Proteomic cardiovascular risk assessment in chronic kidney disease

Abstract

Aims: Chronic kidney disease (CKD) is widely prevalent and independently increases cardiovascular risk. Cardiovascular risk prediction tools derived in the general population perform poorly in CKD. Through large-scale proteomics discovery, this study aimed to create more accurate cardiovascular risk models.

Methods and results: Elastic net regression was used to derive a proteomic risk model for incident cardiovascular risk in 2182 participants from the Chronic Renal Insufficiency Cohort. The model was then validated in 485 participants from the Atherosclerosis Risk in Communities cohort. All participants had CKD and no history of cardiovascular disease at study baseline when ∼5000 proteins were measured. The proteomic risk model, which consisted of 32 proteins, was superior to both the 2013 ACC/AHA Pooled Cohort Equation and a modified Pooled Cohort Equation that included estimated glomerular filtrate rate. The Chronic Renal Insufficiency Cohort internal validation set demonstrated annualized receiver operating characteristic area under the curve values from 1 to 10 years ranging between 0.84 and 0.89 for the protein and 0.70 and 0.73 for the clinical models. Similar findings were observed in the Atherosclerosis Risk in Communities validation cohort. For nearly half of the individual proteins independently associated with cardiovascular risk, Mendelian randomization suggested a causal link to cardiovascular events or risk factors. Pathway analyses revealed enrichment of proteins involved in immunologic function, vascular and neuronal development, and hepatic fibrosis.

Conclusion: In two sizeable populations with CKD, a proteomic risk model for incident cardiovascular disease surpassed clinical risk models recommended in clinical practice, even after including estimated glomerular filtration rate. New biological insights may prioritize the development of therapeutic strategies for cardiovascular risk reduction in the CKD population.

Keywords: Cardiovascular risk; Kidney disease; Mendelian Randomization; Pathway analysis; Prediction; Proteomics.

Structured Graphical Abstract

ARIC, Atherosclerosis Risk in Communities; CRIC, Chronic Renal Insufficiency Cohort; CKD, chronic kidney disease; CV, cardiovascular; CVE, cardiovascular event; ROC AUC, receiver operating characteristic area under the curve.

Figure 1

Methods overview.

Figure 2

Time-dependent area under the curve values across 10 years of follow-up. The area under the curve values are depicted for each model in the Chronic Renal Insufficiency Cohort testing set (internal validation) and the Atherosclerosis Risk in Communities cohort (external validation). Area under the curve values for the Pooled Cohort Equation were the weighted sum of four different equations representing Black women, White women, Black men, and White men. The modified Pooled Cohort Equation included estimated glomerular filtration rate and the Pooled Cohort Equation risk factors (age, sex, race, systolic blood pressure, use of anti-hypertensive therapies, total cholesterol, high-density lipoprotein cholesterol, history of diabetes, and current smoking), all of which were refit to the Chronic Renal Insufficiency Cohort training set. The protein model was derived in the Chronic Renal Insufficiency Cohort training set using elastic net regression and consisted of 32 proteins.

Figure 3

Calibration of proteomic risk model in the Chronic Renal Insufficiency Cohort and Atherosclerosis Risk in Communities cohort. (A) The Chronic Renal Insufficiency Cohort training set was comprised of 1742 participants without a self-reported history of coronary heart disease, myocardial infarction, stroke, or heart failure. Over the 10-year follow-up period, there were 373 incident cardiovascular events. (B) Similarly, the Atherosclerosis Risk in Communities study population was comprised of 485 participants with chronic kidney disease and without a history of cardiovascular disease at the time of proteomic measurements. Over a 10-year follow-up, there were 173 incident cardiovascular events.

Figure 4

Time-dependent area under the curve values for hybrid models across 10 years of follow-up. The area under the curve values are depicted for each hybrid model in the Chronic Renal Insufficiency Cohort testing set (internal validation). Both hybrid models were developed in the Chronic Renal Insufficiency Cohort training set. In the Clinical and Protein Hybrid Model, estimated glomerular filtration rate and all variables from the Pooled Cohort Equation were forced into a final hybrid model that selected protein variables by elastic net regression. The final model consisted of 15 clinical terms and 11 proteins. In the Competing Clinical and Protein Hybrid Model, all clinical terms and proteins were allowed to compete for inclusion in a hybrid model derived by elastic net regression. No clinical variables were selected, and the final model consisted of 27 proteins (all of which are also contained in the 32-protein model shown in Figure 2).

Figure 5

Volcano plots: proteomics of incident cardiovascular disease. The volcano plots depict the hazard ratios per log 2–transformed protein value. Analyses for associations between proteins and incident cardiovascular disease were evaluated in the full Chronic Renal Insufficiency Cohort without a baseline history of cardiovascular disease. There were a total of 2182 participants who contributed 459 incident cardiovascular events over the 10-year timeframe. (A) represents the unadjusted associations between individual proteins and incident cardiovascular disease in univariate analysis, (B) represents adjustment for estimated glomerular filtration rate; and (C) represents adjustment for age, sex, race, body mass index, systolic blood pressure, diastolic blood pressure, hypertension, diabetes, smoking, total cholesterol, HDL cholesterol, estimated glomerular filtration rate, and proteinuria.

Figure 5

Volcano plots: proteomics of incident cardiovascular disease. The volcano plots depict the hazard ratios per log 2–transformed protein value. Analyses for associations between proteins and incident cardiovascular disease were evaluated in the full Chronic Renal Insufficiency Cohort without a baseline history of cardiovascular disease. There were a total of 2182 participants who contributed 459 incident cardiovascular events over the 10-year timeframe. (A) represents the unadjusted associations between individual proteins and incident cardiovascular disease in univariate analysis, (B) represents adjustment for estimated glomerular filtration rate; and (C) represents adjustment for age, sex, race, body mass index, systolic blood pressure, diastolic blood pressure, hypertension, diabetes, smoking, total cholesterol, HDL cholesterol, estimated glomerular filtration rate, and proteinuria.

Figure 5

Volcano plots: proteomics of incident cardiovascular disease. The volcano plots depict the hazard ratios per log 2–transformed protein value. Analyses for associations between proteins and incident cardiovascular disease were evaluated in the full Chronic Renal Insufficiency Cohort without a baseline history of cardiovascular disease. There were a total of 2182 participants who contributed 459 incident cardiovascular events over the 10-year timeframe. (A) represents the unadjusted associations between individual proteins and incident cardiovascular disease in univariate analysis, (B) represents adjustment for estimated glomerular filtration rate; and (C) represents adjustment for age, sex, race, body mass index, systolic blood pressure, diastolic blood pressure, hypertension, diabetes, smoking, total cholesterol, HDL cholesterol, estimated glomerular filtration rate, and proteinuria.