Low BMI predicts mortality in T2DM older adults: cholesterol parameters demonstrate negative mediation

Abstract

Objective

The present study aims to investigate the association of low body mass index (BMI < 22 kg/m²) and core lipid parameters on all-cause mortality among older adults with type 2 diabetes mellitus (T2DM), and to assess the mediating roles of total cholesterol (TC), triglycerides (TG), high-density lipoprotein cholesterol (HDL-C), and low-density lipoprotein cholesterol (LDL-C).

Methods

Data were analyzed from 372,559 T2DM participants aged ≥ 60 years, obtained from the Shanghai Center for Disease Control and Prevention (SCDC) diabetes management cohort, stratified by BMI < 22 kg/m². Associations between low BMI and lipid parameters were examined using linear regression, associations of low BMI and lipid parameters with mortality risk were assessed via Cox regression. Restricted cubic spline (RCS) models were used to evaluate nonlinear trends, and Bayesian mediation models quantified the mediating effects.

Results

The low BMI group had a significantly higher risk of all-cause mortality compared to the non-low BMI group (HR = 1.10, 95% CI: 1.06 ~ 1.13). Among the lipid parameters, TC, HDL-C, and LDL-C were nonlinearly associated with mortality risk (U-shaped for TC and LDL-C, L-shaped for HDL-C), with optimal thresholds identified at 5.03 mmol/L, 0.98 mmol/L, and 3.39 mmol/L, respectively. Mediation analysis demonstrated significant negative mediation for TC (proportion mediated, PE = –16.68%), HDL-C (PE = –111.99%), and LDL-C (PE = –4.12%), with HDL-C showing the strongest masking effect (βACME = 1.44, 95% CI: 1.33 ~ 1.56).

Conclusion

Low BMI is independently associated with an increased risk of mortality in T2DM older adults. Cholesterol parameters, particularly HDL-C, may explain part of this association through negative mediation. These findings suggest the need for clinical attention to lipid metabolic disturbances among underweight older adults.

Background

Type 2 diabetes mellitus (T2DM) has become a major chronic disease threatening the health of the older adult population in China, with prevalence rates exceeding 28.5% and rising sharply with age [1, 2]. Older adults with T2DM face not only challenges in glycemic control but also additional burdens arising from age-related metabolic changes and alterations in body composition. Among these, weight abnormalities, particularly involuntary weight loss, as well as low body mass index (BMI) are key factors influencing disease prognosis. Contrary to the traditional perception that ‘lower weight is always healthier,’ recent studies have suggested that maintaining a moderate body weight is significantly associated with better survival outcomes in this population [3–5].

The link between low BMI and mortality in older adults with T2DM has attracted increasing scholarly attention. A review involving older adults with T2DM found that those with low BMI (20 ~ 25 kg/m2) had a higher risk of mortality compared to those with fit BMI [6]. Retrospective studies conducted in China further corroborate that both low and high BMI increase mortality risk among older adults with T2DM [7]. Importantly, low BMI in T2DM older adults typically reflects more than simple underweight,it is often accompanied by sarcopenia and altered fat distribution. Such changes in body composition can aggravate pathophysiological processes including impaired glucose disposal related to muscle mass loss and reduced energy reserves, which collectively lower resilience to acute illnesses and increase susceptibility to severe metabolic disturbances and multiple organ dysfunction during stress events such as infections or surgery, ultimately contributing to elevated mortality risk [8–10]. These evidences point to the central issue of the biological pathways by which low BMI influences long-term survival in older adults with T2DM.

In this context, dyslipidemia, a hallmark pathophysiological change in diabetes, may be a key correlate linking low BMI and mortality risk. Core lipid parameters, total cholesterol (TC), triglycerides (TG), low-density lipoprotein cholesterol (LDL-C), and high-density lipoprotein cholesterol (HDL-C), are not only biomarkers for cardiovascular risk, but also directly involved in energy homeostasis, inflammatory, and cell membrane stability [11–13]. In the context of low BMI, T2DM older adults experience reduced hepatic lipoprotein synthesis due to depleted energy stores, and diminished skeletal muscle uptake of free fatty acids, resulting in both hypocholesterolemia and dysfunctional HDL-C ([14]). Previous research has noted that 87.7% of T2DM individuals in Northwest China demonstrate distinctive patterns of dyslipidemia. Such changes, highly correlated with malnutrition, are regarded as "metabolic wasting syndrome," with potential adverse mechanisms including impaired steroid hormone synthesis and cell membrane integrity due to low cholesterol, diminished stress response, reduced anti-inflammatory and antioxidant capacity due to dysfunctional HDL-C, leading to accelerated vascular endothelial injury, and impaired prostaglandin synthesis and immune function due to essential fatty acid deficiency [15, 16]. Notably, lipid parameters may play a complex mediating role in the relationship between low BMI and mortality. This concept is supported by evidence suggesting that the association between low BMI and mortality can be substantially explained through measurable mediating pathways [17]. This highlights that dyslipidemia may contribute indirectly to mortality risk by affecting neurocognitive function and activity levels.

To address these knowledge gaps, the present study utilizes systematic data from Shanghai Municipal Center for Disease Control and Prevention (SCDC) to investigate the mediating effects of the main lipid parameters (TC, TG, LDL-C, HDL-C) in the relationship between low BMI (BMI < 22) and all-cause mortality among T2DM participants aged 60 and above. The specific aims are: 1) to characterize the strength and temporal features of the relationship between low BMI, lipid parameters and all-cause mortality, and 2) to examine the potential mediating role of TC, TG, LDL-C, and HDL-C to this association.

Materials and methods

Study population

The participants in this study were drawn from the diabetes management population maintained by SCDC. The diabetes management program utilizes a community-hospital-family-doctor contracting model targeting older adults with T2DM, with the goal of establishing a comprehensive prevention and control system covering the entire population and life course of T2DM. It aims to reduce the burden of diabetes, prevent complications, delay mortality, and enhance the precision and efficiency of services. The study adhered to the principles of the Declaration of Helsinki, with all participants providing written informed consent prior to enrollment. The protocol was approved by the SCDC Ethics Committee (KY-2024–52-C). T2DM was diagnosed by physicians according to clinical criteria. Detailed inclusion and exclusion criteria for participant enrollment, as well as data collection and quality control procedures, have been described previously [18–20]. For the present analysis, the initial diabetes management database was further screened to exclude participants with missing key data, including those lost to follow-up or who refused follow-up, those missing BMI or lipid profile laboratory values, and those > 60 years of age at entry. Ultimately, 372,559 participants were included for analysis (see Fig. 1).

Fig. 1.

Sample selection flow chart

Assessment of low BMI

The primary exposure variable was low BMI. Low BMI was defined as BMI < 22 kg/m². This threshold was chosen based on prior studies in Asian populations that have associated BMI < 22 kg/m² with undernutrition and increased health risks in older adults [21, 22], and aligns with cutoffs suggested for geriatric nutritional assessment. BMI was calculated by dividing weight (kg) by height squared (m²). Height and weight were measured by family doctors using standardized instruments during clinic visits and were automatically entered into the diabetes management system.

Assessment of lipid parameters

The mediators in this study were the four core lipid parameters: TC, TG, HDL-C and LDL-C. Fasting venous blood samples were collected by nurses, with fasting for 8–10 h required prior to collection. Samples were drawn into vacuum tubes containing appropriate anticoagulant, allowed to stand at room temperature for approximately 30 min, and then centrifuged under defined conditions to separate plasma or serum. Separated plasma/serum samples were aliquoted and stored at –80 °C or below until analysis. Lipid measurements were performed in clinical laboratories using standardized enzymatic methods: TC was measured by the cholesterol oxidase–peroxidase (CHOD-PAP) method, TG by the glycerol phosphate oxidase–peroxidase (GPO-PAP) method, and HDL-C by direct homogeneous assay (immunoseparation). LDL-C was calculated using the Friedewald formula LDL-C = TC – HDL-C – (TG/2.2) (in mmol/L) when TG < 4.52 mmol/L, and by direct assay for samples with higher TG [23, 24]. All assays were conducted by trained personnel on calibrated automated analyzers, strictly following standard operating procedures.

Assessment of all-cause mortality and survival time

Mortality outcomes were ascertained via linkage to the Shanghai Mortality Registry System, which automatically captures death data for all participants. The observation period was from May 1, 2020 to Dec 31, 2024, and survival time was computed as months from baseline until the date of death. Death was defined according to standard criteria for both cardiac and brain death. Cases of early death (3 months of follow-up), participants who moved out of Shanghai, and those who refused follow-up were excluded. The median follow-up duration was 19 months.

Assessment of confounding variables

Potential confounding variables were selected a priori based on established biological plausibility and evidence from the literature regarding their associations with body composition, lipid metabolism, and mortality in older adults with diabetes [25, 26]. Data on these confounding variables were collected at baseline through a structured Demographic Questionnaire, administered by community physicians during outpatient interviews. The information was recorded directly into electronic chronic disease management cards. The variables, their sources, and operational definitions are as follows: 1) Socio-demographic factors: Age (treated as a continuous variable in years), sex (male/female), education level (categorized as primary school or below, junior high school, and high school or above), and marital status (classified as married, including cohabiting, or unmarried, including never married, divorced, or widowed). 2) Lifestyle factors: Smoking status, alcohol consumption, and engagement in regular physical activity (all categorized as yes or no).

Statistical analysis

The diabetes management database was constructed and cleaned using SAS software. All subsequent statistical analyzes were conducted using R software (version 4.4). Key analyzes utilized the following R packages: "survival" for Cox regressions, "rms" for regression modeling and RCS, "segmented" for threshold effect analysis, and "mediation" for mediation analysis.

Participants were classified into two groups according to BMI value: the "low BMI" group (22 kg/m²) and the "non-low BMI" group (≥ 22 kg/m²). Group differences in confounding variables and lipid parameters were assessed using the chi-square test for categorical variables and independent-samples t-tests for normally distributed continuous variables. The confounding variables were selected a priori based on biological plausibility and known associations from the literature, while group comparisons were used solely for descriptive purposes.

Pairwise associations among low BMI, lipid parameters, and all-cause mortality were analyzed using regression methods. First, linear regression was employed to assess the association between low BMI (exposure) and each lipid parameter (potential mediator). Second, Cox proportional hazards regression was used to analyze the associations of low BMI and each lipid parameter with all-cause mortality. For both linear and Cox models, a two-step approach was adopted: Model I was the unadjusted (crude) model; Model II adjusted for confounding variables (age, sex, education, smoking, alcohol consumption, and physical activity) selected a priori based on biological plausibility and literature. The adequacy of the Cox proportional hazards models was assessed by evaluating the proportional hazards assumption using Schoenfeld residual tests and by examining deviance residuals to identify potential influential observations.

Kaplan–Meier (KM) survival curves were plotted, and log-rank tests were performed to visually compare survival probabilities between the two BMI groups. For lipid parameters that showed significant associations with all-cause mortality in Cox models, we further investigated potential nonlinear dose–response relationships. This was done by fitting RCS models within the Cox framework using the "rms" package, with knots placed at predefined percentiles (the 5th, 50th, and 95th percentiles).

To quantitatively identify inflection points in the dose–response curves suggested by the RCS analysis, a threshold analysis was performed using the "segmented" package, which implements segmented linear regression to estimate the optimal breakpoint and the linear associations on either side of it. Both crude and adjusted models were fitted for this analysis. Finally, to explore the potential indirect pathways, a quasi-Bayesian Monte Carlo simulation-based mediation analysis was conducted using the "mediation" package. This model quantified the mediating role of each lipid parameter in the association between low BMI and all-cause mortality, estimating the average causal mediation effect (ACME), average direct effect (ADE), total effect (TE), and the proportion mediated (PE).

All statistical tests were two-sided, with a significance level set at α = 0.05.

Results

Demographic characteristics

A total of 372,559 participants were included in the study, comprising 167,695 males (45.01%) and 204,864 females (54.99%), with a mean age of 72.04 ± 7.31 years. Among all participants, 76,326 individuals with BMI < 22 kg/m² were classified as the low BMI group, while the remaining 296,233 were classified as the non-low BMI group. Group comparisons revealed statistically significant differences between the two groups for age, sex, education, smoking, alcohol consumption, and physical activity (all P < 0.001); marital status did not differ significantly between groups (P = 0.680). Therefore, age, sex, education, smoking, alcohol, and physical activity were included as confounding variables in subsequent regression and mediation models. Additionally, significant differences between the two groups were observed in all four lipid parameters (TC, TG, HDL-C, LDL-C), all P < 0.001 (see Table 1).

Table 1.

Demographic characteristics of participants, by low BMI

Predictors	Sample N = 372,559	BMI groups		P
		≥ 22, n = 296,233	< 22, n = 76,326
Age (year), Mean ± SD	72.04 ± 7.31	71.92 ± 7.25	72.51 ± 7.54	< 0.001
Sex, n(%)				< 0.001
Male	167,695 (45.01)	136,868 (46.20)	30,827 (40.39)
Female	204,864 (54.99)	159,365 (53.80)	45,499 (59.61)
Marital, n(%)				0.680
Married	289,717 (77.76)	230,405 (77.78)	59,312 (77.71)
Not married	82,842 (22.24)	65,828 (22.22)	17,014 (22.29)
Education, n(%)				< 0.001
Primary school and below	211,903 (56.88)	170,894 (57.69)	41,009 (53.73)
Junior high school	87,109 (23.38)	67,889 (22.92)	19,220 (25.18)
High school and above	73,547 (19.74)	57,450 (19.39)	16,097 (21.09)
Smoking, n(%)				< 0.001
No	334,442 (89.77)	265,397 (89.59)	69,045 (90.46)
Yes	38,117 (10.23)	30,836 (10.41)	7,281 (9.54)
Alcohol, n(%)				< 0.001
No	310,898 (83.45)	245,557 (82.89)	65,341 (85.61)
Yes	61,661 (16.55)	50,676 (17.11)	10,985 (14.39)
Physical activity, n(%)				< 0.001
No	245,737 (65.96)	198,323 (66.95)	47,414 (62.12)
Yes	126,822 (34.04)	97,910 (33.05)	28,912 (37.88)c
BMI (kg/m2), Mean ± SD	24.35 ± 3.37	25.33 ± 3.06	20.55 ± 1.07	< 0.001
TC (mmol/L), Mean ± SD	4.82 ± 1.22	4.81 ± 1.22	4.90 ± 1.23	< 0.001
TG (mmol/L), Mean ± SD	1.88 ± 2.50	1.91 ± 2.50	1.74 ± 2.49	< 0.001
HDL-C (mmol/L), Mean ± SD	1.35 ± 0.39	1.33 ± 0.38	1.43 ± 0.43	< 0.001
LDL-C (mmol/L), Mean ± SD	2.81 ± 0.95	2.80 ± 0.95	2.84 ± 0.96	< 0.001

SD: standard deviation, BMI: body mass index, TC: total cholesterol, TG: triglyceride, HDL-C: High-density lipoprotein cholesterol, LDL-C: Low density lipoprotein cholesterol; [bold]: P < 0.05, same as below.

Associations between low BMI and lipid parameters

Linear regression analyzes indicated that in Model I, low BMI was significantly associated with all four lipid parameters (TC, TG, HDL-C, LDL-C), with all P < 0.001. In Model II (adjusted for confounding variables), the associations were as follows: TC, β (95% CI) = 0.08 (0.07 ~ 0.09); TG, β (95% CI) = –0.17 (–0.19 ~ –0.15); HDL-C, β (95% CI) = 0.10 (0.10 ~ 0.11); LDL-C, β (95% CI) = 0.03 (0.01 ~ 0.04). All P < 0.001 (see Table 2).

Table 2.

Linear association between low BMI and lipid profile markers

Outcomes	Model I					Model II
	β	S.E	t	P	β (95%CI)	β	S.E	t	P	β (95%CI)
TC	0.09	0.00	18.90	< 0.001	0.09 (0.08 ~ 0.10)	0.08	0.00	15.61	< 0.001	0.08 (0.07 ~ 0.09)
TG	-0.17	0.01	-16.79	< 0.001	-0.17 (-0.19 ~ -0.15)	-0.17	0.01	-17.09	< 0.001	-0.17 (-0.19 ~ -0.15)
HDL-C	0.10	0.00	65.80	< 0.001	0.10 (0.10 ~ 0.11)	0.10	0.00	41.00	< 0.001	0.10 (0.10 ~ 0.11)
LDL-C	0.04	0.00	9.50	< 0.001	0.04 (0.03 ~ 0.04)	0.03	0.01	4.04	< 0.001	0.03 (0.01 ~ 0.04)

Model I: crude, Model II: adjusting for age, sex, education, smoking, alcohol, physical activity, CI: confidence interval

Associations of low BMI, lipid parameters, and all-cause mortality

In Cox regression analyzes of the association between low BMI and all-cause mortality, Model I revealed a statistically significant association (P < 0.001). In Model II, the hazard ratio for low BMI was HR (95% CI) = 1.10 (1.06 ~ 1.13), P < 0.001 (see Table 3). The median follow-up duration for the cohort was 19.0 months (range: 3.0 ~ 56.0 months; Q1 ~ Q3: 12.5 ~ 28.5 months). Kaplan–Meier survival curves and risk tables demonstrated that survival probability was significantly higher in the non-low BMI group compared to the low BMI group, log-rank P < 0.001 (see Fig. 2).

Table 3.

Cox association between low BMI and lipid profile markers with mortality

Predictors	Sample (n)	Person-time (person-months)	Model I		Model II
			HR (95%CI)	P	HR (95%CI)	P
Low BMI
≥ 22 kg/m2	296,233	5,634,828	1.00 (reference)	-	1.00 (reference)	-
< 22 kg/m2	76,326	1,561,988	1.17 (1.13 ~ 1.21)	< 0.001	1.10 (1.06 ~ 1.13)	< 0.001
TC	372,559	7,196,816	0.75 (0.74 ~ 0.76)	< 0.001	0.84 (0.83 ~ 0.85)	< 0.001
TG	372,559	7,196,816	0.97 (0.97 ~ 0.98)	< 0.001	1.00 (1.00 ~ 1.01)	0.296
HDL-C	372,559	7,196,816	0.40 (0.38 ~ 0.42)	< 0.001	0.45 (0.42 ~ 0.48)	< 0.001
LDL-C	372,559	7,196,816	0.77 (0.76 ~ 0.78)	< 0.001	0.92 (0.89 ~ 0.94)	< 0.001

Model I: crude, Model II: adjusting for age, sex, education, smoking, alcohol, physical activity, HR: hazard ratio, CI: confidence interval

Fig. 2.

KM curves of survival probability among BMI groups

For the association between lipid parameters and all-cause mortality, Cox regression in Model I showed that all four lipid indicators were significantly associated with all-cause mortality (all P < 0.001). In Model II, the results were: TC, HR (95% CI) = 0.84 (0.83 ~ 0.85); HDL-C, HR (95% CI) = 0.45 (0.42 ~ 0.48); LDL-C, HR (95% CI) = 0.92 (0.89 ~ 0.94) (all P < 0.001). TG was not significantly associated with mortality (P = 0.296) and was, therefore, excluded from further analysis (see Table 3.

Dose–response relationships of cholesterol markers and all-cause mortality

RCS analyzes revealed: A U-shaped, nonlinear dose–response relationship between TC and all-cause mortality, both before and after confounders adjustment; An L-shaped, nonlinear association between HDL-C and all-cause mortality, both before and after adjustment; A U-shaped, nonlinear association between LDL-C and all-cause mortality, both before and after adjustment. For all three markers, P for overall association and P for nonlinear were < 0.001. Additionally, the shape of the HDL-C curve, particularly at the extremes of the distribution, should be interpreted with caution due to fewer data points and wider CIs. This does not imply a definitive biological effect (see Fig. 3).

Fig. 3.

RCS curves for cox association between cholesterol markers and mortality for both models

Thresholds for associations between cholesterol markers and all-cause mortality

RCS analysis indicated the presence of threshold effects in the relationship between the three cholesterol markers and all-cause mortality. After adjusting for confounding variables, the optimal cutoff values were: TC: 5.03 mmol/L; HDL-C: 0.98 mmol/L; LDL-C: 3.39 mmol/L. All P for likelihood tests were < 0.001 (see Table 4).

Table 4.

Threshold effects of cholesterol markers on mortality

Variables	Model I			Model II
	Outcome	effect	P	Outcome	effect	P
TC	Inflection point	5.44		Inflection point	5.03
	< 5.44	0.66 (0.65—0.67)	< 0.001	< 5.03	0.71 (0.70—0.73)	< 0.001
	≥ 5.44	1.13 (1.09—1.17)	< 0.001	≥ 5.03	1.13 (1.10—1.17)	< 0.001
	P for likelihood test		< 0.001	P for likelihood test		< 0.001
HDL-C	Inflection point	1.08		Inflection point	0.98
	< 1.08	0.03 (0.02—0.03)	< 0.001	< 0.98	0.03 (0.02—0.04)	< 0.001
	≥ 1.08	0.85 (0.80—0.89)	< 0.001	≥ 0.98	0.73 (0.67—0.79)	< 0.001
	P for likelihood test		< 0.001	P for likelihood test		< 0.001
LDL-C	Inflection point	2.98		Inflection point	3.39
	< 2.98	0.59 (0.57—0.61)	< 0.001	< 3.39	0.75 (0.72—0.78)	< 0.001
	≥ 2.98	1.12 (1.08—1.16)	< 0.001	≥ 3.39	1.27 (1.19—1.35)	< 0.001
	P for likelihood test		< 0.001	P for likelihood test		< 0.001

Mediation analyzes of cholesterol markers in the association between low bmi and all-cause mortality

Mediation analysis results showed TC exerted a reverse indirect association: proportion mediated (PE) = –16.68%, indirect effect βACME (95% CI) = 0.20 (0.18 ~ 0.22), direct effect βADE (95% CI) = –1.44 (–1.95 ~ –0.96), and total effect βTE (95% CI) = –1.23 (–1.75 ~ –0.76). HDL-C showed a substantial negative indirect association: PE = –111.99%, indirect effect βACME (95% CI) = 1.44 (1.33 ~ 1.56), direct effect βADE (95% CI) = –2.81 (–3.80 ~ –2.04), and total effect βTE (95% CI) = –1.37 (–2.33 ~ –0.71). LDL-C showed a modest reverse indirect association: PE = –4.12%, indirect effect βACME (95% CI) = 0.06 (0.03 ~ 0.08), direct effect βADE (95% CI) = –1.46 (–2.46 ~ –0.71), and total effect βTE (95% CI) = –1.41 (–2.40 ~ –0.67). These results are visualized in Fig. 4.

Fig. 4.

Mediation role of cholesterol markers in the association between low BMI and mortality

Discussion

Association between low BMI and mortality in t2dm older adults

Based on a large-scale community cohort, this study found that low BMI (22 kg/m²) was independently associated with all-cause mortality among older adult participants with T2DM (HR = 1.10). This finding is consistent with previous research, while also challenging the traditional view that ‘lower weight is always healthier’ [27–29]. As previously mentioned, low BMI in this population rarely reflects simple underweight status but more often coexists with sarcopenia, abnormal fat distribution, and depleted energy reserves. These altered body composition are associated with accelerated aging- and diabetes-related pathophysiological processes, and ultimately with elevated mortality risk. Our findings highlight the urgent need to recognize and address low BMI in the management of older adults with T2DM, as current guidelines tend to overemphasize obesity treatment and overlook the risks associated with underweight status [30].

Nonlinear associations and thresholds of cholesterol parameters with mortality risk

The present study provides novel insights into the nonlinear dose–response relationships between cholesterol parameters (TC, HDL-C, LDL-C) and all-cause mortality in T2DM older adult participants: both TC and LDL-C display U-shaped associations, while HDL-C presents an L-shaped pattern. This indicates that both excessively high and low concentrations of TC and LDL-C are associated with mortality. For HDL-C, a floor effect is observed: below a certain threshold, risk escalates sharply. Threshold analyzes allowed us to pinpoint the inflection points in these relationships; notably, the optimal HDL-C threshold identified in this cohort (0.98 mmol/L) is lower than commonly accepted normal lower limits (typically > 1.0 mmol/L for males and > 1.3 mmol/L for females). This suggests that even slight reductions in HDL-C, within the lower end of the normal range, may signal increased mortality for T2DM older adults with low BMI.

The complex indirect associations involving cholesterol: negative mediating roles and the centrality of HDL-C

Another key finding of the present study is the quantification of negative indirect association of cholesterol parameters in the association between low BMI and mortality. Unlike the conventional positive mediation (PE > 0%), indicating that the exposure (low BMI) raises the mediator (cholesterol) to increase the outcome (mortality) risk, negative mediation (PE < 0%) demonstrates that low BMI reduces TC, HDL-C, and LDL-C levels, which, in turn, may partially explain the association with low BMI on mortality, essentially masking the true magnitude of risk. Notably, HDL-C exhibited the strongest negative indirect association in our analysis. This means that the reduction in HDL-C is strongly involved in the association of low BMI with mortality. Such findings underscore the pivotal role of dysfunctional HDL-C, which manifests as impaired anti-inflammatory, antioxidant, and cholesterol efflux capacities in metabolic wasting syndromes associated with low BMI, which may correlate with accelerated vascular injury and weakening adaptive responses to stress [31–33]. TC and LDL-C also exhibited significant negative indirect associations, although to a lesser extent than HDL-C. Collectively, these findings suggest that hypocholesterolemia secondary to low BMI may compromise steroid hormone synthesis, membrane stability, and immune function, which may be associated with elevated mortality risk [34, 35].

Comparison with prior evidence on composite lipid indices

In recent years, there has been a growing interest in integrating lipid parameters into composite indices for health risk assessment. A notable study based on data from middle-aged and older adults demonstrated that the uric acid to high-density lipoprotein cholesterol ratio (UHR) is significantly associated with both all-cause and cardiovascular disease mortality, highlighting the importance of a comprehensive evaluation of lipid metabolic status [36]. Our study advances this understanding from a different yet complementary perspective. First, we focused specifically on the high-risk population of older adults with T2DM and directly investigated the nonlinear associations between the levels of core lipid parameters themselves and mortality risk, rather than ratios with other metabolites. More importantly, moving beyond simple association analysis, we quantified the potential pathways through which these core lipid parameters operate in the "low BMI -elevated mortality risk" association by applying mediation models. We found that a reduction in HDL-C statistically "masks" part of the total association between low BMI and mortality, suggesting that dysfunctional or low levels of HDL-C may represent a key intermediate biological state in underweight older adults with T2DM. This is inherently consistent with the concept in UHR research that views HDL-C as a protective component. However, our analysis further points to the central mediating role of HDL-C itself within a specific pathophysiological context. Therefore, our work, together with previous research by peers, confirms the core value of meticulously assessing lipid metabolism in predicting health risks in older adults. While their study supports the use of composite indices for risk stratification, our research provides deeper observational evidence for understanding the potential lipid-related pathways contributing to increased mortality in underweight older adults with T2DM by focusing on individual parameters and introducing mediation analysis.

Clinical and public health implications

The results have implications for clinical practice and public health. Clinically, low BMI (< 22 kg/m²) may be considered a high-risk indicator in older adults with T2DM and incorporated into risk assessment frameworks. Lipid management goals in this group, especially regarding HDL-C levels, should be carefully reviewed; values at the lower end of the conventional normal range may warrant attention. Active steps to address reversible causes of low HDL-C, such as malnutrition or chronic inflammation, are recommended. Prioritizing nutritional assessment and support, ensuring adequate energy and protein intake along with resistance or high-intensity interval training may help improve body composition and functional lipid profiles [37, 38]. Comorbidity management should also be optimized to reverse metabolic wasting. From a public health perspective, diabetes management programs should increase focus on low BMI and underweight older adults by strengthening screening, monitoring, and developing targeted interventions to reduce disease burden in this vulnerable group.

Limitations

Several limitations should be considered. First, despite multivariable adjustment, residual confounding cannot be fully excluded due to potential measurement error and unmeasured factors. Second, this study lacked detailed body composition data and relied on BMI rather than direct measures of muscle mass, fat mass, and their distribution, which are clinically relevant in older adults. Third, the median follow-up of 19 months captured short- to mid-term risks; longer follow-up is needed to assess long-term outcomes. Future studies with extended follow-up, detailed body composition assessment, and randomized trials evaluating nutritional, exercise, or pharmacological interventions are warranted to clarify these relationships.

Conclusion

This study suggests that a low BMI (22 kg/m²) is associated with an increased risk of all-cause mortality among older adults with T2DM. Cholesterol parameters (TC, HDL-C, LDL-C) show nonlinear associations with mortality risk: both TC and LDL-C exhibit U-shaped relationships, while HDL-C demonstrates an L-shaped curve. Importantly, these cholesterol parameters showed potential negative mediating roles in the association between low BMI and mortality. Among them, the "masking" effect of HDL-C is most pronounced, suggesting that the association between low BMI and mortality may involve reductions in HDL-C. Similar but weaker mediating pathways were observed for TC and LDL-C. No significant indirect association was found for TG.

Abbreviations

ACME

Average causal mediation effect

ADE

Average direct effect

BMI

Body mass index

CHOD-PAP

Cholesterol oxidase–peroxidase

CI

Confidence interval

GPO-PAP

Glycerol phosphate oxidase–peroxidase

HDL-C

High-density lipoprotein cholesterol

HR

Hazard ratio

LDL-C

Low-density lipoprotein cholesterol

PE

Proportion of mediation effect

RCS

Restricted cubic spline

SCDC

Shanghai Municipal Center for Disease Control and Prevention

TC

Total cholesterol

TG

Triglycerides

TE

Total Effect

T2DM

Type 2 diabetes mellitus

UHR

Uric acid to high-density lipoprotein cholesterol ratio

Author contribution

Conceptualization: HM, QYang; Methodology: HM, QYan; Software: HM, QYan, YW, SS; Validation: All authors; Formal analysis: HM; Investigation: QYang, YS, HY, MC; Resources: QYan, YS, HY, MC; Data Curation: QYang, YS, HY, MC; Writing—Original Draft: HM; Writing—Review & Editing: All authors; Visualization: HM; Supervision: FW, MC; Project administration: QYang, YS, HY, FW, MC; Funding acquisition: MC.

Funding

Non communicable Chronic Diseases-National Science and Technology Major Project (2024ZD0524200).

Data availability

The data is available at the authors.

Declarations

Ethics approval

This study conforms to the relevant provisions of the Declaration of Helsinki and the Guidelines for the Construction of Ethics Review Committees Involving Human Clinical Research. The ethical review was approved by the Ethics Committee of Shanghai Municipal Center for Disease Control and Prevention. All participants signed written informed consent forms.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.