Growth differentiation factor-15 and the incidence, bidirectional progression, and risk prediction of atherosclerotic cardiovascular disease and metabolic dysfunction-associated steatotic liver disease in individuals with cardiovascular-kidney-metabolic syndrome stages 0–3

Abstract

Background

Growth differentiation factor 15 (GDF-15) is a circulating biomarker reflecting oxidative stress, inflammation, and cellular aging. However, its role in disease risk assessment amongst individuals with cardiovascular-kidney-metabolic (CKM) syndrome stages 0–3 remains unclear.

Methods

This study included 29,697 UK Biobank participants with CKM stages 0–3 defined in accordance with the American Heart Association criteria. Associations of GDF-15 with metabolic, inflammatory and liver fibrosis markers and CKM stage were examined using linear or multinomial logistic regression models. Fine-Gray competing risk regression models were used to evaluate associations with incident atherosclerotic cardiovascular disease (ASCVD), metabolic dysfunction-associated steatotic liver disease (MASLD) and their comorbidity (coexistence of both conditions). Bidirectional disease transitions were assessed using a multi-state Markov model. The relative importance of GDF-15 was evaluated using SHapley Additive exPlanations (SHAP) and likelihood ratio (LR) statistics. Improvements in risk prediction models were assessed using time-dependent area under the receiver operating characteristic curve, Brier score, integrated discrimination improvement and continuous net reclassification improvement.

Results

Amongst 29,697 participants (mean age of 56.16 years; 57.32% female), 2,786 developed ASCVD and 456 developed MASLD during follow-up. Higher GDF-15 levels were associated with poorer CKM health and showed the strongest associations with renal function markers, followed by insulin resistance indices. Each 1-unit increase in GDF-15 (normalised protein expression, log2 scale) was associated with increased risks of ASCVD (HR = 1.25, 95%CI 1.15–1.36, P = 1.35 × 10^–7), MASLD (HR = 1.62, 95%CI 1.41–1.86, P = 2.06 × 10^–11) and their comorbidity (HR = 1.62, 95%CI 1.32–2.18, P = 4.36 × 10^–7) after multivariable adjustment for age, sex, smoking status, body mass index, diabetes mellitus, glycated haemoglobin, systolic blood pressure, low-density lipoprotein cholesterol, high-density lipoprotein cholesterol, C-reactive protein, creatinine, estimated glomerular filtration rate, urinary albumin-to-creatinine ratio, cystatin C, Townsend deprivation index, physical activity, and CKM stage. These associations remained consistent across subgroup analyses. Multi-state analyses indicated that GDF-15 predicted bidirectional progression between ASCVD and MASLD, with 10-year cumulative incidences of ASCVD and MASLD reaching 15.75% and 2.03%, respectively, among individuals in the top 10% of GDF-15 levels, and further increasing to 20.05% and 2.32% in those in the top 5%. SHAP and LR analyses showed that GDF-15 had high relative importance in predicting ASCVD and MASLD. Incorporating GDF-15 into established risk scores (PREVENT, SCORE2, FLI, FIB-4 and ARPI) showed modest improvements in risk discrimination, reclassification, and prediction error, particularly for ASCVD. In several settings, GDF-15 outperformed established biomarkers, including insulin resistance, systemic inflammation, apolipoprotein A/B, lipoprotein(a), cardiac troponin I, and N-terminal prohormone of brain natriuretic peptide.

Conclusions

GDF-15 may serve as a promising biomarker for cardiovascular-kidney-liver-metabolic syndrome risk stratification and management.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12933-026-03243-8.

Introduction

Cardiovascular-kidney-metabolic (CKM) health reflects the complex interplay among metabolic risk factors, chronic kidney disease (CKD), and the cardiovascular system, with important implications for morbidity and mortality. In 2023, the American Heart Association (AHA) introduced and defined the concept of CKM syndrome [1]. However, risk assessment in this population still relies largely on traditional cardiometabolic factors, and additional biomarkers reflecting multisystem stress are considered to potentially improve early identification of high-risk individuals [2].

Growth differentiation factor-15 (GDF-15), a member of the transforming growth factor-β superfamily, is a stress-responsive biomarker associated with inflammation, oxidative stress, and cellular aging [3]. Increased GDF-15 levels have been individually linked to CVD, diabetes mellitus (DM), CKD and metabolic dysfunction-associated steatotic liver disease (MASLD) [4–8]. These observations suggest that GDF-15 may capture shared pathophysiological mechanisms across cardiovascular, kidney and metabolic disorders. However, previous studies have not systematically incorporated CKM status, and they have generally been limited by relatively small sample sizes. Consequently, whether incorporating GDF-15 into risk assessment amongst individuals with CKM syndrome stages 0–3 improves the prediction of atherosclerotic cardiovascular disease (ASCVD) and MASLD remains unclear.

This study aimed to evaluate the association of GDF-15 with incident ASCVD and MASLD and their bidirectional progression amongst individuals with CKM syndrome stages 0–3, and determine whether GDF-15 improves risk prediction beyond traditional cardiometabolic risk factors and recently proposed novel biomarkers.

Methods

Study participants and design

This study used data from the UK Biobank (UKB), a prospective cohort of over 500,000 participants recruited across England, Scotland, and Wales between 2006 and 2010 [9]. Blood-based proteomic profiling was performed in 52,995 participants through the UKB Pharma Proteomics Project (UKB-PPP), representing a randomised subset of the cohort. The UKB received ethical approval from the North West Multi-centre Research Ethics Committee, and all participants provided written informed consent. This study was conducted under UKB application number 205837.

The study flow and design are shown in Fig. 1. This study initially included 30,246 participants with available data to determine CKM syndrome components and classified as stages 0–3. Amongst them, 29,771 participants had GDF-15 measurements. After those with outcome events at baseline were excluded, 29,697 participants were included in the final analysis.

Fig. 1.

Study flow diagram and design. Abbreviations: APRI, aspartate aminotransferase to platelet ratio index; ASCVD, atherosclerotic cardiovascular disease; CKM, cardiovascular-kidney-metabolic; FIB-4, fibrosis-4 score; FLI, fatty liver index; GDF-15, growth/differentiation factor-15; IDI, integrated discrimination improvement; MASLD, metabolic dysfunction-associated steatotic liver disease; NRI, net reclassification improvement; PREVENT, Predicting Risk of cardiovascular disease EVENTs; SCORE, Systematic COronary Risk Evaluation

Definition of CKM syndrome

The definition of CKM syndrome in this study followed the Presidential Advisory from the AHA [1]. Detailed criteria for each stage are provided in Supplementary Table 1. Briefly, Stage 0 (no CKM health risk factors) included individuals without overweight or obesity; metabolic risk factors, including hypertension, hypertriglyceridemia, metabolic syndrome (MetS) or DM; CKD; or subclinical/clinical CVD. Stage 1 (excess and/or dysfunctional adiposity) included individuals with overweight or obesity, abdominal obesity, or dysfunctional adipose tissue but without other metabolic risk factors or CKD. Stage 2 (metabolic risk factors and CKD) included individuals with metabolic risk factors such as hypertriglyceridemia, hypertension, MetS, or DM, or CKD. Stage 3 (subclinical CVD in CKM) was defined in this study as a high predicted 10-year CVD risk according to the Predicting Risk of cardiovascular disease EVENTs (PREVENT) equations [2, 10, 11]. Stage 4 (clinical CVD in CKM) included individuals with clinical CVD, including coronary heart disease, heart failure, stroke, peripheral artery disease, or atrial fibrillation, in the presence of excess or dysfunctional adiposity, other metabolic risk factors, or CKD.

Measurement of GDF-15

Proteomic profiling in the UKB was performed within UKB-PPP by using the Olink Explore platform [12–14]. EDTA plasma samples were selected using stratified pseudo-random sampling. Protein levels were measured using a proximity extension assay with next-generation sequencing readout, quantifying 2923 proteins. Protein abundance was reported as Normalized Protein eXpression (NPX) after internal control normalisation, plate and batch adjustment.

GDF-15 was the primary protein of interest in this study. Outliers were not removed because in medical research, outlying values may reflect genuine pathological variation rather than measurement error [15].

Definition of outcomes

In UKB, disease status was identified using International Classification of Diseases, Tenth Revision (ICD-10) codes. The first occurrence date of each disease was determined using information from primary care records, hospital admissions, and death registry records. Follow-up time was calculated from baseline to the earliest of the first disease occurrence, loss to follow-up, death, or the end of follow-up.

In the primary analyses, the outcomes of interest were ASCVD and MASLD. ASCVD was defined as coronary heart disease (I20-I25) and ischaemic stroke (I63) [2]. MASLD was identified using codes K76.0 and K75.8 [16]. These outcomes showed high positive predictive value (PPV), indicating that individuals classified as being positive for an event were likely to have actually experienced it. For example, the PPV was 79% for any stroke (83% for ischaemic stroke) [17], 82.2% for acute myocardial infarction [18], and 91.2% for MASLD [19].

Covariates

This study incorporated covariates from three sources. Firstly, cardiometabolic risk factors, including age, sex (male or female), smoking status (current or not), body mass index (BMI), DM (yes or not), glycated haemoglobin, systolic blood pressure (SBP), low-density lipoprotein cholesterol (LDL-C), high-density lipoprotein cholesterol (HDL-C), C-reactive protein (CRP), creatinine, estimated glomerular filtration rate (eGFR), and urinary albumin-to-creatinine ratio (UACR) were included. Secondly, additional covariates were identified through a data-driven screening procedure. Covariate screening was performed on the basis of a change-in-estimate criterion (> 10% change in effect estimates) whilst ensuring the absence of multicollinearity (variance inflation factor < 5) [20, 21]. The results of the covariate screening are presented in Supplementary Fig. 1. On the basis of this procedure, cystatin C was included in the models. Thirdly, additional potential confounders, including Townsend deprivation index, physical activity (assessed by MET-min/week) and CKM stage, were selected in accordance with prior studies [22, 23].

Participants with missing exposure or follow-up outcome data were excluded, and the missingness for each covariate is shown in Supplementary Table 2. The missing values for all covariates were further imputed using a random forest-based approach, whereas exposure and outcome variables were not imputed to avoid introducing additional bias [10, 24].

Statistical analyses

All analyses were performed using R (version 4.5.2) and Free Statistics software (version 2.4). All statistical tests were two sided. False discovery rate (FDR) correction was applied to account for multiple testing, and P < 0.05 was considered statistically significant. Additional methodological details are provided in the Supplementary Methods.

Descriptive analysis

Continuous variables were presented as mean ± standard deviation (SD) for normally distributed data or median (interquartile range, IQR) for non-normally distributed data, whereas categorical variables were expressed as number (%). Group differences were assessed using t-test, Pearson’s chi-squared test, or the Wilcoxon rank-sum test, as appropriate.

Biochemical and metabolic features of GDF-15

Linear regression analyses were conducted to evaluate the associations of GDF-15 with haematological parameters, blood biochemical markers, insulin resistance indices (triglyceride-glucose index [TyG], C-reactive protein-triglyceride glucose index [CTI], triglyceride to high density lipoprotein cholesterol ratio [TG/HDL-C], and estimated glucose disposal rate [eGDR]) [22], systemic inflammatory indices (systemic immune inflammation index [SII] and systemic inflammation response index [SIRI]) [25], liver-related indices (fatty liver index [FLI], fibrosis-4 score[FIB-4], aspartate aminotransferase to platelet ratio index [APRI]) [26, 27], cardiac troponin I (cTnI) and N-terminal prohormone of brain natriuretic peptide (NT-proBNP). The detailed calculation methods for these indices are described in Supplemental Methods. All variables were standardised to facilitate direct comparison of effect estimates.

Density plots and violin plots were used to visualise the distribution of GDF-15 levels across different CKM stages to further characterise the association between GDF-15 and overall CKM health status. Additionally, multinomial logistic regression was applied to examine the association between increasing GDF-15 levels and the risk of being in a higher CKM stage and each individual stage, and to estimate the predicted probabilities of each CKM stage across different levels of GDF-15.

GDF-15 and incident ASCVD and MASLD

In the primary analysis, Aalen–Johansen cumulative incidence curves were used to depict the cumulative incidence of ASCVD and MASLD during follow-up across participants with extreme (top 10% and bottom 10%) and median (middle 10%) levels of GDF-15, with all-cause mortality treated as a competing event and between-group differences assessed by Gray’s test. Separate analyses were performed across GDF-15 quartiles.

Fine-Gray competing risk regression models were applied to evaluate the association between increasing GDF-15 levels and the risk of incident ASCVD and MASLD and their comorbidity (coexistence of both conditions, regardless of which condition occurred first) during follow-up, with outcomes assessed as single-state events. Model 1 was adjusted for age and sex. Model 2 was additionally adjusted for cardiometabolic risk factors, including BMI, DM, SBP, LDL-C, creatinine, CRP, and smoking status. Model 3 was additionally adjusted for HDL-C, eGFR, UACR and glycated haemoglobin. Model 4 further included additional potential confounders, including cystatin C, Townsend deprivation index, physical activity, and CKM stage.

Subgroup analyses were conducted in accordance with age (< 65 years vs. ≥ 65 years), sex (male vs. female), overweight status (yes vs. no), hypertension (yes vs. no), hypertriglyceridemia (yes vs. no), DM (yes vs. no), MetS (yes vs. no), CKD (yes vs. no), CRP levels (< 1, 1–3 or > 3 mg/L), and CKM stage (stages 0/1, 2 or 3). Interaction effects were evaluated using likelihood ratio (LR) tests.

GDF-15 and bidirectional progression between ASCVD and MASLD

Multi-state Markov model (MSM) was applied to investigate the association between GDF-15 levels and the progression between ASCVD and MASLD during follow-up [28]. Four possible transition pathways were considered: baseline (free of disease) → ASCVD, baseline → MASLD, ASCVD → MASLD and MASLD → ASCVD. For participants entering two states on the same date, the entry date of the earlier state was defined as 0.5 days before that of the subsequent state, consistent with previous studies [29].

Transition probabilities for each disease state at 1, 5, 10 and 15 years were further estimated for participants with GDF-15 levels in the top 10% and top 5%. Additionally, whether incident CKD, hypertension or type 2 DM during follow-up modified the association between GDF-15 and the outcomes was examined using the MSM framework [30].

Incremental value of GDF-15 within established disease prediction frameworks

Two complementary approaches were used to evaluate the importance of GDF-15 relative to cardiometabolic risk factors in outcome prediction. Firstly, a Fine-Gray competing risk regression model that included GDF-15 and cardiometabolic risk factors was constructed, and the survival SHapley Additive ExPlanations (SurvSHAP) framework was applied to quantify and visualise the average variable importance across follow-up and to model the relationship between GDF-15 and model output [31, 32]. Secondly, variable importance was assessed using the LR Chi² statistic minus the degrees of freedom (Chi² − df), where larger values indicate greater contribution to the model [23].

This study further examined whether adding GDF-15 and other novel biomarkers improved the performance of established risk prediction scores, including PREVENT [2] and Systematic COronary Risk Evaluation 2 (SCORE2) [33] for ASCVD, and FLI, FIB-4 and APRI for MASLD prediction [27]. Risk discrimination was assessed using the time-dependent area under the receiver operating characteristic curve (AUC), overall prediction error was evaluated using the Brier score, and reclassification performance was evaluated using continuous net reclassification improvement (NRI) and integrated discrimination improvement (IDI) [34, 35]. These metrics were evaluated at 1, 5, 10, and 15 years, with confidence intervals estimated using perturbation-resampling. To summarize biomarker performance across multiple metrics and time horizons, standardized effect sizes and statistical evidence, defined as − log10(P value), were combined to generate an overall composite score for biomarker ranking, with higher ranks indicating greater incremental predictive value.

Sensitivity analyses

Several sensitivity analyses were performed. Firstly, haemorrhagic stroke (ICD-10 codes I60-I62 and I64) was additionally included in the definition of ASCVD. Secondly, among participants with CKM stage 3, cardiovascular risk was evaluated using the SCORE2 risk algorithm. High predicted 10-year CVD risk was defined according to age-specific SCORE2 thresholds as follows: ≥ 7.5% for individuals aged < 50 years, ≥ 10% for those aged 50–69 years, and ≥ 15% for those aged ≥ 70 years [33]. Thirdly, Cox proportional hazards regression models were used to replace the competing risk regression models in the primary analyses.

Results

Baseline characteristics

The baseline characteristics of participants are shown in Table 1 and Supplementary Table 3. A total of 29,697 participants with CKM stages syndrome 0–3 were included in the study. The mean age was 56.16 years (SD = 8.20), and 57.32% were female. The median follow-up duration was 16.15 years (IQR 15.29–16.97 years) for ASCVD and 15.63 years (IQR 14.91–16.38 years) for MASLD. During follow-up, 2786 participants developed ASCVD and 456 developed MASLD. Participants who developed disease were more likely to be male, and they had a higher Townsend deprivation index; a higher prevalence of current smoking; and higher levels of BMI, DM prevalence, SBP, CRP, Glycated haemoglobin, Creatinine, and Cystatin C.

Table 1.

Baseline characteristics of participants according to incident disease status during follow-up

Characteristic	Overall N = 29,697	Without ASCVD N = 26,911	ASCVD N = 2,786	P value	Without MASLD N = 29,241	MASLD N = 456	P value
Age at recruitment, years				< 0.001			0.74
Mean (SD)	56.16 (8.20)	55.73 (8.20)	60.36 (6.90)		56.17 (8.20)	56.04 (8.10)
Sex, n (%)				< 0.001			0.03
Female	17,022 (57.32)	15,885 (59.03)	1,137 (40.81)		16,785 (57.40)	237 (51.97)
Male	12,675 (42.68)	11,026 (40.97)	1,649 (59.19)		12,456 (42.60)	219 (48.03)
Townsend deprivation index				0.003			< 0.001
Median (IQR)	−2.21 (−3.69, 0.39)	−2.22 (−3.69, 0.35)	−2.04 (−3.67, 0.76)		−2.22 (−3.70, 0.37)	−1.33 (−3.15, 1.95)
Current smoking, n (%)				< 0.001			< 0.001
No	26,728 (90.00)	24,346 (90.47)	2,382 (85.50)		26,343 (90.09)	385 (84.43)
Yes	2,969 (10.00)	2,565 (9.53)	404 (14.50)		2,898 (9.91)	71 (15.57)
Physical activity, MET-min/week				0.01			< 0.001
Median (IQR)	2,552.65 (1,102.00, 3,554.31)	2,533.45 (1,112.00, 3,542.46)	2,763.66 (1,042.50, 3,722.24)		2,556.00 (1,112.00, 3,558.00)	2,259.50 (693.00, 3,400.56)
BMI, kg/m2				< 0.001			< 0.001
Mean (SD)	26.98 (4.44)	26.86 (4.39)	28.14 (4.70)		26.93 (4.42)	29.95 (4.72)
DM, n (%)				< 0.001			< 0.001
No	28,059 (94.48)	25,630 (95.24)	2,429 (87.19)		27,666 (94.61)	393 (86.18)
Yes	1,638 (5.52)	1,281 (4.76)	357 (12.81)		1,575 (5.39)	63 (13.82)
SBP, mmHg				< 0.001			0.003
Mean (SD)	136.11 (18.02)	135.39 (17.84)	143.07 (18.26)		136.07 (18.02)	138.57 (18.03)
LDL-C, mmol/L				0.02			0.49
Mean (SD)	3.60 (0.85)	3.60 (0.84)	3.64 (0.91)		3.60 (0.85)	3.58 (0.84)
HDL-C, mmol/L				< 0.001			< 0.001
Mean (SD)	1.48 (0.37)	1.49 (0.37)	1.37 (0.36)		1.48 (0.37)	1.32 (0.36)
CRP, mg/L				< 0.001			< 0.001
Median (IQR)	1.24 (0.63, 2.55)	1.20 (0.61, 2.48)	1.60 (0.83, 3.29)		1.22 (0.62, 2.52)	2.35 (1.16, 4.54)
Glycated haemoglobin, mmol/mol				< 0.001			< 0.001
Mean (SD)	35.64 (6.05)	35.42 (5.67)	37.74 (8.67)		35.60 (5.99)	38.07 (8.99)
Creatinine, μmol/L				< 0.001			0.41
Mean (SD)	71.71 (16.86)	71.15 (15.76)	77.11 (24.51)		71.70 (16.85)	72.39 (17.59)
Cystatin C, mg/L				< 0.001			< 0.001
Mean (SD)	0.90 (0.16)	0.89 (0.15)	0.98 (0.24)		0.89 (0.16)	0.95 (0.18)
GDF-15				< 0.001			< 0.001
Median (IQR)	−0.08 (−0.37, 0.25)	−0.10 (−0.39, 0.21)	0.18 (−0.14, 0.57)		−0.08 (−0.37, 0.24)	0.10 (−0.23, 0.53)
CKM stage, n (%)				< 0.001			< 0.001
Stage 0	2447 (8.24)	2,383 (8.86)	64 (2.30)		2,437 (8.33)	10 (2.19)
Stage 1	2489 (8.38)	2,383 (8.86)	106 (3.80)		2,474 (8.46)	15 (3.29)
Stage 2	23,981 (80.75)	21,634 (80.39)	2,347 (84.24)		23,581 (80.64)	400 (87.72)
Stage 3	780 (2.63)	511 (1.90)	269 (9.66)		749 (2.56)	31 (6.80)

ASCVD, atherosclerotic cardiovascular disease; BMI, body mass index; CKM, cardiovascular-kidney-metabolic; CRP, C-reactive protein; DM, diabetes mellitus; GDF-15, growth/differentiation factor-15; IQR, interquartile range; LDL-C, low-density lipoprotein cholesterol; MASLD, metabolic dysfunction-associated steatotic liver disease; SBP, systolic blood pressure; SD, standard deviation; HDL-C, high-density lipoprotein cholesterol

The median NPX of GDF-15 amongst individuals who developed ASCVD and MASLD was 0.18 and 0.10, respectively. Amongst these participants, 84.24% and 87.72% were in CKM stage 2, whereas 9.66% and 6.80% were in stage 3, respectively.

GDF-15, biochemical and metabolic features, and CKM health

After adjustment for confounders, GDF-15 was found to be significantly associated with 55 biomarkers. The strongest associations were observed with renal function markers, including cystatin C and creatinine, followed by FLI and insulin resistance indices (CTI and eGDR), as shown in Supplementary Fig. 2.

As shown in Fig. 2A and B, descriptive analyses indicated that the NPX of GDF-15 increased in a stepwise manner with advancing CKM stage. Multinomial logistic regression further showed that each 1-unit increase in GDF-15 was associated with a 44% increase in the risk of a higher CKM stage (OR = 1.44, 95% CI 1.32–1.57), after adjustment for confounders (Fig. 2C). Stage-specific analyses demonstrated a progressively stronger association between increased GDF-15 and the risk of more advanced CKM stages. Figure 2D illustrates the predicted probability distribution of CKM stages across different GDF-15 levels. As GDF-15 increased, the predicted probability of CKM stages 0 or 1 gradually declined. The probability of stage 2 initially increased and then decreased, peaking at an NPX of 0.57, whereas the probability of stage 3 increased continuously. When NPX exceeded 4, the model predicted that most individuals would be classified as stage 3.

Fig. 2.

Distribution and associations of GDF-15 across CKM stages. A Density distribution of GDF-15 by CKM stage. B Violin plots of GDF-15 levels across CKM stages. C Association between GDF-15 and CKM stages using multinomial logistic regression. D Predicted probabilities of CKM stages across GDF-15 levels. Models were adjusted for age, sex, BMI, DM, SBP, LDL-C, creatinine, CRP and smoking status. Abbreviations: CKM, cardiovascular-kidney-metabolic; GDF-15, growth differentiation factor-15

Association of GDF-15 with ASCVD and MASLD

Supplementary Figure 3 presents Aalen-Johansen curves for the cumulative incidence of ASCVD and MASLD. The risk of events increased progressively with increasing levels of the NPX of GDF-15, and the differences between groups were statistically significant (Gray’s Test, P < 0.001).

Each 1-unit increase in the NPX of GDF-15 was associated with a 104% increase (HR = 2.04, 95%CI 1.89–2.22) in the risk of ASCVD, and this association remained significant after adjustment for demographic characteristics, cardiometabolic risk factors and additional potential confounders (Fig. 3A). The C-statistic for predicting ASCVD by using GDF-15 alone was 0.661, which increased to 0.702 after adding age and sex and further to 0.719 after additional adjustment for cardiometabolic risk factors. For MASLD, each 1-unit increase in the NPX of GDF-15 was associated with 89% (HR = 1.89, 95%CI 1.70–2.11) increased risk, with a C-statistic of 0.611 for GDF-15 alone. For the comorbidity of ASCVD and MASLD, each 1-unit increase in the NPX of GDF-15 was associated with 134% increased risk (HR = 2.34, 95%CI 2.00–2.74), with a C-statistic of 0.692 when using GDF-15 alone. Subgroup analyses across different risk strata showed no significant heterogeneity (P for interaction > 0.05) in the associations after correction for multiple testing (Fig. 3B–D).

Fig. 3.

Associations of GDF-15 with incident ASCVD and MASLD and their comorbidity during follow-up. A Hazard ratios for incident ASCVD and MASLD and their comorbidity according to GDF-15 increase. Model 1 was adjusted for age and sex. Model 2 was further adjusted for BMI, DM, SBP, LDL-C, creatinine, CRP and smoking status. Model 3 was further adjusted for HDL-C, eGFR, UACR and glycated haemoglobin. Model 4 was further adjusted for cystatin C, Townsend deprivation index, physical activity, and CKM stage. B Subgroup analysis for ASCVD. C Subgroup analysis for MASLD. D Subgroup analysis for the comorbidity of ASCVD and MASLD. “ns” indicated no statistical significance (P for interaction > 0.05), whereas “*” indicated P < 0.05. Abbreviations: ASCVD, atherosclerotic cardiovascular disease; BMI, body mass index; CI, confidence interval; CKD, chronic kidney disease; CKM, cardiovascular-kidney-metabolic; CRP, C-reactive protein; DM, diabetes mellitus; eGFR, estimated glomerular filtration rate; FDR, false discovery rate; GDF-15, growth differentiation factor-15; HDL-C, high-density lipoprotein cholesterol; HR, hazard ratio; LDL-C, low-density lipoprotein cholesterol; MASLD, metabolic dysfunction–associated steatotic liver disease; MetS, metabolic syndrome; SBP, systolic blood pressure; SCORE, Systematic COronary Risk Evaluation; UACR, urinary albumin-to-creatinine ratio

Association of GDF-15 with bidirectional progression between ASCVD and MASLD

Figure 4A and B illustrate the associations of the NPX of GDF-15 with bidirectional transitions between ASCVD and MASLD. Amongst participants with ASCVD as the initial event, each 1-unit increase in the NPX of GDF-15 was associated with a 55% increase (HR = 1.55, 95%CI, 1.03–2.34) in the risk of subsequent MASLD. Conversely, amongst those with MASLD as the initial event, each 1-unit increase in the NPX of GDF-15 was associated with a 59% increase (HR = 1.59, 95%CI 1.01–2.50) in the risk of subsequent ASCVD.

Fig. 4.

Associations of GDF-15 with the bidirectional progression between ASCVD and MASLD. A State transition diagram of multi-state model. B Results of multi-state model for each transition pathway. The model was adjusted for age, sex, BMI, DM, SBP, LDL-C, creatinine, CRP, smoking status, HDL-C, eGFR, UACR, glycated haemoglobin, cystatin C, Townsend deprivation index, physical activity, and CKM stage. When the number of outcome events was limited, covariate selection was reperformed to avoid model overfitting and overadjustment. C Transition probabilities for each disease state at 1, 5, 10, and 15 years amongst participants with GDF-15 levels in the top 10% of the study population. D Transition probabilities amongst participants with GDF-15 levels in the top 5%. Abbreviations: ASCVD, atherosclerotic cardiovascular disease; CI, confidence interval; GDF-15, growth differentiation factor-15; HR, hazard ratio; MASLD, metabolic dysfunction–associated steatotic liver disease

Figure 4C and Supplementary Table 4 show the transition probabilities across disease states at 1, 5, 10 and 15 years for participants with the NPX of GDF-15 in the top 10%. From baseline to ASCVD, the transition probability increased from 1.05 to 23.24% over 15 years, markedly higher than that for baseline to MASLD (0.20–3.08%). Moreover, progression from MASLD to ASCVD was substantially more common than the reverse (15-year probability: 34.95% vs. 5.73%). Amongst individuals with the NPX of GDF-15 in the top 5%, the probability of developing ASCVD increased further, reaching 20.05% within 10 years, whereas the probability of MASLD remained relatively low at 2.32% (Fig. 4D and Supplementary Table 4).

As shown in Supplementary Table 5, when CKD, hypertension, and type 2 DM were considered as potential mediators, the association between GDF-15 and outcomes remained significant amongst individuals without these intermediate events. This finding suggested that the effects of GDF-15 on ASCVD and MASLD could not be fully explained by increased risks of CKD, hypertension, or type 2 DM.

Incremental value of GDF-15

SHAP and LR analyses indicated that GDF-15 played an important role in predicting outcomes beyond cardiometabolic risk factors (Fig. 5). For ASCVD, GDF-15 ranked third in importance based on SHAP, following age and sex, whereas it ranked fifth in LR analysis, following age, sex, SBP, and smoking status (Fig. 5A). For MASLD, GDF-15 ranked third in SHAP, following BMI and HDL-C, and second in LR analysis, second only to BMI (Fig. 5B). For the comorbidity of ASCVD and MASLD, GDF-15 ranked second in SHAP, following HDL-C, and ranked first in LR analysis (Fig. 5C).

Fig. 5.

Variable importance of GDF-15 in predicting ASCVD and MASLD and their comorbidity amongst cardiometabolic risk factors. A Variable importance for ASCVD prediction assessed using SHAP (beeswarm plot) and likelihood ratio Chi² statistics minus degrees of freedom (lollipop plot). B Variable importance for MASLD prediction. C Variable importance for comorbidity of ASCVD and MASLD prediction. Abbreviations: ASCVD, atherosclerotic cardiovascular disease; CI, confidence interval; GDF-15, growth differentiation factor-15; HR, hazard ratio; MASLD, metabolic dysfunction–associated steatotic liver disease; SHAP, SHapley Additive exPlanations

Figure 6 shows the improvements in risk discrimination overall prediction error, and risk reclassification after incorporating novel biomarkers into established risk scores. For ASCVD prediction, GDF-15 provided the greatest improvement amongst individuals with CKM stages 0–3 (Fig. 6A–B). Adding GDF-15 to SCORE2 resulted in a modest but significant increase 10y-AUC (Δ = 0.0056), accompanied by an improvement in the Brier score (Δ =  − 0.00051), an IDI of 0.013, and an NRI of 0.119. Similar improvements were observed for the PREVENT score (Brief score improvement = −0.00039, IDI = 0.012, NRI = 0.107), outperforming several other biomarkers related to insulin resistance, systemic inflammation, apolipoprotein A/B, lipoprotein(a), cTnI, and NT-proBNP. For MASLD prediction, GDF-15 showed the greatest overall incremental predictive value for FLI and ranked second only to CTI for FIB-4 and APRI (Fig. 6C–D).

Fig. 6.

Incremental predictive value of novel biomarkers beyond established risk scores for ASCVD and MASLD prediction. A Ranking of biomarkers according to overall composite scores for incremental predictive value in ASCVD prediction. B Incremental changes in time-dependent AUC, Brier score, IDI, and NRI after adding GDF-15 to SCORE2 and PREVENT for ASCVD prediction. C Ranking of biomarkers according to overall composite scores for incremental predictive value in MASLD prediction. D Incremental changes in time-dependent AUC, Brier score, IDI, and NRI after adding GDF-15 to FLI, FIB-4, and APRI for MASLD prediction. Abbreviations: ApoA, apolipoprotein A; ApoB, apolipoprotein B; APRI, aspartate aminotransferase to platelet ratio index; ASCVD, atherosclerotic cardiovascular disease; AUC, area under the receiver operating characteristic curve; CRP, C-reactive protein; CTI, C-reactive protein-triglyceride glucose index; cTnI, cardiac troponin I; eGDR, estimated glucose disposal rate; FIB-4, fibrosis-4 score; FLI, fatty liver index; GDF-15, growth differentiation factor-15; IDI, integrated discrimination improvement; Lp(a), lipoprotein(a); MASLD, metabolic dysfunction-associated steatotic liver disease; NRI, net reclassification improvement; NT-proBNP, N-terminal prohormone of brain natriuretic peptide; PREVENT, Predicting Risk of cardiovascular disease EVENTs; SCORE, Systematic COronary Risk Evaluation; SII, systemic immune-inflammation index; SIRI, systemic inflammation response index; TG/HDL-C, triglyceride to high-density lipoprotein cholesterol ratio; TyG, triglyceride-glucose index

Sensitivity analyses

Consistent and significant associations of GDF-15 with ASCVD, MASLD, and their comorbidity were observed when altering the definition of ASCVD outcomes (Supplementary Figs. 4–5), re-evaluating CKM stages (Supplementary Figs. 6–7), and replacing competing risk regression with Cox regression models (Supplementary Figs. 8–9), although some numerical estimates and variable rankings changed slightly across analyses.

Discussion

In this cohort study from UKB-PPP, the results showed that higher levels of GDF-15 were associated with poorer CKM health status and independently related to the incidence and bidirectional progression of ASCVD and MASLD amongst individuals with CKM stages 0–3. These associations remained consistent across disease states and subgroups. Moreover, GDF-15 showed substantial importance amongst traditional cardiometabolic risk factors and provided significant incremental value for existing risk prediction systems, outperforming several recently proposed biomarkers.

Beyond traditional cardiometabolic risk factors, AHA has highlighted that biomarkers reflecting target organ injury (e.g., high-sensitivity cardiac troponin [hs-TnT] and BNP), inflammation (e.g., CRP), or subclinical disease (e.g., coronary artery calcium), although not included in recent cardiovascular risk prediction equations such as PREVENT and SCORE2, may provide additional value when integrated into risk assessment for the general population [2]. Several emerging biomarkers have recently been proposed to improve risk prediction among individuals with CKM syndrome stages 0–3. For example, using data from 7,364 participants with CKM syndrome in the China Health and Retirement Longitudinal Study, Hong et al. reported that the TyG index and its derived indicators independently predicted CVD, particularly TyG-WC and TyG-WHtR, with C-statistics around 0.621 [36]. These findings have been confirmed in multiple cohorts from China, the United Kingdom, and the United States [11, 37, 38]. Additionally, Qin et al. analysed 423,701 participants from UKB and showed that systemic inflammatory biomarkers, such as SII and SIRI, were independently and dose-dependently associated with cardiovascular outcomes, although their addition to SCORE2 resulted in only minimal improvements in discrimination (ΔC-statistic: 0.0004 and 0.0018, respectively) [25]. Similarly, Kurt et al. reported that CRP showed greater predictive importance for major adverse cardiovascular events than diabetes, BMI, LDL-C, or creatinine in the general population, with a continuous NRI of 2.4% when added to SCORE2 [23]. This finding is also supported by analyses within the UKB-PPP cohort used in the present study.

Some researchers recently proposed the concept of cardiovascular-kidney-liver-metabolic (CKLM) syndrome [39]. CKM syndrome and MASLD share substantial pathophysiological mechanisms, including visceral adiposity, insulin resistance, and atherosclerosis, with nearly 99% of individuals with MASLD having at least one core metabolic risk factor [39]. These shared mechanisms may trigger cross-organ inflammatory and fibrotic cascades that amplify systemic risk. Evidence from several international cohorts, such as the VCTE-Prognosis cohort, NHANES III and the Chinese Kailuan cohort, has demonstrated that CKM syndrome is associated with increased risks of advanced fibrosis and adverse outcomes in MASLD [40, 41]. Although FIB-4 is widely used as a non-invasive test for MASLD screening, its performance can be suboptimal in certain metabolic or high-risk populations, leading to relatively high rates of missed diagnoses [42]. Other studies have suggested that metabolic indicators related to insulin resistance may better capture the pathophysiological features of MASLD and predict its development and related cardiovascular complications [43, 44]. In the present study, several of these biomarkers showed incremental value in existing prediction models in terms of risk discrimination and reclassification. However, their contributions to ASCVD prediction were generally smaller than that of GDF-15. For MASLD prediction, indices reflecting inflammation and insulin resistance, such as CTI, provided greater incremental improvement.

GDF-15 has been proposed as a potential biomarker for CKLM syndrome management and as a possible therapeutic target [45]. It has been recognised as a systemic biomarker reflecting oxidative stress, inflammation, and cellular aging [3]. Under physiological conditions, GDF-15 is expressed at low levels but markedly upregulated in response to inflammation, oxidative stress, hypoxia, tissue injury, and oncogene activation [46, 47]. In many contexts, GDF-15 appears to trigger adaptive tissue-protective responses, including antiproliferative, anti-inflammatory, and anti-apoptotic effects [48]. The metabolic effects of GDF-15 are mediated through binding to the specific receptor GDNF family receptor α-like (GFRAL) [47]. Upon ligand binding, GFRAL forms signaling complex with the co-receptor RET, which activates downstream pathways in hindbrain neurons, particularly in the area postrema [48]. Experimental studies demonstrated that GDF-15 primarily regulated energy balance by reducing food intake and enhancing energy expenditure in muscle, thereby leading to decreased body weight and improved metabolic parameters. These effects were shown to require a GFRAL–β-adrenergic–dependent signaling axis [47, 49].

From the perspective of disease management and screening, GDF-15 may represent a marker of subclinical disease risk through distinct biological pathways. The present study and previous studies suggest that its predictive value was independent of, and in some cases superior to, traditional subclinical risk markers such as NT-proBNP and hs-TnT [50]. These findings suggest that GDF-15 may capture residual biological risk not fully reflected by conventional cardiometabolic assessment frameworks. Notably, participants with GDF-15 levels in the top 5% had an approximately 30% probability of developing ASCVD within 15 years, suggesting that GDF-15 measurement may help identify individuals with CKM syndrome who could benefit from intensified preventive strategies, closer longitudinal monitoring, or broader cardiometabolic evaluation. However, because GDF-15 levels may increase across a wide range of pathological conditions [13, 14], further studies are needed to determine its optimal clinical thresholds, integration with existing risk algorithms, and potential implications for treatment decision-making before routine clinical implementation. From a therapeutic perspective, the pleiotropic biological effects of GDF-15 suggest potential value as a treatment target. The development of GDF-15–targeted therapies remains at an early stage. Replacement strategies, such as CIN-109, are being explored for obesity, whereas inhibition of excessive GDF-15 signaling may benefit conditions characterised by pathological weight loss, highlighting its context-dependent effects [51]. For example, Ponsegromab, a monoclonal antibody targeting GDF-15, increased body weight and improved appetite and physical activity in patients with cancer cachexia in a phase II trial [52].

This study, based on a large proteomics cohort, clarified the predictive and potential management value of GDF-15 for ASCVD and MASLD amongst individuals with CKM stages 0–3. However, several limitations should be acknowledged. Firstly, although the sample size was large, the analysis was conducted within a single-country cohort, which may limit the generalisability of the findings. Future studies incorporating multi-country and multi-ethnic populations are needed. Secondly, causal relationships could not be inferred considering that this study was an observational cohort study. Thirdly, in the UKB-PPP dataset, all proteins were quantified using NPX, and therefore, the original concentration values of GDF-15 and specific diagnostic thresholds could not be determined. Fourthly, GDF-15 was measured only once at baseline, which may not capture longitudinal changes in biomarker levels over time. Fifthly, the outcomes in this study were defined based on ICD-10 codes; although these outcomes have relatively high PPVs, misclassification remains possible.

Conclusions

GDF-15 was significantly associated with CKM health status, and it predicted the incidence and bidirectional progression of ASCVD and MASLD amongst individuals with CKM syndrome stages 0–3. These findings suggest that GDF-15 may serve as a promising biomarker for risk stratification and management of CKM-related diseases and represent a potential therapeutic target within current CKM prevention and treatment frameworks.

Author contributions

Authorship contribution statement: Conceptualization: X.C., H.T., Y.C.; Data Management and Analysis: X.C., H.T., Z.X.; Figure Creation: X.C., H.T., Z.X., XY.C, C.T.; Writing—Original Draft Preparation: N.L., J.H., H.L, X.Z., Q.Y., K.L.; Writing—Review and Editing: P.C., XH.C., L.J., W.L., W.C., Y.Z.; Provided Critical Revisions to the Manuscript: X.C., H.T., J.L., X.T. Y.C.; Project Management: J.L., X.T., Y.C.

Funding

This research was supported by Provincial Science and Technology Special Fund of Guangdong in 2021 (No. 2021-88-53), Provincial Science and Technology Special Fund of Guangdong in 2022 (No. 2022-124-6), Fund from National Health Commission Medical and Health Science and Technology Development and Research Center (No. WKZX2022JG0138), Innovation Team Project of Guangdong Universities, China (Natural, No. 2024KCXTD019), and Grant for Key Disciplinary Project of Clinical Medicine under the High-level University Development Program, Guangdong, China (No. 2024–2025). Role of the funding source: The funders of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. All authors had full access to all the data in the study and accepted responsibility to submit for publication.

Data availability

Data Availability: The raw data supporting the findings of this study are available from the UK Biobank (https://www.ukbiobank.ac.uk/) under restricted access due to data privacy and ethical restrictions. Access to these data can be obtained by submitting an application directly to the UK Biobank. This research was conducted under UK Biobank application number 205837.

Declarations

Ethics approval

The UK Biobank was approved by the North West Multicenter Research Ethics Committee, with all participants providing written informed consent. Ethical approval and informed consent were waived as the UK Biobank data is publicly available and does not include identifiable information.

Competing interests

The authors declare no competing interests.