A Polygenic Score for Reduced Kidney Function and Adverse Outcomes in a Chronic Kidney Disease Cohort

Graphical Abstract

Introduction

Chronic kidney disease (CKD) is a global health burden of increasing importance that affects >10% of the general adult population.¹ CKD is defined as the sustained presence of abnormalities of kidney structure or function, and classified using estimated glomerular filtration rate (eGFR) and the urinary albumin-to-creatinine ratio (UACR).^S1 The eGFR based on serum creatinine is the most common measure of kidney function. Persons with CKD are at increased risk of adverse outcomes such as kidney failure (KF), cardiovascular diseases, and premature death.^S2

Genome-wide association studies enable the calculation of polygenic scores (PGS). PGS aggregate the effects of many genetic variants into a single number and permit a straight-forward investigation of the association between a genetic predisposition with a given outcome. Yet, evaluations of such PGS in external studies, including different target populations, are often limited. Large CKD cohorts like the prospective German Chronic Kidney Disease (GCKD)² study represent a valuable opportunity to study whether a PGS for reduced eGFR (termed “eGFR PGS”), is associated with adverse outcomes in CKD, and whether the PGS adds information above and beyond established risk factors. Therefore, the aim of this study was to investigate if (i) a polygenic predisposition to lower eGFR is associated with KF, major adverse cardiovascular events (3P-MACE), and mortality in persons with CKD, and if (ii) the eGFR PGS carries predictive ability by itself and in addition to established risk factors.

Results

In the PGS development stage, the PGS with the best performance (Supplementary Methods S5 and Supplementary Table S1) assumed a fraction of causal variants of 0.1 and was subsequently applied in the GCKD study. The mean eGFR among 4 924 participants (39.8% women) with complete data was 49.4 mL/min/1.73m² (Supplementary Table S2). As expected in a CKD cohort that recruited mostly participants with stage G3 CKD, Spearman correlation of the eGFR PGS with eGFR at the enrolment visit was low (0.04, p=0.008). Over a median follow-up of 6.5 years (Q1: 6.5; Q3: 6.5), 470 (9.5%) participants experienced KF, 551 (11.2%) 3P-MACE, and 630 (12.8%) died.

A PGS for eGFR is associated with adverse outcomes

A higher continuous PGS, corresponding to a lifelong genetic predisposition to lower eGFR, was significantly associated with higher risk for all three outcomes (KF: HR 1.22, 95% confidence interval (CI) [1.11;1.34], 3P-MACE: 1.19 [1.09;1.29], death: 1.15 [1.07;1.25]) after controlling for age, sex, baseline eGFR and genetic principal components (model 2; Figure 1, panel A; Supplementary Tables S3–S5). Results with and without (model 1) adjustment for eGFR were similar. Upon adjustment for the UACR (model 3), the associations remained significant, with similar estimates for death and 3P-MACE, but attenuated ones for KF (KF: HR 1.12, 95% CI [1.02;1.23], 3P-MACE: 1.16 [1.07;1.26], death: 1.12 [1.04;1.21]). Additional adjustment for prevalent diabetes (model 4) had little effect (Supplementary Tables S3–S5). Subdistribution analyses (Supplementary Methods S7) delivered similar results, indicating the absence of indirect effects through the competing event. The risk of KF over 6.5 years was 1.42 (95% CI 1.16–1.72) times higher comparing those in the highest PGS quartile with the remainder of the cohort (Supplementary Table S3). Across deciles of the eGFR PGS, participants in the highest decile, corresponding to genetically lowest eGFR, had a more than 1.5 times increased risk of KF compared to those in the lower deciles (HR=1.52 [1.18;1.97]).

Figure 1: Association results of the eGFR PGS with all studied outcomes.

A: Hazard ratios of the continuous eGFR PGS and comparing the top quartile and top decile with the rest of the distribution for different adjustment models. The studied outcomes are color-coded. Adjustment: model 1: age, sex, PC1, PC2; model 2: model 1 + baseline eGFR; model 3: model 2 + log(UACR). B: Cumulative incidence function (unadjusted) for KF comparing the top decile and quartile of the eGFR PGS to the respective rest of the distribution. C: Cumulative incidence function (unadjusted) for 3P-MACE comparing the top decile and quartile of the eGFR PGS to the respective rest of the distribution. eGFR: estimated glomerular filtration rate; UACR: urinary albumin-to-creatinine ratio. PC: Principal Component.

The unadjusted cumulative incidence function for KF as well as for 3P-MACE was most pronounced for the highest decile of the eGFR PGS compared to the rest of the distribution, in line with the Cox regression results (Figure 1, panels B and C). Notably, the function for KF clearly distinguished participants in the highest and lowest decile and quartile of the PGS (cumulative incidence of KF of 14.9% (highest decile) vs. 9.6% (rest of the disribution)). The same trends were observed for the individual outcomes combined in 3P-MACE, acute myocardial infarction, cerebral haemorrhage and stroke (Supplementary Table S5), albeit mostly not significant.

We compared these results with two previously published eGFR PGS available from the PGS Catalog by Yu et al.³ and Khan et al.⁴ (see Supplementary Methods S6). As shown in Supplementary Table S6 and Supplementary Figure S1, the results for the main model (model 2) were similar both in magnitude and in statistical significance for each of the evaluated endpoints across the evaluated PGS, with the PGS by Khan et al. showing the highest hazard ratios in the most extreme evaluated tail (top decile vs. rest).

Predictive ability of the PGS

A likelihood ratio test (LRT) comparing the well-established kidney failure risk equation^S3 (KFRE) for incident KF with the same model after addition of the eGFR PGS was nominally significant, which also held true for the two previously published PGS (Supplementary Table S7), indicating potential for improvement of model performance. Yet, neither calibration nor the prediction error curves, the c-index or the receiver operating characteristic (ROC) curve at year 6 improved when adding the eGFR PGS to the well-established KFRE model (Figure 2). Addition of the eGFR PGS to the KFRE resulted in no change of the c-index of 0.84. Similarly, the integrated prediction error curve (IPEC), where higher values indicate worse prediction, changed from 56.37 to 56.68 for KF. The same trends in c-index and IPEC were observed for the two previously published PGS, supporting the initial observation.

Figure 2: Predictive ability of the eGFR PGS.

A: Calibration plot at year six; B: Prediction error curves for KF over time; C: Six-year ROC curve for KF; Each panel displays different adjustments: Null model: unadjusted; PGS only: Polygenic Score (PGS) only; Kidney Failure Risk Equation (KFRE): adjustment for age, sex, baseline eGFR, log(UACR); KFRE plus PGS: KFRE + PGS. eGFR: estimated glomerular filtration rate; UACR: urinary albumin-to-creatinine ratio.

Discussion

In this study of an eGFR PGS in persons with CKD, we found significant associations with KF, 3P-MACE and death, but limited predictive ability with respect to KF.

While various studies focused primarily on the association of an eGFR PGS with clinical outcomes or the prediction of the presence of CKD in the general population (e.g. Yun et al.^S4, Khan et al.⁴), little is known about the relationship of an eGFR PGS to adverse outcomes in persons with already diagnosed moderate CKD, a novel aspect addressed in this study. A previous study by Yu et al.³ reported a significant positive association of a similarly developed eGFR PGS with incident CKD and KF in the general population. Our results extend these findings by showing that an eGFR PGS developed in the general population is associated with adverse kidney outcomes also among persons with already existing CKD.

At the same time, our study showed no improvement in risk prediction, underlining the finding of Gleiss et al.⁵ that significant associations are only a necessary but not a sufficient condition for the usefulness of prognostic markers in prediction studies. The limited use of PGS for risk prediction has been pointed out by Wald et al. previously.⁶ While several studies concluded that eGFR PGS were suitable for prediction of the presence of CKD^3,7, they did not evaluate prediction of incident outcomes, although these are two important separate aspects in PGS validity according to Torkamani et al.⁸ A previous population-based prospective study of incident CKD³ on the other hand did not report on risk prediction. We therefore compared association and prediction results of our eGFR PGS with two previously published PGS and found similar results across these PGS, supporting our initial findings. Our results suggest that in order to achieve the goal of clinical translation of an eGFR PGS^S5, further validation in CKD cohorts is warranted in order to identify settings in which such a PGS can add above and beyond established risk prediction equations, which do not require genetic information.

Strengths of our study are its new focus on persons with CKD and a large sample size with systematic outcome ascertainment over 6.5 years of follow-up, and the comparison to previously published PGS with consistent findings. Despite the large cohort of participants with CKD, however, the power to study finer distinctions, such as the very top percentiles of the PGS or rare outcomes, was limited, as discussed previously.⁹ Another limitation is that findings in the GCKD study may not be generalizable to non-European ancestry populations.

In conclusion, our study revealed significant associations between a polygenic predisposition to lower eGFR and KF, 3P-MACE, and death among persons with moderate CKD, emphasising the importance of genetic background even after disease onset. However, the eGFR PGS carried no added predictive ability with regard to KF beyond the well-performing KFRE among patients with established CKD.

Data availability statement

The newly developed eGFR PGS will be made publicly available in the PGS Catalog^S6 under pgscatalog.org when published.