 adults aged ≥ 45 years in the China Health and Retirement Longitudinal Study (CHARLS). Cardiometabolic multimorbidity (CMM) was defined as having two or more physician-diagnosed conditions among heart disease, stroke, and diabetes. Associations of remnant cholesterol (RC) and body mass index (BMI) with CMM were examined using restricted cubic spline, multivariable regression, interaction, and bidirectional mediation analyses, adjusting for demographic, lifestyle, and metabolic factors. A total of 6,646 participants (mean age 58.52 years) were included. Both RC and BMI were significant independent risk factors for CMM with combined, with a joint association. Elevated RC (≥ median) significantly increased the risk of CMM (adjusted OR = 1.35, p = 0.003), whereas BMI showed a dose-dependent risk pattern, with adjusted ORs (95% CIs) of 1.32 (1.10–1.58) for overweight and 1.78 (1.42–2.22) for obesity compared with normal weight (p < 0.01). The combined effect of high RC and high BMI significantly increased CMM risk (adjusted OR = 2.23, p < 0.0001). In adjusted mediation models, BMI accounted for 32.6% of the total association of RC on CMM (β = 0.00021, P < 0.001), whereas RC accounted for 6.9% of the total effect of BMI on CMM (β = 0.00005, P = 0.016). This study suggests that elevated remnant cholesterol and higher BMI are jointly associated with increased cardiometabolic multimorbidity risk, underscoring the importance of considering lipid-related and metabolic factors together in risk stratification.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-025-25958-6.

Introduction

Cardiometabolic multimorbidity (CMM), defined as the coexistence of diabetes, heart disease, and stroke, has emerged as a major global health concern due to its increasing prevalence and complex pathogenic mechanisms¹. Recent large-scale cohort studies have reported that the prevalence of CMM has risen substantially worldwide, affecting approximately 3–6% of the adult population and posing substantial challenges to healthcare systems^2,3. Cardiovascular and metabolic disorders together account for nearly 31% of global deaths according to the World Health Organization, highlighting the enormous public health burden caused by these interrelated conditions⁴. Given the shared metabolic and inflammatory pathways among these diseases, understanding modifiable risk factors and their interactions is crucial for early prevention and management of CMM.

Disorders of lipid metabolism are widely recognized as important risk factors for cardiovascular and metabolic diseases. Among lipid fractions, remnant cholesterol (RC)—the cholesterol content of triglyceride-rich lipoproteins including very-low-density lipoproteins (VLDL), intermediate-density lipoproteins (IDL), and chylomicron remnants—is calculated as total cholesterol minus LDL-C and HDL-C (RC = TC − LDL-C − HDL-C)^5–75–7. Although some previous reports have used the term “residual cholesterol” interchangeably with RC, these concepts are distinct. Residual cholesterol typically refers to cholesterol that remains after lipid-lowering therapy, whereas RC represents a directly calculated biochemical parameter that reflects the atherogenic lipid fraction independent of treatment status. Transitioning from LDL-C to RC provides a logical extension of lipid-related risk assessment, as elevated RC has been linked to persistent cardiovascular and metabolic risk even when LDL-C is well controlled⁸, mainly through vascular infiltration, macrophage-mediated inflammation^9,10, and insulin resistance^5,11. These observations highlight RC as a complementary lipid marker that bridges dyslipidemia and cardiometabolic multimorbidity (CMM).

Body Mass Index (BMI) is another key indicator of adiposity and metabolic health. While higher BMI generally increases the risk of diabetes¹² and cardiovascular disease¹³, the association is not strictly linear¹⁴ and may vary with fat distribution and inflammatory status. Abdominal obesity is closely related to insulin resistance, whereas peripheral fat accumulation exerts less metabolic burden. Considering the well-established Asian-specific thresholds (24.0 kg/m² for overweight and 28.0 kg/m² for obesity)¹⁵, we primarily analyzed BMI as a categorical variable to enhance clinical interpretability and comparability with public-health guidelines. At the same time, restricted cubic-spline (RCS) models were applied to verify potential non-linear dose–response patterns across the full BMI range, ensuring that categorization did not obscure underlying trends.

Although it has been shown that elevated RC^16,17 as well as BMI¹⁷ have independent predictive effects on the development of cardiometabolic multimorbidity (CMM), they capture different yet complementary aspects of metabolic risk. RC reflects the atherogenic lipid burden and remnant inflammation from triglyceride-rich lipoproteins^9,10,18, while BMI indicates overall adiposity and its associated systemic metabolic burden. Recent Mendelian randomization studies have provided causal evidence linking elevated RC or triglycerides to CMM risk¹⁶, supporting their independent metabolic roles. Nevertheless, limited evidence has explored how RC and BMI may jointly contribute to the development of CMM and interact through shared biological mechanisms—particularly insulin resistance and chronic inflammation—which are central to both lipid metabolism and adiposity-related metabolic dysfunctions^19–21. Therefore, exploring their combined and bidirectional effects may provide a more comprehensive understanding of metabolic health and CMM risk than evaluating either marker alone. Based on data from the China Health and Retirement Longitudinal Study (CHARLS), we systematically assessed the independent and joint associations of RC and BMI with CMM risk and used mediation analysis to explore their potential interrelations, acknowledging that causal inference cannot be established in this observational study. These findings may offer scientific insights for integrated risk assessment and targeted prevention strategies for CMM. Few studies have quantified whether adiposity mediates the RC–CMM association and, conversely, whether remnant lipoprotein-related dyslipidemia mediates the BMI–CMM pathway. By jointly evaluating bidirectional mediation and additive joint effects in a nationally representative cohort, our study extends current evidence beyond single-risk-factor models.

Methods

Study design and population

The China Health and Retirement Longitudinal Study (CHARLS) is a longitudinal survey study initiated and managed by the National Development Research Institute of Peking University (NDRI), focusing on China’s household-based population aged 45 years or older. The survey began in 2011, covering 28 provinces, with representative samples obtained through a multi-stage, stratified, probability sampling method. CHARLS is an ongoing survey study, examined at intervals of 2–3 years, and focuses on the collection of basic demographic information, personal health status, healthcare service utilization, economic status, and social support. To date, CHARLS has released four waves of survey data, in 2011, 2013, 2015, and 2018. Previous literature has described CHARLS in detail²². CHARLS was conducted in accordance with the principles of the Declaration of Helsinki, and the survey project was approved by the Biomedical Ethics Committee of Peking University (IRB00001052-11015), and all participants were required to sign an informed consent form.

Data from 2011 to 2018 were used in this study as the baseline and follow-up waves, respectively. The total sample size at baseline (2011) was 19,752 participants. Exclusions were made according to the following criteria: (1) missing follow-up CMM data (n = 12075); (2) missing information on sex (n = 6) or education (n = 2) or Hypertension data (n = 34) or other chronic disease variables (n = 75); (3) having cardiometabolic multimorbidity (CMM) at baseline (2011) (n = 198); (4) outliers based on the interquartile range (IQR) method (n = 391). After applying these exclusion criteria, a total of 6,646 participants without pre-existing CMM at baseline were included in the analysis. The detailed inclusion and exclusion process is shown in Fig. 1. Because the CHARLS dataset does not record the exact onset time of each disease, time-to-event analyses (e.g., Cox regression) were not feasible. Therefore, we used multivariable logistic regression to estimate odds ratios (ORs) and 95% confidence intervals (CIs) for incident CMM during the 7-year follow-up (2011–2018).

Fig. 1.

Flow chart for inclusion of this study population.

Variables and outcomes

Remnant cholesterol (mg/dL) was defined as total cholesterol (TC) (mg/dL) minus LDL-C (mg/dL) minus HDL-C (mg/dL)^20,21,23,24. Although there is no standardized definition, this formula has been used in previous studies. Body mass index (BMI) is used to assess the obesity of an individual by dividing the weight in kilograms by the height in meters¹⁵. It is categorized as normal weight (BMI < 24.0 kg/m2), overweight (BMI ≥ 24.0 kg/m2), and obese (BMI ≥ 28.0 kg/m2) according to the Asian population classification criteria²⁴. To evaluate the combined metabolic effects of lipid and adiposity burden, two grouping strategies were adopted. First, RC was dichotomized at the sample median (low vs. high) and cross-classified with BMI (< 24.0 vs. ≥24.0 kg/m²) to create four categories, allowing assessment of the overall combined effect between RC and overweight status. Second, to explore potential dose–response trends, BMI was subdivided into three categories—normal, overweight, and obese—and cross-classified with RC (low vs. high) to generate six combined groups. This dual-grouping design enabled both the evaluation of interactive effects and the gradient association of RC–BMI combinations with cardiometabolic multimorbidity (CMM) risk. Following previous CHARLS studies, three physician-diagnosed conditions were used to ascertain CMM^16,25: diabetes or high blood sugar, heart disease (including coronary heart disease, angina, myocardial infarction, or heart failure), and stroke. Heart disease and stroke were identified based on participants’ self-reported physician diagnoses in response to the questions ‘Has a doctor ever told you that you have a heart problem such as heart attack, coronary heart disease, angina, congestive heart failure, or other heart diseases?’ and ‘Have you ever been diagnosed with stroke by a doctor?’ respectively. Diabetes was determined by self-reported physician diagnosis, medication use, or laboratory criteria (fasting plasma glucose ≥ 7.0 mmol/L or HbA1c ≥ 6.5%)^26,27. Individuals were considered to have CMM if two or more of these conditions were present.

The distribution of remnant cholesterol (RC) was examined using histograms and violin plots. As shown in Supplementary Figures S1 and S2, the raw RC data were markedly right-skewed, while the cleaned RC data after outlier removal by the interquartile range (IQR) method approximated a normal distribution but remained mildly skewed. Therefore, RC was dichotomized by the median value for subsequent analyses.

Covariates

Consistent with the covariates included in most previous studies, sociodemographic factors included gender, age, marital status, education, and place of residence; other variables included smoking, alcohol consumption, sleep duration, and Hypertension, dyslipidemia, and laboratory data (e.g., Glucose, Glycated Hemoglobin, Creatinine, Blood Urea Nitrogen, Total Cholesterol, High-Density Lipoprotein Cholesterol, Low-Density Lipoprotein Cholesterol, Triglycerides). Marital status was categorized as married and unmarried; educational attainment was categorized as No formal education, Primary school, Secondary school (including middle, high, and vocational education), and Higher education (college or above); residence was categorized as urban and rural; smoking and drinking status was categorized as no or yes, respectively; and self-reported The diagnosis of hypertension was based on self-reported doctor’s diagnosis (positive response to “Have you been diagnosed with hypertension?”) and/or recent use of anti-hypertensive medications (positive response to “Are you currently taking any anti-hypertensive medications to treat or control your blood pressure? blood pressure”) and/or systolic/diastolic blood pressure (SBP/DBP) ≥ 140/90 mmHg²⁷. Blood pressure was measured by trained interviewers using a validated electronic sphygmomanometer, and the average of three consecutive readings taken after a 5-minute rest was used for analysis.

Statistical methods

Participants’ baseline characteristics were analyzed using descriptive statistics. Continuous variables were presented as mean and standard deviation (SD), while categorical variables were presented using frequencies and percentages. Baseline characteristics were compared using one-way ANOVA or χ² tests as appropriate. Since RC has larger and smaller values, to prevent bias in the results, outliers were removed in this study based on the interquartile range (IQR) method. The RC was categorized into two groups based on the cleaned median RC (median = 18.56 mg/dL), and BMI was divided into high and low categories using 24 as the cutoff point. Using this categorization method, participants were further divided into four groups by combining the categories pairwise (Q1: RC < median & BMI < 24; Q2: RC < median & BMI ≥ 24; Q3: RC ≥ median & BMI < 24; Q4:RC ≥ median & BMI ≥ 24).

To assess the association between RC, BMI and CMM, we first performed multivariable-adjusted multivariable logistic regression models to estimate odds ratios (OR) and 95% confidence intervals (CI) to observe the associations of RC and BMI with CMM, respectively. Covariates were selected based on clinical relevance and prior literature. Variables including demographic (age, sex, residence, marital status, education), behavioral (smoking, drinking), and metabolic factors (hypertension, treatment for dyslipidemia) were adjusted for in the multivariable models. Multicollinearity was assessed using generalized variance inflation factors (GVIFs), and all adjusted GVIF values (GVIF^(1/(2×Df))) were < 2.0, indicating no significant collinearity (Supplementary Table S1). Having established independent associations between RC, BMI and CMM, we further assessed RC and BMI jointly using the same methodology by categorizing participants into four categories as described above. To further explore the joint effect of RC and BMI on CMM, additive interaction was evaluated using three indices: the relative excess risk due to interaction (RERI), the attributable proportion (AP), and the synergy index (S). The 95% confidence intervals were estimated using the delta method. Dose-response associations of RC and BMI levels with CMM were shown using restricted cubic spline functions at the 10th, 50th, and 90th percentiles using 3 sections. We also performed stratified analyses in age, sex, smoking, and drinking status. Additionally, we performed mediation analyses in unadjusted models as well as in adjusted full models to assess the direct and indirect relationships between RC and BMI and CMM. To facilitate joint analyses, residual cholesterol (RC) and BMI were dichotomized using data-driven cutoffs (18.6 mg/dL and 23.1 kg/m², respectively) derived from the inflection points of restricted cubic spline (RCS) curves. These cutoffs closely approximated the sample medians and provided optimal subgroup balance and discrimination for interaction testing. The dose–response patterns observed in RCS plots further supported the use of these data-driven thresholds. Subgroup analyses were performed according to medication status. Subgroup 1 included participants not receiving antihypertensive or lipid-lowering therapy at baseline (n = 5396). Subgroup 2 excluded those receiving such therapy either at baseline or during follow-up (n = 3980). All statistical analyses were performed using R software (version 4.4.1). Mediation analyses were conducted using the mediation package in R, with 5,000 bootstrap resamples to obtain bias-corrected 95% confidence intervals²⁸. RC and BMI were alternately treated as the exposure and mediator in two models, each adjusted for the same covariates as in the main regression analyses. The mediation proportion was calculated as the ratio of the indirect to total effect. A two-sided test with a p-value < 0.05 was defined as statistically significant.

Results

Baseline characteristics of participants

This study included 6646 participants with a mean age of 58.52 years and 54.29% female. Participants were categorized into four groups based on RC and BMI levels: RC < median & BMI < 24 (2306 individuals), RC < median & BMI ≥ 24 (1064 individuals), RC ≥ median & BMI < 24 (1718 individuals), and RC ≥ median & BMI ≥ 24 (1558 individuals), with significant differences in demographic characteristics as well as laboratory data between the four groups. Detailed baseline characteristics of the participants are shown in Table 1. As shown in Supplementary Figures S1 and S2, the raw remnant cholesterol (RC) distribution was highly right-skewed, whereas the cleaned RC data became approximately normal with mild skewness. Violin and box plots further confirmed consistent distributional patterns of RC across BMI categories and RC × BMI subgroups. Overall, the high RC high BMI group was younger (57.31 ± 8.15), had a higher proportion of females (63.16%), and lived in a more urban area (59.63%), and had a significantly higher HbA1c, Glucose, TC, LDL, and TG compared to the other three groups (p < 0.0001), whereas HDL was significantly decreased (p < 0.0001). Among CMM patients, the highest incidence was found in the high RC high BMI group (11.49%), while it was only 4.12% in the low RC low BMI group.

Table 1.

Baseline characteristics of articipants.

Variable	Total (n = 6646)	Q1 (n = 2306)	Q2 (n = 1064)	Q3 (n = 1718)	Q4 (n = 1558)	P value
Age	58.52 ± 8.75	59.66 ± 9.14	56.67 ± 8.16	59.23 ± 8.77	57.31 ± 8.15	< 0.0001
Sex						< 0.0001
female	3608(54.29)	1077(46.70)	680(63.91)	867(50.47)	984(63.16)
male	3038(45.71)	1229(53.30)	384(36.09)	851(49.53)	574(36.84)
Marital						< 0.0001
Married	5957(89.63)	2016(87.42)	974(91.54)	1514(88.13)	1453(93.26)
Non-Married	689(10.37)	290(12.58)	90(8.46)	204(11.87)	105(6.74)
Education						< 0.0001
Higher education	80(1.20)	22(0.95)	21(1.97)	19(1.11)	18(1.16)
Secondary school	1933(29.09)	609(26.41)	366(34.40)	456(26.54)	502(32.22)
No formal education	3105(46.72)	1126(48.83)	458(43.05)	848(49.36)	673(43.20)
Primary school	1528(22.99)	549(23.81)	219(20.58)	395(22.99)	365(23.43)
Residence						< 0.0001
rural	4505(67.79)	1691(73.33)	668(62.78)	1217(70.84)	929(59.63)
urban	2141(32.21)	615(26.67)	396(37.22)	501(29.16)	629(40.37)
Smoke						< 0.0001
no	4627(69.62)	1464(63.49)	863(81.11)	1072(62.40)	1228(78.82)
yes	2019(30.38)	842(36.51)	201(18.89)	646(37.60)	330(21.18)
Drink						< 0.0001
no	4457(67.06)	1465(63.53)	757(71.15)	1102(64.14)	1133(72.72)
yes	2189(32.94)	841(36.47)	307(28.85)	616(35.86)	425(27.28)
Glucose	106.56 ± 30.37	102.50 ± 23.07	104.41 ± 24.40	108.02 ± 35.78	112.45 ± 35.58	< 0.0001
HbA1c	5.21 ± 0.70	5.13 ± 0.60	5.20 ± 0.63	5.20 ± 0.73	5.36 ± 0.83	< 0.0001
Creatinine	0.77 ± 0.18	0.78 ± 0.19	0.74 ± 0.16	0.78 ± 0.18	0.78 ± 0.18	< 0.0001
Urea	15.68 ± 4.40	16.16 ± 4.58	15.62 ± 4.24	15.53 ± 4.46	15.17 ± 4.08	< 0.0001
TC	191.95 ± 36.75	185.37 ± 35.25	188.86 ± 34.82	195.99 ± 37.13	199.36 ± 37.87	< 0.0001
HDL	52.55 ± 14.79	59.74 ± 15.02	53.23 ± 12.60	50.15 ± 13.98	44.09 ± 10.97	< 0.0001
LDL	118.40 ± 33.17	115.55 ± 31.63	124.26 ± 32.07	115.52 ± 33.80	121.79 ± 34.58	< 0.0001
TG	113.82 ± 55.52	76.63 ± 24.71	86.34 ± 27.97	138.37 ± 53.38	160.57 ± 56.22	< 0.0001
DM						< 0.0001
no	5830(87.72)	2149(93.19)	905(85.06)	1534(89.29)	1242(79.72)
yes	816(12.28)	157(6.81)	159(14.94)	184(10.71)	316(20.28)
Hypertension						< 0.0001
no	5174(77.85)	1971(85.47)	753(70.77)	1422(82.77)	1028(65.98)
yes	1472(22.15)	335(14.53)	311(29.23)	296(17.23)	530(34.02)
Dyslipidemia						< 0.0001
no	6172(92.87)	2225(96.49)	957(89.94)	1639(95.40)	1351(86.71)
yes	474(7.13)	81(3.51)	107(10.06)	79(4.60)	207(13.29)
Heart disease						< 0.0001
no	5233(78.74)	1906(82.65)	795(74.72)	1386(80.68)	1146(73.56)
yes	1413(21.26)	400(17.35)	269(25.28)	332(19.32)	412(26.44)
Stroke						< 0.01
no	6115(92.01)	2155(93.45)	971(91.26)	1586(92.32)	1403(90.05)
yes	531(7.99)	151(6.55)	93(8.74)	132(7.68)	155(9.95)
CMM						< 0.0001
no	6183(93.03)	2211(95.88)	970(91.17)	1623(94.47)	1379(88.51)
yes	463(6.97)	95(4.12)	94(8.83)	95(5.53)	179(11.49)

Continuous data are presented as mean (SD) and categorical data as percentage (%).

Divide the population into four groups:.

Q1: RC < median & BMI < 24; Q2: RC < median & BMI ≥ 24; Q3: RC ≥ median & BMI < 24; Q4:RC ≥ median & BMI ≥ 24.

Abbreviation: RC, Remnant Cholesterol; BMI, Body Mass Index; HbA1c, Glycated Hemoglobin; Urea, Blood Urea Nitrogen; TC, Total Cholesterol; HDL, High-Density Lipoprotein Cholesterol; LDL, Low-Density Lipoprotein Cholesterol; TG, Triglycerides; DM, Diabetes Mellitus; CMM, Cardiometabolic multimorbidity.

Relationship between RC, BMI and CMM

Restricted cubic spline (RCS) analyses demonstrated significant overall associations between RC and BMI and CMM risk (P < 0.001), with evidence of non-linearity (P for non-linearity = 0.0266 for RC and 0.0028 for BMI) (Fig. 2). The risk of CMM tended to increase significantly with increasing levels of RC and BMI, and Table 2 shows the independent effects of RC and BMI on the risk of CMM after adjusting for the confounding variables of age, sex, place of residence, marital status, education, smoking and drinking status, hypertension, and the use of lipid-lowering medications. Specifically, after grouping RC at the median and BMI at a cutoff of 24 and combining these, the combination of RC ≥ median and BMI ≥ 24 significantly increased the risk of CMM compared with the group with RC below median and BMI below 24 (OR = 2.23, 95% CI: 1.69–2.94, P < 0.0001). Table S4 Sensitivity analyses were performed in which Subgroup 1 included individuals who did not receive antihypertensive or lipid-lowering therapy at baseline, whereas Subgroup 2 included individuals who did not receive any of these therapies both at baseline and during follow-up. The results showed a consistent overall risk trend after adjusting for the same variables. In addition, Table 3 adjusting for the same covariates, it was observed that the group with RC ≥ median and BMI ≥ 28 had a significantly higher risk of CMM compared to the group with low RC and the group with low BMI (OR = 2.62, 95% CI: 1.84–3.72, p < 0.0001). These results demonstrate that the combination of high RC and high BMI substantially increases CMM risk, indicating a higher concurrent risk rather than statistically significant supra-additivity. However, the additive interaction analysis showed that the coexistence of elevated RC and obesity (BMI ≥ 28) did not exert a statistically significant supra-additive effect on CMM risk (RERI = 0.31, AP = 0.08, S = 0.06; all p > 0.05; Table S3).

Fig. 2.

Dose-response relationship between RC and BMI and CMM risk. RC, remnant cholesterol; RCS, restricted cubic spline. Panels a–b illustrate the association between RC (Remnant Cholesterol) and BMI (Body Mass Index) with the risk of CMM (Cardiometabolic multimorbidity) using restricted cubic spline (RCS) models. Panel a depicts the relationship between RC and CMM risk, while panel b shows the relationship between BMI and CMM risk. The trends are presented as odds ratios (OR) with 95% confidence intervals. The model adjusted for age, sex, residence, marital, education, smoke, drink, hypertension, treatments for dyslipidemia.

Table 2.

Joint effects of RC and BMI (≥ 24 kg/m²) on CMM risk.

character	Unadjusted		Adjusted
	OR(95%CI)	P value	OR(95%CI)	P value
RC < median	ref		ref
RC ≥ median	1.54(1.27,1.86)	< 0.0001	1.35(1.11,1.65)	0.003
BMI < 24	ref		ref
BMI ≥ 24	2.35(1.94,2.84)	< 0.0001	1.83(1.48,2.26)	< 0.0001
RC < median & BMI < 24	ref		ref
RC < median & BMI ≥ 24	2.26(1.68,3.03)	< 0.0001	1.83(1.34,2.50)	< 0.001
RC ≥ median & BMI < 24	1.36(1.02,1.82)	0.04	1.3(0.96,1.74)	0.09
RC ≥ median & BMI ≥ 24	3.02(2.33,3.91)	< 0.0001	2.23(1.69,2.94)	< 0.0001
p for trend		< 0.0001		< 0.0001

BMI was dichotomized at 24 kg/m² to define normal weight and overweight groups, creating four RC×BMI joint categories.

Crude model unadjusted for covariates.

Adjusted model: age, sex, residence, marital, education, smoke, drink, hypertension, treatments for dyslipidemia.

Abbreviations: OR, Odds Ratio; CI, Confidence interval; RC, Remnant Cholesterol; BMI, Body Mass Index; CMM, Cardiometabolic multimorbidity.

Table 3.

Joint effects of RC and BMI categories (< 24, 24–27.9, ≥ 28 kg/m²) on CMM risk.

character	Unadjusted		Adjusted
	OR(95%CI)	P value	95%CI	OR(95%CI)
RC < median	ref		ref
RC ≥ median	1.54(1.27,1.86)	< 0.0001	1.35(1.11,1.65)	0.003
BMI < 24	ref		ref
BMI 24–28	1.93(1.56,2.39)	< 0.0001	1.64(1.31,2.07)	< 0.0001
BMI ≥ 28	3.55(2.76,4.57)	< 0.0001	2.36(1.78,3.13)	< 0.0001
p for trend		< 0.0001		< 0.0001
RC < median & BMI < 24	ref		ref
RC < median & 24 ≤ BMI < 28	1.85(1.33,2.58)	< 0.001	1.57(1.11,2.22)	0.01
RC < median & BMI ≥ 28	3.76(2.47,5.74)	< 0.0001	2.82(1.80,4.41)	< 0.0001
RC ≥ median & BMI < 24	1.36(1.02,1.82)	0.04	1.3(0.97,1.75)	0.08
RC ≥ median & 24 ≤ BMI < 28	2.52(1.89,3.35)	< 0.0001	2.08(1.54,2.81)	< 0.0001
RC ≥ median & BMI ≥ 28	4.26(3.08,5.88)	< 0.0001	2.62(1.84,3.72)	< 0.0001
p for trend		< 0.0001		< 0.0001

To further assess dose–response consistency, BMI was categorized into three groups (< 24, 24–27.9, ≥ 28 kg/m²), yielding six RC×BMI joint categories.

Crude model unadjusted for covariates; Adjusted model: age, sex, residence, marital, education, smoke, drink, hypertension, treatments for dyslipidemia.

Abbreviations: OR, Odds Ratio; CI, Confidence interval; RC, Remnant Cholesterol; BMI, Body Mass Index; CMM, Cardiometabolic multimorbidity.

Supplementary threshold-based joint analysis

To further verify the nonlinear relationships identified in the restricted cubic spline (RCS) models, participants were reclassified according to the inflection points of RC (18.6 mg/dL) and BMI (23.1 kg/m²). As shown in Supplementary Table S2, compared with the low RC + low BMI group, individuals with either high RC or high BMI alone exhibited moderately higher odds of CMM (OR = 2.09 [95% CI: 1.52–2.86] and 1.44 [1.03–2.02], respectively), while those with both high RC and high BMI had the greatest risk (OR = 2.41 [1.80–3.24], P < 0.001; P for trend < 0.001). These findings confirmed the presence of threshold-dependent and combined effects between RC and BMI, consistent with the results of the primary analysis.

In the present study, subgroup analyses were conducted to examine potential heterogeneity in the associations between RC, BMI, and CMM risk (Table 4). In participants aged < 60 years, those with RC ≥ median and BMI ≥ 24 kg/m² showed higher odds of CMM compared with their older counterparts (OR = 4.01 vs. 2.73, P < 0.0001). Men tended to have higher odds of CMM than women (OR = 3.80 vs. 2.44, P < 0.0001), and smokers or alcohol drinkers exhibited greater risk than non-smokers and non-drinkers (OR = 4.58 vs. 2.54 and OR = 4.33 vs. 2.57, respectively). Although the p for interaction values were not statistically significant (all p > 0.05), these trends suggest that the combined influence of elevated RC and higher BMI may be more pronounced among younger or high-risk individuals.

Table 4.

Subgroup analyses of the association between RC、BMI and CMM incidence.

Character	Q1	Q2	p	Q3	p	Q4	p	p for trend (character2 integer)	P for interaction
Age group									0.248
<60	ref	3.193(2.057,5.038)	< 0.0001	1.442(0.885,2.356)	0.141	4.011(2.692,6.130)	< 0.0001	< 0.0001
≥60	ref	1.917(1.253,2.900)	0.002	1.350(0.935,1.948)	0.108	2.731(1.940,3.863)	< 0.0001	< 0.0001
Sex									0.405
female	ref	1.897(1.304,2.768)	< 0.001	1.137(0.765,1.686)	0.522	2.435(1.752,3.423)	< 0.0001	< 0.0001
male	ref	2.610(1.599,4.225)	< 0.001	1.660(1.074,2.573)	0.023	3.796(2.538,5.745)	< 0.0001	< 0.0001
Smoke									0.113
no	ref	1.952(1.393,2.740)	< 0.001	1.369(0.969,1.934)	0.074	2.539(1.888,3.443)	< 0.0001	< 0.0001
yes	ref	3.087(1.634,5.716)	< 0.001	1.369(0.789,2.379)	0.262	4.577(2.774,7.686)	< 0.0001	< 0.0001
Drink									0.082
no	ref	2.223(1.590,3.113)	< 0.0001	1.150(0.808,1.634)	0.434	2.567(1.908,3.480)	< 0.0001	< 0.0001
yes	ref	1.996(1.041,3.746)	0.033	1.989(1.172,3.425)	0.012	4.334(2.643,7.291)	< 0.0001	< 0.0001
Hypertension									0.063
no	ref	2.336(1.542,3.527)	< 0.0001	1.372(0.922,2.041)	0.118	2.963(2.065,4.284)	< 0.0001	< 0.0001
yes	ref	1.267(0.818,1.969)	0.290	1.186(0.756,1.860)	0.457	1.615(1.108,2.386)	0.014	0.016
Dyslipidemia									0.064
no	ref	2.382(1.729,3.280)	< 0.0001	1.376(1.004,1.886)	0.047	2.746(2.065,3.672)	< 0.0001	< 0.0001
yes	ref	0.780(0.351,1.740)	0.540	1.031(0.453,2.345)	0.942	1.606(0.852,3.190)	0.157	0.034

Q1: RC < median & BMI < 24; Q2: RC < median & BMI ≥ 24; Q3: RC ≥ median & BMI < 24; Q4:RC ≥ median & BMI ≥ 24.

Abbreviations: OR, Odds Ratio; CI, Confidence interval; RC, Remnant Cholesterol; BMI, Body Mass Index; CMM, Cardiometabolic multimorbidity.

Mediator analysis

The mediation analyses (Fig. 3) demonstrated that BMI significantly mediated the relationship between RC and CMM, with an average indirect effect (ACME) of β = 0.00021 (95% CI: 0.00015–0.00029, P < 0.001) and a corresponding proportion mediated of 32.6% (95% CI: 19.1%–70.0%) after full adjustment. Conversely, RC partially mediated the association between BMI and CMM, with an average indirect effect of β = 0.00005 (95% CI: 0.00005–0.00006, P = 0.032) and a mediated proportion of 6.9% (95% CI: 0.7%–14.0%).Both mediation pathways remained significant in unadjusted analyses (β = 0.00032 and 0.00029, respectively), indicating bidirectional yet modest mediation effects between RC and BMI in influencing CMM risk.

Fig. 3.

Bidirectional mediation between RC and BMI on CMM risk.OR, Odds Ratio; CI, Confidence interval; RC, Remnant Cholesterol; BMI, Body Mass Index; CMM, Cardiometabolic multimorbidity. The model adjusted for age, sex, residence, marital, education, smoke, drink, hypertension, treatments for dyslipidemia.

Predictive performance of RC and BMI for CMM

To evaluate the predictive ability of remnant cholesterol (RC) and body mass index (BMI) for cardiometabolic multimorbidity (CMM), receiver operating characteristic (ROC) analyses were performed (Fig. 4). The area under the ROC curve (AUC) was 0.578 for RC and 0.650 for BMI, indicating that BMI had a better discrimination ability than RC alone. When both RC and BMI were included in the model, the AUC increased to 0.657, suggesting an additive predictive effect. However, the RC/BMI ratio showed a lower AUC of 0.542, indicating limited predictive value.

Fig. 4.

Receiver operating characteristic (ROC) curves for predicting cardiometabolic multimorbidity (CMM).

Discussion

This study systematically evaluated the independent and joint associations of remnant cholesterol (RC) and body mass index (BMI) with cardiometabolic multimorbidity (CMM) in 6,646 adults from the CHARLS cohort. Both elevated RC and higher BMI were significant independent predictors of CMM, exhibiting an additive rather than synergistic pattern. Participants with high RC (≥ median) showed a 35% higher odds of CMM (adjusted OR = 1.35, 95% CI 1.11–1.65, P = 0.003), while those with BMI of 24–28 and ≥ 28 kg/m² had 1.64- and 2.36-fold higher risks, respectively, compared with BMI < 24 kg/m². The combined exposure to high RC and high BMI conferred the greatest overall risk (adjusted OR = 2.23, 95% CI 1.69–2.94, P < 0.0001). However, formal additive-interaction tests (RERI/AP/S) were not significant, indicating a cumulative metabolic burden rather than supra-additivity. These findings were consistent across threshold-based and spline-derived analyses (RC = 18.6 mg/dL; BMI = 23.1 kg/m²), underscoring the robustness of the associations.

Our findings are broadly consistent with previous cohort studies showing that evated remnant cholesterol (RC)^23,29 and higher BMI¹³ each increase the risk of cardiometabolic diseases, including diabetes and cardiovascular events. However, prior investigations have primarily focused on single outcomes, whereas our study extends these observations to cardiometabolic multimorbidity (CMM), capturing the coexistence of multiple metabolic disorders in the same individuals. By jointly analyzing RC and BMI, we demonstrate that their effects are additive rather than synergistic, a distinction supported by formal interaction testing. This additive pattern suggests that RC-related dyslipidemia and adiposity-related metabolic stress contribute to CMM through parallel yet interlinked pathways, such as insulin resistance and low-grade inflammation. From a broader perspective, these results deepen understanding of metabolic clustering and highlight the need for integrated management strategies targeting both lipid and adiposity burdens to prevent multimorbidity.

Remnant cholesterol (RC) is a novel metabolic index proposed in recent years, and several studies have shown that RC is recognized as a potential predictor of cardiovascular disease²⁹. It consists mainly of HDL cholesterol with the fraction of LDL cholesterol removed and is dominated by cholesterol from triglyceride-rich lipoproteins¹⁸. Regardless of the level of LDL-C, RC is consistently and independently associated with the development of CVD⁶ and reflects the residual risk after treatment with statins and PCSK9 inhibitors⁸. In addition, RC not only induces atherosclerosis, but also triggers an inflammatory response⁹, which is closely related to the pathogenesis of CMM, and also shows a unique value in metabolic diseases. The role of RC in metabolic diseases is not limited to cardiovascular diseases. Studies have shown that RC is more strongly associated with DM than LDL-C, and that insulin resistance and pro-inflammatory status may be key links³⁰. Diabetic patients are often associated with dyslipidemia, and despite statins being the primary treatment, residual risk remains. Considering that cardiovascular disease and diabetes share a common pathogenesis, RC may enhance the risk of CMM through complex metabolic pathways, especially by promoting atherosclerosis²³, activating pro-inflammatory factors and exacerbating insulin resistance. Experimental work provides mechanistic support linking RC to low-grade inflammation and insulin resistance. In human in-vivo and ex-vivo models, remnant-like lipoprotein particles were shown to activate endothelial and immune pathways and to increase pro-inflammatory cytokines (e.g., interleukin-6 and tumor necrosis factor-α), thereby aggravating vascular inflammation and impairing insulin signaling⁹. These data offer biological plausibility for the associations observed in our cohort. However, because our analyses are epidemiological, the inflammatory pathway should be regarded as a putative mechanism rather than a causal explanation; future population studies integrating lipidomic profiling with inflammatory biomarkers are warranted to validate these pathways. These findings are generally consistent with prior evidence, although several contextual differences—such as population characteristics, outcome definitions, and exposure measurements—may influence the observed effect estimates. Importantly, while the Mendelian randomization study by Zhao et al. (2024)¹⁶ inferred causality by minimizing unmeasured confounding, our analyses are observational in nature and based on multivariable adjustment. Accordingly, the reported odds ratios should be interpreted as associations rather than causal estimates. Differences in study design, confounder control, and temporal assessment likely account for the modest variations in effect size between studies.

BMI is now a commonly used indicator for clinical assessment of the degree of obesity and associated health risks in individuals. Obesity has been widely shown to be a risk factor for diabetes and metabolic diseases, especially high BMI obesity³¹. Adipose tissue is categorized as either white adipose tissue (WAT) or brown adipose tissue (BAT), with visceral fat (which is WAT) having a more significant metabolic impact³². Visceral fat, compared to subcutaneous fat, releases large amounts of free fatty acids (NEFAs) and pro-inflammatory cytokines³³, metabolites that influence glucose-lipid metabolism via the liver, leading to insulin resistance and chronic inflammatory responses. In addition, high BMI-type obesity (especially abdominal obesity) not only increases the inflammatory burden but may also further exacerbate lipid metabolism disorders through adipose tissue remodeling and negative metabolic feedback mechanisms³⁴.

These mechanisms highlight the intricate metabolic interplay between RC and BMI, which may differ across populations and contribute to the observed heterogeneity in effect estimates. Heterogeneity may arise from (i) outcome definitions (CMM versus single endpoints), (ii) population structure (≥ 45 years in CHARLS with mixed urban–rural profiles), and (iii) exposure assessment (RC calculated from TC–LDL–HDL with medical therapy and lifestyle factors variably controlled). Moreover, our additive interaction analysis did not show statistically significant supra-additivity (RERI/AP/S not significant), suggesting that RC and BMI may operate through partly parallel pathways rather than amplifying each other beyond additivity. These design and population differences can yield effect sizes that diverge from single-disease cohorts while remaining directionally consistent.

The interaction between RC and BMI may create positive feedback effects through common metabolic pathways (e.g., lipid metabolism disorders and inflammatory responses), thereby contributing to the development of CMM. For example, pro-inflammatory factors released from visceral fat in individuals with high BMI may enhance the overall inflammatory milieu associated with elevated RC, while higher RC levels may, in turn, aggravate insulin resistance through increased lipid metabolic burden. This metabolic interplay suggests potential synergistic patterns at the physiological level, which help explain the observed statistical association between RC, BMI, and CMM, rather than implying direct causal amplification. These interrelationships provide valuable perspectives for designing comprehensive metabolic intervention strategies.

The mediation analysis showed that BMI mediated 32.6% of the total effect between RC and CMM, whereas RC mediated 6.9% of the total effect between BMI and CMM, further supporting the complex interaction between the two in metabolic abnormalities.

These statistical findings not only clarify the metabolic linkage between RC and BMI but also have practical implications for clinical risk assessment. In risk stratification, BMI alone showed modest discrimination (AUC 0.650) and RC alone was lower (AUC 0.578), whereas combining RC with BMI provided incremental, albeit limited, improvement (AUC 0.657). Thus, RC and BMI should be incorporated as part of multi-parameter assessment rather than used in isolation for screening. The RCS-derived turning points (RC ≈ 18.6 mg/dL; BMI ≈ 23.1 kg/m²) may serve as pragmatic early-warning thresholds in middle-aged and older adults, prompting closer follow-up, lifestyle programs targeting adiposity, and lipid-lowering strategies when appropriate. These findings support a dual-target approach (weight reduction plus triglyceride-rich lipoprotein lowering) in routine care to reduce multimorbidity burden.

The present study is consistent with previous studies and expands on some aspects. It has been shown that RC and elevated BMI are independently associated with the incidence of CMM. However, the combined effect of RC and BMI on CMM has been less studied. In this study, for the first time, RC and BMI were combined and quantitatively analyzed for combined effect studies, revealing the cumulative effect of both on CMM. Further stratified analyses showed that the combined effect of RC and BMI was more pronounced in specific populations, such as men, smokers, and advanced age. These groups may have higher levels of inflammation and metabolic burden, making them more sensitive to the combined effects of RC and BMI. This finding suggests that individual metabolic status and environmental factors may significantly modulate the effects of RC and BMI on CMM in different populations, emphasizing the importance of accurate screening and individualized interventions.

This study has potential implications for public health and clinical practice, particularly among middle-aged and older adults with similar demographic characteristics to the CHARLS cohort. First, RC and BMI serve as important joint indicators of CMM occurrence. It may be difficult for any single indicator to fully and accurately capture metabolic abnormalities status, and the joint assessment helps to accurately identify high-risk individuals, especially those with high levels of both BMI and RC. Second, based on the results of mediation analysis, intervention through weight loss not only directly reduces the risk of CMM, but may also indirectly mitigate the risk of CMM by lowering RC levels; meanwhile, lipid-lowering therapy not only mitigates the risk of CMM from high BMI, but may also improve the overall metabolic status by lowering RC levels. Finally, increased public awareness, early screening, and comprehensive risk assessment could play a positive role in reducing the burden of CMM.

There are several limitations that should be acknowledged in this study. First, due to the longitudinal cohort study, it is difficult to infer causality among RC, BMI, and CMM. Second, although multiple confounders were adjusted for, residual factors such as diet, exercise, and socioeconomic status might still have influenced the results. Third, disease identification in CHARLS was based mainly on self-reported physician diagnoses rather than ICD-coded medical records, which may have introduced recall or misclassification bias, particularly among participants with lower education levels or limited access to healthcare. Nevertheless, such misclassification is likely to be non-differential with respect to RC and BMI, which would bias the estimates toward the null rather than create spurious associations. In addition, remnant cholesterol (RC) was derived from a calculated formula (TC – LDL-C – HDL-C) instead of being directly measured in laboratory assays, which may introduce estimation error and potentially attenuate the true associations. Furthermore, the standardized CHARLS questionnaires have been validated in previous studies and shown acceptable agreement with hospital records. In addition, potential model misspecification cannot be fully excluded; despite using multivariable logistic regression with spline and interaction analyses, remaining nonlinear effects or unmeasured interactions might have biased the estimates. Moreover, this study did not capture the temporal dynamics of RC and BMI, while their cumulative changes over time may be relevant for CMM development. Finally, since all participants were aged ≥ 45 years and drawn from specific regions in China, the findings may not be fully generalizable to younger populations or other areas. Future studies should employ longitudinal and interventional designs to clarify causal relationships, assess cumulative and dynamic effects, and validate the generalizability of our findings across age groups and regions.

Conclusion

The present study demonstrated that RC and BMI were independent predictors of CMM with significant combined effects. Mediation analyses showed that BMI played a crucial role in amplifying the effect of remnant cholesterol on CMM risk, and remnant cholesterol partially mediated the relationship between BMI and CMM. These findings emphasize the importance of considering both RC and BMI in clinical assessments to more effectively identify at-risk individuals. Further studies are warranted to explore the underlying mechanisms and develop targeted interventions to reduce both RC and BMI levels, thereby reducing the burden of CMM and promoting metabolic health.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

RC

Remnant cholesterol

CVD

Cardiovascular disease

BMI

Body mass index

CMM

Cardiometabolic multimorbidity

CHARLS

China Health and Retirement Longitudinal Study

SD

Standard deviation

OR

Odds ratio

CI

Confidence interval

WAT

White adipose tissue

BAT

Brown adipose tissue

NEFAs

non-esterified fatty acids

RCS

Restricted cubic spline

Author contributions

Jiaxin Li and Kuanlu Fan were involved in the experiment design. Jiaxin Li performed the data analysis. Jiaxin Li, Mingxue Fan, Mingyue Yang, Wei Huang, Wenjing Shi and Kuanlu Fan wrote the manuscript. Mingxue Fan, Daowen Zhang, Weiwei Zhu, Cancan Liu, Nannan Xu, Huijun Zhang and Kuanlu Fan reviewed the manuscript and provided critical suggestions. Jiaxin Li, Mingxue Fan, Mingyue Yang, Wei Huang, Wenjing Shi, Daowen Zhang, Weiwei Zhu, Cancan Liu, Nannan Xu, Huijun Zhang and Kuanlu Fan revised the manuscript and improve the writing quality. Jiaxin Li and Kuanlu Fan approved the final version of the manuscript.

Funding

This work was supported by Development Fund of Affiliated Hospital of Xuzhou Medical University (grant number XYFC22020005).

Data availability

The datasets used and/or analyzed during the current study are publicly available or can be obtained from the corresponding authors upon reasonable request.

Declarations

Competing interests 

The authors declare no competing interests.

Ethics approval and consent to participate

The CHARLS study was conducted in accordance with the principles of the Declaration of Helsinki and was approved by the Biomedical Ethics Committee of Peking University (IRB00001052-11015). All participants provided written informed consent prior to participation in the CHARLS study.