Abstract
BACKGROUND:
Cardio-kidney-metabolic (CKM) disease represents a significant public health challenge. While proteomics-based risk scores (ProtRS) enhance cardiovascular risk prediction, their utility in improving risk prediction for a composite CKM outcome beyond traditional risk factors remains unknown.
METHODS:
We analyzed 23 815 UK Biobank participants without baseline CKM disease, defined by International Classification of Diseases-Tenth Revision codes as cardiovascular disease (coronary artery disease, heart failure, stroke, peripheral arterial disease, atrial fibrillation/flutter), kidney disease (chronic kidney disease or end-stage renal disease), or metabolic disease (type 2 diabetes or obesity). The sample was randomly divided into ProtRS training (70%, N=16 671) and validation (30%, N=7144) cohorts. A least absolute shrinkage and selection operator-based Cox regression model of 2913 Olink-based proteins was utilized to develop the ProtRS in the training cohort. We then assessed the association of the ProtRS with incident CKM disease risk in the validation cohort with competing-risk regression after adjusting for traditional risk factors and evaluated its ability to discriminate incident CKM disease risk with C-indices.
RESULTS:
The study sample had a mean age of 56.1 years; 44% were male, and 94% were White. Over a median follow-up of 13.5 years, 3235 and 1407 incident CKM disease events occurred in the training and validation cohorts, respectively. A ProtRS based on the weighted sum of the 238 least absolute shrinkage and selection operator-selected proteins was significantly associated with incident CKM disease risk (subdistribution hazard ratio per 1-SD, 1.87 [95% CI, 1.73–2.03]; P<0.001) in the validation cohort after adjustment for traditional risk factors. The addition of the ProtRS to a traditional risk factor model significantly improved incident CKM disease risk discrimination beyond the traditional risk factor model (C-index, 0.73 [0.72–0.74] versus 0.71 [0.69–0.72]; ΔC-index, 0.03 [0.02–0.04]).
CONCLUSIONS:
A ProtRS was independently associated with incident CKM disease risk and improved risk prediction beyond traditional risk factors in a population free of CKM disease at baseline.
Keywords: atrial fibrillation, heart failure, kidney, proteomics, risk factors
Cardio-kidney-metabolic (CKM) diseases represent a significant public health challenge. CKM diseases affect >25% of American adults and were the leading cause of death in 2021.1–4 The American Heart Association has recognized the interrelated nature of cardiovascular, kidney, and metabolic disease (MD)s, emphasizing their shared preventability and potential lethality. The underlying processes driving the development of CKM diseases remain a topic of active research, with inflammation, metabolic dysregulation, and dysfunctional adipose tissue thought to be implicated among the many pathobiological pathways involved in CKM disease pathogenesis.5 Given the substantial burden of CKM disease-related morbidity and mortality, novel risk prediction techniques are needed to better predict and ultimately reduce the burden of CKM disease worldwide.
High-throughput proteomics platforms allow for the quantification of thousands of proteins from a single blood specimen, enabling large-scale estimation of proteomic profiles across population samples.6,7 Such technology provides the ability to uncover relationships between circulating proteins and disease risk with unprecedented scale and precision. Proteomics-based risk scores (ProtRS) have previously been shown to improve risk discrimination for incident cardiovascular disease (CVD) beyond traditional risk factors in primary prevention populations, and similar large-scale proteomics-based profiling has been utilized to identify proteins associated with increased risk for incident kidney and MDs.8–12 However, it remains unknown whether a ProtRS could improve risk prediction for incident composite CKM disease beyond traditional risk factors. A ProtRS for incident CKM disease incorporating proteins contributing to the shared risk of individual CKM diseases might more accurately predict a composite CKM outcome.
Herein, we leveraged the UK Biobank, a resource with rich clinical, risk factor, and proteomics data, to develop and validate a ProtRS for incident CKM disease and demonstrate its potential utility in risk prediction.
METHODS
Study Population
The UK Biobank is a population-based, prospective registry of ≈500 000 participants aged 40 to 69 years who were recruited from across the United Kingdom between 2006 and 2010.13 The recruitment criteria for the UK Biobank have been described previously.14 UK Biobank data are available to researchers who gain approval from the UK Biobank access management committee (https://www.ukbiobank.ac.uk/). The current study included UK Biobank participants without CKM disease at baseline. CKM disease was defined by the distinct disease states represented in the American Heart Association definition of CKM syndrome by International Statistical Classification of Diseases and Related Health Problems (ICD), Ninth (ICD-9) and Tenth Revision (ICD-10), codes and consisted of CVD (coronary artery disease, heart failure, stroke, atrial fibrillation or atrial flutter, and peripheral vascular disease), kidney disease (KD; chronic KD or end-stage renal disease), and MD (type 2 diabetes or obesity; Table S1).5 Participants’ ICD codes were mapped to primary care records, hospital inpatient data, death register records, and self-reported medical conditions at the baseline UK Biobank assessment visit. In addition to exclusion by ICD-10 codes as above, participants were also excluded from this study for an estimated glomerular filtration rate, as calculated by the 2021 chronic KD-EPI equation, of <60 mL/min per 1.73 m2, hemoglobin A1c ≥6.5%, or body mass index of ≥30 kg/m2 at UK Biobank enrollment.15 Proteomics data collected at the time of enrollment were available in a subsample of ≈50 000 UK Biobank participants through the UK Biobank Pharma Proteomics Project.6 Only participants from this subsample and without missing clinical covariate data were included in our analysis, resulting in a study sample free of preexisting CKM disease and with proteomics data available (Figure S1).
The UK Biobank has received approval from the North West Multicenter Research Ethics Committee and holds current approval as a Research Tissue Bank, such that researchers do not require separate ethical clearance. All UK Biobank participants provided informed consent before registry enrollment.
Follow-Up and Outcomes
UK Biobank data are linked to primary care data, as well as Hospital Episode Statistics data, which include all hospital admissions until September 2023 dating back to 1997 for England, 1998 for Wales, and 1981 for Scotland. Death registries included all deaths until September 2023. The primary outcome was defined as the first occurrence of any incident CKM disease mapped to ICD-9 and ICD-10 codes, including death due to CKM disease as the primary or secondary cause, through linkage to hospital inpatient records, primary care data, death registries, and self-reported conditions. Secondary outcomes included incident CVD, KD, or MD, including outcome-related death, analyzed as separate end points. Follow-up time was defined as the time from UK Biobank enrollment until incident outcome occurrence, death, end of follow-up, or loss to follow-up.
Proteomics Measurement
In participants of the UK Biobank Pharma Proteomics Project, 2924 protein levels were measured with the Olink Explore 3072 platform, which utilized an antibody-based proximity extension assay in nonfasting venous EDTA plasma samples.6,16 Normalized protein expression levels were calculated using individual participants’ gene product levels, resulting in protein expression levels described as values relative to other UK Biobank Pharma Proteomics Project participants. These processes have been described in detail elsewhere.6,17
Traditional Risk Factors and Risk Factor Modeling
Traditional risk factors utilized in multivariable regression modeling were selected based on their inclusion in the American Heart Association Predicting Risk of CVD EVENTS (PREVENT) base equation.18 In addition to demographics (age, sex, and race), additional clinical risk factors included were smoking status at time of UK Biobank enrollment, body mass index, hemoglobin A1c, estimated glomerular filtration rate, total cholesterol, high-density lipoprotein cholesterol, hypertension, blood pressure medication use, and cholesterol medication use. For the primary analysis, ICD-10 defined prevalent hypertension was chosen over systolic blood pressure measurement at enrollment, due to a larger degree of missing systolic blood pressure data.
Proteomics-Based Risk Score Development
Participants missing ≥20% proteomics data were excluded from the study sample. Proteins with ≥20% missing values were also excluded from analyses, resulting in a total of 2913 proteins being utilized for ProtRS development. The ProtRS was developed in a training cohort consisting of a randomly selected subsample of 70% of the entire sample (16 671 participants, 3235 incident CKM disease events) and tested in a separate validation cohort, comprised of the remaining 30% (7144 participants, 1407 incident CKM disease events). Missing protein values were imputed using the minimum observed value of each respective protein, and rank-based inverse normal transformation was performed. The ProtRS was then constructed in the training cohort using all 2913 transformed protein levels in a least absolute shrinkage and selection operator (LASSO)-based Cox proportional hazards model for incident CKM disease risk.19 For this analysis, the full available follow-up was utilized to maximize protein discovery and the number of incident events. Nonzero beta coefficients of the remaining proteins after application of the LASSO penalty were multiplied by their respective protein levels in the validation cohort, the product of which was added to a cumulative score, representing individual-level risk attributable to the sum of the LASSO-selected proteins. This ProtRS was then Z-scored to ensure comparability across study participants.
Shared Protein Identification
To identify which of the LASSO-selected proteins were predictive of all individual CKM disease outcomes, LASSO-based Cox regressions for incident CVD, KD, and MD were performed for each individual outcome. All proteins with nonzero beta coefficients from the LASSO models were compared with identify shared protein predictors across individual CKM disease outcomes.
Univariable Protein Associations With Incident CKM Disease Risk
To evaluate the individual association of each protein with incident CKM disease risk in the validation cohort, we performed univariable competing-risk regression modeling of all 2913 Olink-based proteins for the outcome of incident CKM disease and adjusted for multiple comparisons with the Benjamini-Hochberg false discovery rate procedure. Proteins with false discovery rate-adjusted P<0.05 were considered statistically significant.
Pathway Analysis
Gene ontology overrepresentation analysis was then performed to identify biological processes enriched among proteins significantly associated with incident CKM disease risk in validation cohort participants with univariable competing-risk regression modeling. This analysis was conducted using the topGO R package.20 Gene ontology terms with enrichment P<0.05 were considered significantly overrepresented.
Statistical Analysis
Descriptive Statistics
Baseline characteristics of the study sample were reported as absolute counts (percentages) for categorical variables and means (SD) or medians (interquartile range) for continuous variables, as appropriate. The distribution of the ProtRS in the validation cohort was assessed with density and Q-Q plots. Multivariable regression was utilized in the validation cohort to assess the association of demographic and traditional risk factor variables with the ProtRS.
Survival Analyses
The independent association between the ProtRS and incident CKM disease risk was assessed in the validation cohort by Fine and Gray subdistribution hazard modeling, while accounting for competing non-CKM disease-related death.21 Regression modeling was adjusted for traditional risk factors (age, sex, race, smoking history, body mass index, hemoglobin A1c, estimated glomerular filtration rate, total cholesterol, high-density lipoprotein cholesterol, hypertension, blood pressure medication use, and cholesterol medication use). The association between the ProtRS and individual incident CVD, KD, and MD outcomes was conducted similarly. Subdistribution hazard ratios (sHR) for each model were reported per a 1-SD increase in the ProtRS, and comparisons between ProtRS quartiles were also performed after the normality of the ProtRS was assessed. Calibration of the ProtRS for CKM disease risk was assessed in the internal validation cohort. Participants were grouped into deciles based on predicted risk, and observed CKM event rates were estimated for each decile using the Aalen-Johansen estimator, which accounts for the competing risk of non-CKM-related death. Predicted versus observed risks were then visualized in a calibration plot. We additionally analyzed the association of the ProtRS with incident CVD risk censored at 10 years, independent of participants’ PREVENT scores, in participants of the validation cohort with nonmissing relevant clinical variables for PREVENT score calculation (N=6447/7144). For all analyses including the PREVENT score, the incident 10-year CVD outcome was defined per the PREVENT equation as the first occurrence of any incident myocardial infarction, coronary heart disease death, stroke, stroke death, heart failure, or heart failure death, and the same ProtRS developed using the full available follow-up was applied to ensure consistency in the risk score across time horizons. Furthermore, all PREVENT-related analyses utilized continuous systolic blood pressure data, rather than ICD-10-based hypertension used in the primary analysis, consistent with the PREVENT score derivation.18
The incremental value of the ProtRS to incident CKM disease risk discrimination beyond traditional factors was estimated with the C-index, and 95% CIs were calculated by bootstrapping with 200 iterations. The ProtRS was also compared with the PREVENT score for incident 10-year CVD risk discrimination, utilizing the PREVENT-based CVD definition. The survC1 R package was utilized for these analyses.22
Sensitivity analyses were performed to explore heterogeneity of effect in the ability of the ProtRS to predict incident CKM disease risk across age (≥65 years or <65 years), sex, and race groups.
All analyses were performed with R 4.2.2 (https://www.R-project.org). P values <0.05 were considered statistically significant.
RESULTS
Descriptive Statistics
The baseline characteristics of the training and validation cohorts are shown in Table 1. The mean age of the entire sample was 56.1 (SD, 8.3) years; 56% (N=13 335) were female, and 94% (N=22 457) were White. Of those who identified as non-White race (N=1358, 6%), 29% (N=398) identified as South Asian (Indian, Pakistani, Bangladeshi, or other Asian background), 27% (N=367) identified as African, Black, or Caribbean, 11% (N=150) identified as mixed race, 7% (N=97) identified as Chinese, 19% (N=255) as another ethnic group, and 7% (N=91) did not know or preferred not to answer. The distribution of all clinical risk factors and diseases was largely similar between the training and validation cohorts, except for a slight overrepresentation of participants of White race in the validation cohort (95% versus 94% in training cohort, Table 1). Over a median follow-up of 13.5 (interquartile range, 1.8) years, 4632 (19.5%) CKM disease, 3378 (14.2%) CVD, 765 (3.2%) KD, and 1272 (3.5%) MD events occurred (Table S2).
Table 1.
Baseline Characteristics of Study Sample
| Variable | Cohort | P value† | |
|---|---|---|---|
| ProtRS training cohort, N=16 671* | ProtRS validation cohort, N=7144* | ||
| Age, y | 56.1 (8.3) | 56.1 (8.2) | 0.770 |
| Male sex | 7334 (44%) | 3146 (44%) | 0.950 |
| White race | 15 681 (94%) | 6776 (95%) | 0.016 |
| Smoking history | 7216 (43%) | 3107 (43%) | 0.768 |
| BMI, kg/m2 | 25.2 (2.7) | 25.2 (2.7) | 0.233 |
| Hemoglobin A1c, % | 5.3 (0.3) | 5.3 (0.3) | 0.789 |
| eGFR, mL/min per 1.73 m2 | 96.1 (11.6) | 96.0 (11.4) | 0.541 |
| Total cholesterol, mg/dL | 224.1 (42.1) | 224.5 (41.9) | 0.769 |
| HDL-cholesterol, mg/dL | 58.7 (14.9) | 58.9 (14.9) | 0.233 |
| Hypertension | 3092 (19%) | 1325 (19%) | >0.999 |
| Incident cardio-kidney-metabolic disease | 3235 (19%) | 1407 (20%) | 0.752 |
| Incident cardiovascular disease | 2352 (14%) | 1026 (14%) | 0.780 |
| Incident kidney disease | 531 (3.2%) | 234 (3.2%) | 0.807 |
| Incident metabolic disease | 882 (5.3%) | 390 (5.5%) | 0.575 |
BMI indicates body mass index; eGFR, estimated glomerular filtration rate; HDL, and high-density lipoprotein.
Counts (percentages) are shown for categorical variables, and means (SD) and medians (interquartile range) are shown for continuous variables, as appropriate.
Wilcoxon rank sum and Pearson χ2 tests were utilized to assess for between-group differences.
Proteomics-Based Risk Score Development and Prognostic Performance
Following LASSO regression of all 2913 proteins, the ProtRS developed in the training cohort was comprised of 238 proteins (Table S3), with 48 of these proteins being shared across each individual CKM disease outcome (Nproteins=48/238, Figure S2 and Table S4). The ProtRS was significantly associated with several demographic and traditional risk factor variables (Table S5). After weighting and standardization of the ProtRS in the validation cohort, CKM disease incidence was 6.0% (N=107/1786) in the lowest ProtRS quartile and 40.4% (N=721/1786) in the highest ProtRS quartile. As the ProtRS was found to be normally distributed in the validation cohort (Figure S3), unadjusted cumulative incidence curves were constructed to visualize the association between ProtRS quartiles and the incidence of CKM disease. A significant difference in CKM disease incidence across ProtRS quartiles was observed, such that participants in the highest ProtRS quartile had the highest incidence of CKM disease and those in the lowest ProtRS quartile had the lowest (Figure).
Figure. Cumulative incidence of cardio-kidney-metabolic disease by proteomics-based risk score quartile. Cumulative incidence curves demonstrate the significant difference in cardio-kidney-metabolic (CKM) disease incidence between proteomics-based risk score (ProtRS) quartiles.
The highest incidence of CKM disease is seen in the highest ProtRS quartile (red), whereas the lowest incidence of CKM disease is seen in the lowest ProtRS quartile (green).
In a univariable competing-risk regression model accounting for competing non-CKM-related death, the ProtRS was a significant predictor of incident CKM disease risk (sHR, 2.17 per SD [95% CI, 2.06–2.29]; P<0.001). After adjustment for age, sex, race (White versus non-White), body mass index, hemoglobin A1c, total cholesterol, high-density lipoprotein cholesterol, HTN, blood pressure medication use, and cholesterol medication use, the ProtRS remained significantly associated with incident CKM disease risk (sHR, 1.87 per SD [95% CI, 1.73–2.03], P<0.001; Table 2). When compared with participants in the lowest ProtRS quartile (N=1786), those in the highest ProtRS quartile (N=1786) had a markedly higher risk of incident CKM disease risk (sHR, 4.44 per SD [95% CI, 3.40–5.78]; P<0.001) after adjustment for the aforementioned traditional risk factors (Table S6). We next assessed the calibration of the ProtRS for predicting CKM disease risk in the internal validation cohort and observed that predicted risks closely aligned with observed risks across the full risk spectrum (Figure S4). Regarding secondary outcomes, the ProtRS was also an independent predictor of CVD (sHR, 1.84 per SD [95% CI, 1.68–2.02]; P<0.001), KD (sHR, 1.94 per SD [95% CI, 1.60–2.36]; P<0.001), and MD (sHR, 1.70 per SD [95% CI, 1.46–1.97]; P<0.001), after adjustment for traditional risk factors (Table 2).
Table 2.
Association of Proteomics-Based Risk Score With Cardio-Kidney-Metabolic Disease Outcomes
| Outcome (No. participants/events) | Predictor | Covariates | Subdistribution hazard ratio (95% CI)*† |
|---|---|---|---|
| Incident cardio-kidney-metabolic Disease (7144/1407) | ProtRS | … | 2.17 (2.06–2.29) |
| ProtRS | Traditional risk factors‡ | 1.87 (1.73–2.03) | |
| Incident cardiovascular disease (7144/1026) | ProtRS | … | 2.20 (2.07–2.33) |
| ProtRS | Traditional risk factors‡ | 1.84 (1.68–2.02) | |
| Incident kidney disease (7144/234) | ProtRS | … | 2.54 (2.28–2.84) |
| ProtRS | Traditional risk factors‡ | 1.94 (1.60–2.36) | |
| Incident metabolic disease (7144/390) | ProtRS | … | 1.90 (1.73–2.09) |
| ProtRS | Traditional risk factors§ | 1.70 (1.46–1.97) |
ProtRS indicates proteomics-based risk score.
Subdistribution hazard ratio expressed per 1-SD increase in ProtRS.
All P values are <0.001.
Traditional risk factors include age, sex, race (White vs non-White), smoking history, body mass index, hemoglobin A1c, estimated glomerular filtration rate, total cholesterol, high-density lipoprotein cholesterol, hypertension, blood pressure medication use, and cholesterol medication use.
Traditional risk factors include age, sex, race (White vs non-White), body mass index, hemoglobin A1c, total cholesterol, high-density lipoprotein cholesterol, hypertension, blood pressure medication use, and cholesterol medication use
When compared with a traditional risk factor model for the discrimination of incident CKM disease risk (C-index, 0.705 [95% CI, 0.690–0.720]), a model containing only the ProtRS provided significantly stronger risk prediction (C-index, 0.720 [95% CI, 0.704–0.735]; ΔC-index, 0.015 [95% CI, 0.002–0.028]). The addition of the ProtRS to the traditional risk factor model, however, provided the most robust improvement in CKM disease risk discrimination (C-index, 0.732 [95% CI, 0.719–0.744]; ΔC-index, 0.027 [95% CI, 0.020–0.035]) when compared with traditional risk factors alone. Stability testing to ensure that our results were not overly dependent on a single 70%/30% partitioning demonstrated consistent model performance across 5 random partitions, with similar C-index values observed in each validation cohort. The mean C-index across all 5 random partitions was 0.736 for the combined ProtRS and traditional risk factor model compared with 0.694 for the risk factors only model, with an average ΔC-index of 0.042 (Table S7). The ProtRS also significantly improved risk discrimination for incident CVD (C-index, 0.738 [95% CI, 0.722–0.754]; ΔC-index, 0.023 [95% CI, 0.014–0.031]), incident KD (C-index, 0.857 [95% CI, 0.829–0.886]; ΔC-index, 0.026 [95% CI, 0.007–0.046]), and incident MD (C-index, 0.796 [95% CI, 0.773–0.820]; ΔC-index, 0.011 [95% CI, 0.003–0.020]), when compared with the base traditional risk factor model (Table 3).
Table 3.
Proteomics-Based Risk Score and Cardio-Kidney-Metabolic Disease Risk Discrimination
| Outcome (No. participants/events) | C-index (95% CI) | ΔC-index (95% CI) (compared with traditional risk factors) |
|---|---|---|
| Covariates | ||
| Incident cardio-kidney-metabolic disease (7144/1407)* | ||
| Traditional risk factors | 0.705 (0.689–0.720) | (Reference) |
| ProtRS | 0.720 (0.704–0.735) | 0.015 (0.002 to 0.028) |
| ProtRS+traditional risk factors | 0.732 (0.719–0.744) | 0.027 (0.020 to 0.035) |
| Incident cardiovascular disease (7144/1026)* | ||
| Traditional risk factors | 0.716 (0.700–0.732) | (Reference) |
| ProtRS | 0.720 (0.705–0.736) | 0.005 (−0.013 to 0.023) |
| ProtRS +traditional risk factors | 0.738 (0.723–0.754) | 0.023 (0.014 to 0.031) |
| Incident kidney disease (7144/234)* | ||
| Traditional risk factors | 0.831 (0.797–0.864) | (Reference) |
| ProtRS | 0.781 (0.744–0.818) | −0.050 (−0.088 to −0.012) |
| ProtRS +traditional risk factors | 0.857 (0.829–0.886) | 0.026 (0.007 to 0.046) |
| Incident metabolic disease (7144/390)† | ||
| Traditional risk factors | 0.785 (0.761–0.809) | (Reference) |
| ProtRS | 0.700 (0.667–0.732) | −0.086 (−0.115 to −0.056) |
| ProtRS +traditional risk factors | 0.796 (0.773–0.820) | 0.011 (0.003 to 0.020) |
ProtRS indicates proteomics-based risk score.
Traditional risk factors include age, sex, race (White vs non-White), smoking history, body mass index, hemoglobin A1c, estimated glomerular filtration rate, total cholesterol, high-density lipoprotein cholesterol, hypertension, blood pressure medication use, and cholesterol medication use.
Traditional risk factors include age, sex, race (White vs non-White), body mass index, hemoglobin A1c, total cholesterol, high-density lipoprotein cholesterol, hypertension, blood pressure medication use, and cholesterol medication use.
The ProtRS was also examined as an independent predictor of 10-year incident PREVENT-based CVD risk.18 Both the PREVENT score (sHR, 1.15 per SD [95% CI, 1.12–1.17]; P<0.001) and ProtRS (sHR, 2.40 per SD [95% CI, 2.13–2.70]; P<0.001) were predictors of incident PREVENT-based CVD risk in univariable competing-risk regression modeling accounting for non-CVD-related death. The ProtRS remained an independent predictor of incident CVD risk even after adjustment for the PREVENT score (sHR, 2.22 per SD [95% CI, 1.90–2.61], P<0.001; Table 4). When compared with the C-index of the PREVENT score to discriminate the 10-year risk of incident PREVENT-based CVD (C-index, 0.708 [95% CI, 0.686–0.734]), the addition of the ProtRS to this model significantly improved risk discrimination (C-index, 0.757 [95% CI, 0.730–0.784]; ΔC-index, 0.049 [95% CI, 0.028–0.071]; Table 5).
Table 4.
Association of Proteomics-Based Risk Score With 10-Year PREVENT-Based Cardiovascular Disease Risk
| Outcome (No. participants/events) | Predictor | Covariates | Subdistribution hazard ratio, per 1-SD (95% CI)* |
|---|---|---|---|
| Incident PREVENT-based cardiovascular disease† (6447/361) | PREVENT score | … | 1.15 (1.12–1.17) |
| ProtRS | … | 2.40 (2.13–2.70) | |
| ProtRS | PREVENT score | 2.22 (1.90–2.61) |
PREVENT indicates Predicting Risk of CVD EVENTS; and ProtRS, proteomics-based risk score.
All P<0.001.
Defined per cardiovascular disease definition utilized in development and validation of PREVENT equations (cardiovascular disease=incident myocardial infarction, coronary heart disease death, stroke, stroke death, heart failure, or heart failure death).
Table 5.
Proteomics-Based Risk and PREVENT Scores and 10-Year PREVENT-Based Cardiovascular Disease Risk Discrimination
| Outcome (No. participants/events) | C-index (95% CI) | ΔC-index (95% CI) (compared with base PREVENT score) |
|---|---|---|
| Covariates | ||
| Incident cardiovascular disease* (6447/361) | ||
| PREVENT score | 0.708 (0.682–0.734) | (Reference) |
| ProtRS | 0.753 (0.721–0.784) | 0.044 (0.016–0.073) |
| ProtRS +PREVENT score | 0.757 (0.730–0.784) | 0.049 (0.028–0.071) |
PREVENT indicates Predicting Risk of CVD EVENTS; and ProtRS, proteomics-based risk score.
Defined per cardiovascular disease definition utilized in development and validation of PREVENT equations; cardiovascular disease=incident myocardial infarction, coronary heart disease death, stroke, stroke death, heart failure, or heart failure death.
Univariable CKM Disease Risk Modeling and Pathway Analysis
Among the 2913 proteins tested in univariable competing-risk regression models, 1310 were significantly associated with incident CKM disease risk after false discovery rate correction (Table S8; Figure S5). Pathway analysis identified 107 biological pathways significantly overrepresented by the 1310 proteins significantly associated with incident CKM disease in univariable competing-risk regression analyses (P<0.05). These pathways included several thematic categories, including immune and inflammatory signaling (eg, inflammatory response, monocyte chemotaxis, chemokine-mediated signaling pathway), vascular remodeling and angiogenesis (eg, regulation of blood vessel endothelial cell migration, positive regulation of vasoconstriction, regulation of angiogenesis) and metabolic and hormonal regulation (eg, regulation of glucose metabolic process, cellular response to peptide hormones, response to aldosterone (Table S9).
Sensitivity Analyses
Regarding sensitivity analyses, the ProtRS was a significantly but modestly stronger predictor of incident CKM disease risk in women (sHR, 1.95 per SD [95% CI, 1.72–2.20]; P<0.001) when compared with men (sHR, 1.85 per SD [95% CI, 1.66–2.05], P<0.001, P-interaction=0.048). No statistical interaction was observed between the ProtRS and age or White race for the prediction of incident CKM disease risk.
DISCUSSION
In this large, prospective cohort study, we developed and validated a ProtRS that was significantly associated with incident CKM disease risk and improved risk prediction beyond traditional risk factors. The ProtRS was independently associated with a nearly 90% higher risk of CKM disease per ProtRS SD, and it significantly improved risk discrimination for incident CKM disease, as well as for individual CVD, KD, and MD outcomes when compared with traditional risk factors alone. In addition to demonstrating strong discrimination, the ProtRS showed good internal calibration for predicting CKM disease risk, further supporting its potential use for individualized risk stratification. Participants in the highest ProtRS quartile had a nearly 4.5-fold higher risk of incident CKM disease when compared with those in the lowest quartile after adjustment for traditional risk factors. Furthermore, the ProtRS was a significant predictor of the PREVENT-based 10-year incident CVD outcome, independent of participants’ PREVENT scores, and also significantly improved risk discrimination for PREVENT-based CVD risk beyond the PREVENT score. These findings demonstrate the utility of proteomic profiling from a single plasma sample for the identification of individuals at a higher risk of CKM disease and highlight the potential of ProtRS to augment existing risk stratification tools.
Our findings build upon previous research, some of which has also been conducted in the UK Biobank, that has explored proteomics-based methods for predicting cardiovascular, kidney, and MD outcomes individually.8–12,23 By focusing on the integrated risk of incident CKM disease, however, we extend the clinical relevance of proteomics-based risk profiling to a broader, interrelated spectrum of clinically important diseases. Moreover, the attenuation of nearly every risk estimate after the addition of the ProtRS to a traditional risk factor model highlights the ability of a ProtRS to capture complex associations across a diverse array of risk factors (Table S10). These findings underscore the capacity of proteomics-based risk modeling to not only improve risk prediction for CKM disease but also to provide mechanistic insights into which pathobiological processes could be driving CKM disease and potentially uncover targets for future therapeutic interventions through the identification of unique proteomic signatures.
The development of proteomics-based risk scores, such as the score developed in this study, offer prospective advancement toward more precise risk prediction for CKM disease. This approach could enable earlier detection of high-risk individuals, while also identifying underlying pathophysiological pathways implicated in those at higher risk of CKM disease. As our findings demonstrate significant improvement in CKM disease risk discrimination with the addition of a ProtRS to a traditional risk factor model, we show that a ProtRS may aid in capturing high-risk patients who might otherwise have been missed by a traditional risk assessment, which could in turn assist in implementing more timely and targeted preventive interventions. And despite traditional risk factor models outperforming the ProtRS alone for the discrimination of incident kidney and MD risk, the ability of the ProtRS to further significantly improve the discrimination for incident kidney and MD beyond already robust traditional risk factor models indicates that the ProtRS was able to capture important elements of risk undetected by traditional risk factors. Moreover, as our ProtRS also demonstrated incremental improvement even over the newly established PREVENT base equation for incident CVD risk discrimination, ProtRS hold promise for offering an additional layer of precision in cardiovascular risk stratification to the addition of prominent risk equations. A central theme of implicated biological pathways in our overrepresentation analysis were the presence of inflammatory and immune-related pathways, such as the chemokine-mediated signaling (P<0.001), inflammatory response (P=0.001), and monocyte chemotaxis pathways (P=0.002) being significantly overrepresented. These findings build upon emerging evidence identifying systemic inflammation and immune dysregulation as being significantly associated with a multitude of CKM morbidities, including insulin resistance, dysfunctional cholesterol transport, endothelial dysfunction, and renal fibrosis.24 Our identification of such pathways suggests that the ProtRS is able to integrate previously unmeasured biological vulnerability into risk assessment, providing complementary prognostic information that enhances traditional risk stratification approaches. Additional replication studies will however be required to further refine and build upon our initial findings.
A notable strength of this study was the utilization of high-throughput proteomics technology to derive the ProtRS, which provided a wide-ranging examination of the proteomic profiles of UK Biobank participants and their association with incident CKM disease risk. The significant representation of women in our cohort allowed for the detection of a statistical interaction between the ProtRS and sex for the prediction of incident CKM disease, potentially due to sex-specific differences in inflammatory processes and hormonal influences related to CKM disease risk. Last, our utilization of competing-risk modeling allowed for the direct handling of important non-CKM-related causes of death, such as noncardiovascular death, strengthening the reliability of our findings.
LIMITATIONS
As the UK Biobank primarily consists of White participants, our findings may not be generalizable to non-White individuals. This limited ethnic diversity within the UK Biobank necessitated the use of race as a binary variable in our analyses, which may oversimplify the potential impact of racial and ethnic diversity on CKM disease risk. As the ProtRS was developed in a primary prevention population without preexisting CKM disease at baseline, its performance and utility may not be applicable to a secondary prevention population with preexisting CKM disease; however, emerging data have demonstrated that proteomic predictors and pathways underlying primary and secondary cardiovascular events might be less dissimilar than previously held.25 Application of our ProtRS, which was derived using the full available follow-up (median 13.5 year), in models utilizing 10-year PREVENT-based CVD risk was performed to ensure consistency; however, this modest temporal mismatch could introduce discrepancies. The incompleteness of comprehensive primary care within the UK Biobank may have led to underdiagnosis of conditions such as type 2 diabetes, which could impact the identification of MD at baseline and potentially represent a source of confounding. Despite our efforts to subdivide the sample into randomly assigned cohorts for ProtRS development and testing, both cohorts were drawn from the same overall population within the same registry, which could result in overestimation of the ability of the ProtRS to predict risk and may limit overall generalizability. Although nearly 3000 proteins were utilized for ProtRS training, we were ultimately limited by the proteins available in the Olink Explore 3072 platform. As such, proteins not measured by this platform were unable to be accounted for. And as with all proteomics-based methods, our findings are limited by the confines of proteomics tools themselves, which include inconsistent feature annotation and variability in the quantification of protein levels between proteomic platforms, as well as current platforms lacking an absolute measurement of circulating protein abundance.
CONCLUSIONS
Our findings demonstrate that a ProtRS significantly improved risk prediction for incident CKM disease beyond traditional risk factors and improved prediction of PREVENT-based 10-year CVD risk beyond participants’ PREVENT scores. Further research is needed to determine the feasibility of implementing proteomics-based risk stratification in racially and ethnically diverse primary prevention clinical settings and to more deeply explore mechanistic insights of the individual proteins identified in our ProtRS. Additional studies are warranted to explore whether this score could be adapted or recalibrated for use in populations with established CKM diseases.
Supplementary Material
Supplemental Material is available at https://www.ahajournals.org/doi/suppl/10.1161/CIRCGEN.124.005125.
Acknowledgments
This research has been conducted using the UK Biobank Resource under Application Number 34031. The authors like to express their deep appreciation to all UK Biobank study participants.
Sources of Funding
This work was supported by the National Heart, Lung and Blood Institute of the National Institutes of Health (Bethesda, MD) under award numbers 5T32HL007745-30 to Dr Yadalam and T32 HL130025-06 to Dr Razavi.
Nonstandard Abbreviations and Acronyms
- CKM
cardio-kidney-metabolic
- CVD
cardiovascular disease
- ICD
International Statistical Classification of Diseases
- KD
kidney disease
- LASSO
least absolute shrinkage and selection operator
- MD
metabolic disease
- sHR
subdistribution hazard ratio
- PREVENT
Predicting Risk of CVD EVENTS
- ProtRS
proteomics-based risk scores
Footnotes
Disclosures
None.
The abstract for this project was presented at the American College of Cardiology Scientific Sessions 2025 on March 31, 2025 in New Orleans, LA.
Contributor Information
Adithya K. Yadalam, Division of Cardiology, Emory University School of Medicine, Atlanta, GA; Emory Clinical Cardiovascular Research Institute, Division of Cardiology, Emory University School of Medicine, Atlanta, GA.
Chang Liu, Department of Epidemiology, Emory University Rollins School of Public Health, Atlanta, GA.
Qin Hui, Department of Epidemiology, Emory University Rollins School of Public Health, Atlanta, GA.
Alexander C. Razavi, Division of Cardiology, Emory University School of Medicine, Atlanta, GA; Emory Clinical Cardiovascular Research Institute, Division of Cardiology, Emory University School of Medicine, Atlanta, GA.
Laurence S. Sperling, Division of Cardiology, Emory University School of Medicine, Atlanta, GA; Emory Clinical Cardiovascular Research Institute, Division of Cardiology, Emory University School of Medicine, Atlanta, GA.
Arshed A. Quyyumi, Division of Cardiology, Emory University School of Medicine, Atlanta, GA; Emory Clinical Cardiovascular Research Institute, Division of Cardiology, Emory University School of Medicine, Atlanta, GA.
Yan V. Sun, Division of Cardiology, Emory University School of Medicine, Atlanta, GA; Department of Epidemiology, Emory University Rollins School of Public Health, Atlanta, GA; Atlanta Veterans Affairs Healthcare System, Decatur, GA.
REFERENCES
- 1.Ostrominski JW, Arnold SV, Butler J, Fonarow GC, Hirsch JS, Palli SR, Donato BMK, Parrinello CM, O’Connell T, Collins EB, et al. Prevalence and overlap of cardiac, renal, and metabolic conditions in US adults, 1999–2020. JAMA Cardiol. 2023;8:1050–1060. doi: 10.1001/jamacardio.2023.3241 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Aggarwal R, Ostrominski JW, Vaduganathan M. Prevalence of cardiovascular-kidney-metabolic syndrome stages in US adults, 2011–2020. JAMA. 2024;331:1858–1860. doi: 10.1001/jama.2024.6892 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Xu J, Murphy SL, Kochanek KD, Arias E. Mortality in the United States, 2021. NCHS Data Brief. 2022;1–8. [PubMed] [Google Scholar]
- 4.Minhas AMK, Mathew RO, Sperling LS, Nambi V, Virani SS, Navaneethan SD, Shapiro MD, Abramov D. Prevalence of the cardiovascular-kidney-metabolic syndrome in the United States. J Am Coll Cardiol. 2024;83:1824–1826. doi: 10.1016/j.jacc.2024.03.368 [DOI] [PubMed] [Google Scholar]
- 5.Ndumele CE, Rangaswami J, Chow SL, Neeland IJ, Tuttle KR, Khan SS, Coresh J, Mathew RO, Baker-Smith CM, Carnethon MR, et al. ; American Heart Association. Cardiovascular-kidney-metabolic health: a presidential advisory from the American Heart Association. Circulation. 2023;148:1606–1635. doi: 10.1161/CIR.0000000000001184 [DOI] [PubMed] [Google Scholar]
- 6.Sun BB, Chiou J, Traylor M, Benner C, Hsu Y-H, Richardson TG, Surendran P, Mahajan A, Robins C, Vasquez-Grinnell SG, et al. ; Alnylam Human Genetics. Plasma proteomic associations with genetics and health in the UK Biobank. Nature. 2023;622:329–338. doi: 10.1038/s41586-023-06592-6 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Ferkingstad E, Sulem P, Atlason BA, Sveinbjornsson G, Magnusson MI, Styrmisdottir EL, Gunnarsdottir K, Helgason A, Oddsson A, Halldorsson BV, et al. Large-scale integration of the plasma proteome with genetics and disease. Nat Genet. 2021;53:1712–1721. doi: 10.1038/s41588-021-00978-w [DOI] [PubMed] [Google Scholar]
- 8.Helgason H, Eiriksdottir T, Ulfarsson MO, Choudhary A, Lund SH, Ivarsdottir EV, Hjorleifsson Eldjarn G, Einarsson G, Ferkingstad E, Moore KHS, et al. Evaluation of large-scale proteomics for prediction of cardiovascular events. JAMA. 2023;330:725–735. doi: 10.1001/jama.2023.13258 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Royer P, Björnson E, Adiels M, Josefson R, Hagberg E, Gummesson A, Bergström G. Large-scale plasma proteomics in the UK Biobank modestly improves prediction of major cardiovascular events in a population without previous cardiovascular disease. Eur J Prev Cardiol. 2024;31:1681–1689. doi: 10.1093/eurjpc/zwae124 [DOI] [PubMed] [Google Scholar]
- 10.Dubin RF, Deo R, Ren Y, Wang J, Zheng Z, Shou H, Go AS, Parsa A, Lash JP, Rahman M, et al. ; CRIC Study Investigators. Proteomics of CKD progression in the chronic renal insufficiency cohort. Nat Commun. 2023;14:6340. doi: 10.1038/s41467-023-41642-7 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Ye Z, Zhang Y, Zhang Y, Yang S, He P, Liu M, Zhou C, Gan X, Huang Y, Xiang H, et al. Large-scale proteomics improve prediction of chronic kidney disease in people with diabetes. Diabetes Care. 2024;47:1757–1763. doi: 10.2337/dc24-0290 [DOI] [PubMed] [Google Scholar]
- 12.Cronjé HT, Mi MY, Austin TR, Biggs ML, Siscovick DS, Lemaitre RN, Psaty BM, Tracy RP, Djoussé L, Kizer JR, et al. Plasma proteomic risk markers of incident type 2 diabetes reflect physiologically distinct components of glucose-insulin homeostasis. Diabetes. 2023;72:666–673. doi: 10.2337/db22-0628 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Sudlow C, Gallacher J, Allen N, Beral V, Burton P, Danesh J, Downey P, Elliott P, Green J, Landray M, et al. UK biobank: an open access resource for identifying the causes of a wide range of complex diseases of middle and old age. PLoS Med. 2015;12:e1001779. doi: 10.1371/journal.pmed.1001779 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Huang Y, Hui Q, Gwinn M, Hu Y-J, Quyyumi AA, Vaccarino V, Sun YV. Sexual Differences in genetic predisposition of coronary artery disease. Circ Genom Precis Med. 2021;14:e003147. doi: 10.1161/CIRCGEN.120.003147 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Inker Lesley A, Eneanya Nwamaka D, Coresh J, Tighiouart H, Wang D, Sang Y, Crews DC, Doria A, Estrella MM, Froissart M, et al. New creatinine- and cystatin C–based equations to estimate GFR without race. N Engl J Med. 2021;385:1737–1749. doi: 10.1056/NEJMoa2102953 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Wik L, Nordberg N, Broberg J, Björkesten J, Assarsson E, Henriksson S, Grundberg I, Pettersson E, Westerberg C, Liljeroth E, et al. Proximity extension assay in combination with next-generation sequencing for high-throughput proteome-wide analysis. Mol Cell Proteomics. 2021;20:100168. doi: 10.1016/j.mcpro.2021.100168 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Elliott P, Peakman TC, Biobank UK. The UK Biobank sample handling and storage protocol for the collection, processing and archiving of human blood and urine. Int J Epidemiol. 2008;37:234–244. doi: 10.1093/ije/dym276 [DOI] [PubMed] [Google Scholar]
- 18.Khan SS, Matsushita K, Sang Y, Ballew SH, Grams ME, Surapaneni A, Blaha MJ, Carson AP, Chang AR, Ciemins E, et al. ; Chronic Kidney Disease Prognosis Consortium and the American Heart Association Cardiovascular-Kidney-Metabolic Science Advisory Group. Development and validation of the American Heart Association’s PREVENT equations. Circulation. 2024;149:430–449. doi: 10.1161/CIRCULATIONAHA.123.067626 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Tibshirani R The lasso method for variable selection in the cox model. Stat Med. 1997;16:385–395. doi: 10.1002/(sici)1097-0258(19970228)16:4<385::aid-sim380>3.0.co;2-3 [DOI] [PubMed] [Google Scholar]
- 20.Alexa A, Rahnenfuhrer J. topGO: enrichment analysis for gene ontology. R package version 2.58.0 [Google Scholar]
- 21.Fine JP, Gray RJ. A proportional hazards model for the subdistribution of a competing risk. J Am Stat Assoc. 1999;94:496–509. doi: 10.2307/2670170 [DOI] [Google Scholar]
- 22.Uno H, Cai T, Pencina MJ, D’Agostino RB, Wei LJ. On the C-statistics for evaluating overall adequacy of risk prediction procedures with censored survival data. Stat Med. 2011;30:1105–1117. doi: 10.1002/sim.4154 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Ho FK, Mark PB, Lees JS, Pell JP, Strawbridge RJ, Kimenai DM, Mills NL, Woodward M, McMurray JJV, Sattar N, et al. A proteomics-based approach for prediction of different cardiovascular diseases and dementia. Circulation. 2025;151:277. doi: 10.1161/CIRCULATIONAHA.124.070454 [DOI] [PubMed] [Google Scholar]
- 24.Gao C, Gao S, Zhao R, Shen P, Zhu X, Yang Y, Duan C, Wang Y, Ni H, Zhou L, et al. Association between systemic immune-inflammation index and cardiovascular-kidney-metabolic syndrome. Sci Rep. 2024;14:19151. doi: 10.1038/s41598-024-69819-0 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Deo R, Dubin RF, Ren Y, Wang J, Feldman H, Shou H, Coresh J, Grams ME, Surapaneni AL, Cohen JB, et al. Proteomic assessment of the risk of secondary cardiovascular events among individuals with CKD. J Am Soc Nephrol. 2025;36:231–241. doi: 10.1681/ASN.0000000502 [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.