Scoping Review of Polygenic Risk Scores in Kidney-Related Traits

Abstract

Introduction

Large-scale genome-wide association studies (GWAS) have identified numerous loci associated with various diseases and traits, enabling the development of polygenic risk scores (PRS) and genetic risk scores (GRS) that capture cumulative effects of variants. Although these scores have been applied in nephrology, no review has systematically evaluated their implementation or performance. This review aimed to summarize the current landscape of PRS and GRS related to kidney traits and diseases.

Methods

In April 2025, we searched PubMed, Embase, Scopus, and Web of Science for peer-reviewed articles and preprints following the Preferred Reporting Items for Systematic Reviews and Meta-Analysis extension for Scoping Review guidelines. Studies were included if they developed or evaluated PRS and GRS for kidney-related traits. Data on study characteristics, score construction, and performance metrics were extracted.

Results

Of the 1947 records identified, 104 met the inclusion criteria, encompassing 129 unique scores or evaluations. The most frequently studied traits were chronic kidney disease (CKD, n = 43) and estimated glomerular filtration rate (eGFR, n = 18). Additional outcomes such as albuminuria, eGFR rate slope, end-stage kidney disease (ESKD), acute kidney injury (AKI), and specific subgroups of CKD, were evaluated in a few studies. The key challenges include a marked ancestry imbalance, with a predominance of European populations, modest incremental predictive performance for CKD, and frequent mismatches between derivation traits and evaluated outcomes.

Conclusion

Current evidence suggests that the clinical utility of PRS and GRS in nephrology depends on the context. Improved diversity, phenotype definition, and methodological standardization are essential for enhancing translational potential.

Graphical abstract

Susceptibility to CKD and kidney-related laboratory measures such as eGFR and proteinuria are influenced by both genetic and environmental factors.^1,2 Genetic contributions include rare variants with large effect sizes and numerous common variants with relatively small effects that act cumulatively. Detection of common variants requires large-scale GWAS, which have been enabled by recent technological advances and international collaboration. Consequently, numerous loci associated with kidney function and CKD have been identified.3, 4, 5

One approach to assess how these loci affect individuals is through PRS or GRS.6, 7, 8 Typically, a GRS includes only a few variants that reach genome-wide significance (P < 5.0 × 10⁻⁸), whereas a PRS aggregates the effects from a large number of variants (often thousands or more) across the genome to capture the polygenic architecture of complex traits. By combining multiple P-value thresholds and linkage disequilibrium clumping strategies, PRS can capture the cumulative effect of many variants, providing a more refined estimate of individual disease risk. In this review, the term “PRS” is used broadly to include GRS reported in the included studies.

In recent years, methodological advances have further refined the construction of PRS. Although traditional clumping and thresholding approaches^6,9 are simple and widely used, they may not fully capture the joint effects of the correlated variants. Penalized regression methods, such as lasso and elastic net,^10,11 enable joint modeling with shrinkage. More recently, Bayesian LD-aware approaches (e.g., LDpred¹² and PRS-CS¹³) explicitly account for linkage disequilibrium across the genome, often improving predictive performance. However, differences in methods and parameter choices can contribute to variability in PRS performance across studies.

However, the utility of a constructed score depends on the data from which it is derived and the population to which it is applied. Most GWAS have been conducted in populations of European ancestry, and caution is required when extrapolating PRS to other populations because predictive accuracy may decline and population stratification may influence performance.^14,15 In nephrology, PRS could be useful for predicting kidney-related laboratory measures, CKD onset, progression, and related complications; however, the underlying data and performance evaluations have not been comprehensively investigated, and clinical evidence remains limited.

CKD is a highly prevalent condition that significantly affects mortality and imposes a substantial burden on health care systems worldwide.^16,17 Despite this considerable global impact, nephrology has relatively fewer clinical trials and limited therapeutic options for CKD than other medical fields.¹⁸ Thus, accurate stratification of genetic disease risk from a precision medicine perspective may be valuable. In this context, evaluating the current development of PRS in nephrology and clarifying the remaining challenges are important. Although such efforts have been undertaken in the cardiovascular field,¹⁹ reports in the kidney field remain scarce.

This scoping review aimed to provide a comprehensive overview of PRS research in nephrology, including study populations, variant selection and score construction methods, clinical applications, and applicability across multiethnic populations. This study offers insights that can inform future study designs and clinical implementations of PRS for kidney disease and related phenotypes.

Methods

Literature Search

This systematic literature review was conducted and reported in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analysis extension for Scoping Review guidelines.²⁰ The completed Preferred Reporting Items for Systematic Reviews and Meta-Analysis checklist is presented in the Supplementary Material. The MEDLINE, Embase, Web of Science, and Scopus databases were systematically searched for peer-reviewed full papers, preprints, and other documents accessible via the Internet in April 2025. In Supplementary Material, we list the search terms used in each database. During the extraction process, if a study used a PRS that had been developed in a previous publication, but the original PRS development study was not initially included in the final list, it was subsequently added to ensure completeness. The review was registered with Open Science Forum Registries (https://doi.org/10.17605/OSF.IO/CSBRH).

Inclusion and Exclusion Criteria

The inclusion criteria were as follows: (i) literature written in any language from any country; (ii) studies in the context of CKD, AKI, ESKD, or relevant laboratory measures; and (iii) studies in the context of PRS or relevant scores calculated from genomic information.

The following studies were excluded: (i) protocols and proposals; (ii) systematic reviews, commentaries, and scoping reviews; and (iii) missing full text. Publications in languages other than English were included and artificial intelligence–assisted translation was used when required to evaluate their content. Artificial intelligence was used solely for translation purposes and not for data interpretation, extraction, or analysis.

Study Selection

Two investigators (YS and YH) independently screened the titles and abstracts of the identified studies using the Covidence systematic review software (Veritas Health Innovation, Melbourne, Australia). Subsequently, these investigators independently reviewed the full texts of the selected studies based on the inclusion and exclusion criteria. Disagreements regarding the classification of the references were resolved by a third reviewer (KI) by conducting an additional review and confirming the final classification.

Data Collection and Summary

Data regarding study details were extracted using the Covidence software (Veritas Health Innovation, Melbourne, Australia) and Excel (Microsoft Corp., Redmond WA). One reviewer (YS) extracted additional data from eligible studies, including the following: (i) details of the citation, (ii) details of the data underlying the construction of the PRS, (iii) details of the PRS calculation, (iv) details of the test sample population, (v) PRS performance, and (vi) main study conclusions. All extracted data were cross-checked for accuracy by a second reviewer (YH), and any discrepancies were resolved through discussion. A meta-analysis was deemed inappropriate because of heterogeneity in the study design, populations, and outcome measures for quantitative studies. Quantitative study results were narratively synthesized.

In this review, the term "sex" is used throughout to refer to the biological sex variable (typically self-reported or genotype-derived) as defined and adjusted for in the included primary studies, in accordance with the Sex and Gender Equity in Research (SAGER) guidelines. The term "gender" was not used, as none of the included studies addressed gender as a social or psychological construct.

Results

Literature Search and Study Selection

Initially, 1947 studies were considered relevant based on the inclusion criteria. However, 992 duplicate studies were excluded before screening. Furthermore, 785 studies were excluded after screening the titles and abstracts, and 12 studies were excluded after reading the articles in greater depth during the assessment of the extracted data. In addition, 6 studies were identified as citations in the studies that were processed for eligibility assessment. Finally, 104 studies were included in this analysis. A Preferred Reporting Items for Systematic Reviews and Meta-Analysis flowchart summarizing the article selection process is shown in Figure 1. A list of all 104 studies and their information is provided in Supplementary Table S1.

Figure 1.

Preferred Reporting Items for Systematic Reviews and Meta-Analysis 2020 flow diagram for scoping review. Study selection process following the Preferred Reporting Items for Systematic Reviews and Meta-Analysis guidelines showing the flow of information through the scoping review. Several studies were identified through database searches, screening of titles and abstracts, full-text assessment for eligibility, and inclusion in the final analysis. Reasons for exclusion from the full-text review were categorized and quantified.

Characteristics of Included Studies

A total of 129 PRS analyses were reported in 104 studies. Most studies were published after 2019 (n = 90), and the number of reports increased markedly during this period (before 2019: 1.6 studies/yr; 2019 onward: 12.9 studies/yr). The sample size of the derivation GWAS ranged from 517 to 1,159,871 individuals.

Most articles used derivation cohorts of European ancestry (n = 40/104; Figure 2, Supplementary Table S2), followed by cohorts that included all superpopulations (European, East Asian, African, South Asian, and Admixed American; n = 15/104) and East Asian populations (n = 11/104). A similar pattern was observed for the evaluation cohorts; European ancestry was the most common, followed by East Asian and all-superpopulation cohorts (49, 8, and 6 of 104 studies, respectively). However, relatively few studies have evaluated African, South Asian, or Admixed American populations. Overall, 35 studies used the same ancestry for both derivation and evaluation, whereas 69 studies used different ancestry groups at these stages.

Figure 2.

Ancestry of derivation and evaluation datasets in the reviewed studies. We classified the ancestry of the derivation and evaluation datasets used in the included studies according to 5 superpopulations: European, East Asian, South Asian, African, and Admixed-American. Studies that included ≥ 2 superpopulations but not all 5 were categorized separately, as were those that included all 5 superpopulations.

The outcomes used to evaluate the performance of PRS included kidney-related laboratory measures and clinical outcomes or diseases (Table 1). The most assessed kidney laboratory traits were the eGFR, urine albumin-to-creatinine ratio (UACR), and eGFR slope (n = 18, 11, and 10, respectively). Among the kidney-related clinical outcomes or diseases, CKD was the most frequently analyzed (n = 43), followed by IgA nephropathy, membranous nephropathy, and diabetic kidney disease (n = 9, 7, and 6, respectively). The traits used for the GWAS underlying the PRS construction did not always match the outcomes used for performance evaluation. For example, several studies have evaluated the performance of PRS derived from GWAS of eGFR using outcomes such as incident CKD or progression to ESKD.

Table 1.

Measures or outcomes used to evaluate the performance of polygenic risk scores

Measures or outcomes	N = 129 (%)
Kidney-related laboratory measures
eGFR (continuous)	18 (14.0)
Albuminuria (continuous or binary)	11 (8.5)
eGFR slope (continuous or binary)	10 (7.8)
Kidney-related conditions or diseases
Chronic kidney disease (binary)	43 (33.3)
IgA nephropathy (binary)	9 (7.0)
Membranous nephropathy (binary)	7 (5.4)
Diabetic kidney disease (binary)	6 (4.7)
End-stage kidney disease (binary)	7 (5.4)
Acute kidney injury (binary)	5 (3.9)
Others	13 (10.1)

eGFR, estimated glomerular filtration rate.

The number of variants included in the construction of PRS varies widely, ranging from <10 to several hundred thousands. Although some older studies calculated scores without applying variant weights, nearly all recent studies incorporated weighting into score computation.

APOL1 risk alleles have a strong impact on kidney function and CKD development, with substantial differences in allele frequencies across ancestries. Among the studies identified in this review that evaluated the PRS for kidney-related traits, 2 included APOL1 risk alleles within the score itself, whereas 3 accounted for APOL1 risk allele status as a covariate in the performance evaluation.

Across the 104 included studies, the methods used for PRS derivation were as follows: clumping and thresholding in 23 studies, Bayesian LD-aware methods in 12 studies, and penalized regression approaches in 4 studies (counts are not mutually exclusive, because some studies evaluated multiple methods). A substantial number of studies (n = 63) used approaches that did not fall within these categories, including those based on curated sets of previously reported variants. Explicit reporting of parameter tuning was observed in 18 studies. Grid search was the most cused approach for hyperparameter tuning, reported in 14 studies.

For performance evaluation, most studies have reported effect estimates (β coefficients, odds ratios, or hazard ratios) or discrimination metrics such as area under the receiver operating characteristic curve (AUC) or C statistics from models adjusted for clinical parameters. However, the units used to express odds ratios or hazard ratios, as well as the definitions of the comparison groups, varied substantially across studies. Only a limited number of studies (n = 5) evaluated clinically informative performance metrics beyond AUC, such as the net reclassification index and integrated discrimination improvement.

Kidney-Related PRS or GRS

This scoping review includes the PRS developed for various traits and kidney-related diseases. A detailed narrative synthesis of the findings for each trait or disease is provided in Supplementary Tables S3 to S12. Although identifying the single “best-performing” PRS from the existing literature is challenging, we summarize the representative PRS for major kidney-related traits (eGFR, albuminuria, eGFR decline, CKD, ESKD, and AKI) in Table 2. The selection was based on several considerations, including concordance between derivation and validation traits, inclusion of diverse ancestries, adequate sample size of the derivation dataset, and reported evidence of predictive utility.

Table 2.

Representative polygenic risk scores for major kidney-related traits (eGFR, albuminuria, eGFR decline, CKD, ESKD, and AKI) from previous studies

Study		Derivation			Validation			Performance
Trait	Study	Phenotype of base data	Superpopulation	No. of participants	Validated Phenotype	Superpopulation	No. of participants	PRS β or OR/HR (95% CI)
eGFR	Zhou et al.22	eGFRcre	(1) EUR, (2) AFR, EUR	(1) 567,460, (2) 280,722	eGFRcre	AMR	11,534	eGFR β (SE), [PRS-CS EU GWAS] -0.181 (0.074, P = 0.0151), [PRS-CS TE GWAS] -0.275 (0.083, P = 0.0009) among prevalent CKD at visit 1 EU: European, TE: trans-ethnic
Albuminuria	Rashkin 2021	Albuminuria	ALL	564,257	Albuminuria	AFR	288	Time to albuminuria adjusted HR (95% CI), [45-SNP PRS] 1.46 (1.09–2.04), [3-SNP GRS] 1.91 (1.40–2.62)
eGFR decline	Gorski et al.39	eGFRcre decline (Rapid3, CKDi25)	EUR (Most), AFR, AMR, SAS	337,109	(1) eGFRcre decline (Rapid3), CKDi25), (2) ESKD, (3) AKI	(1) EUR, (2) EUR, (3) EUR	(1) 12,262 for Rapid3, 22,367 for CKDi25, (2) 6828, (3) 16,492	CKDi25 OR (95% CI) for high-risk vs. low-risk group, 1.29 (1.06–1.57),
CKD	Zhang 2023	CKD	ALL	1,046,070	CKD	Not specified	89,296	CKD adjusted HR (95% CI) vs. low genetic risk, [intermediate risk] 1.263 (1.133–1.408), [high risk] 1.713 (1.546–1.899)
ESKD	Collins et al.40	(1) eGFRcre, (2) rapid decline in eGFRcre	(1) EUR, (2) ALL	(1) 567,460, (2) 195,145	Kidney failure development	AMR, EUR	10,586	Kidney failure development before the age of 60 yrs Adjusted OR (95% CI) for top 10% PRS vs. others, [PRS of decreased eGFR] 1.26 (1.08–1.46)
AKI	Gorski et al.5	eGFRcre/cys decline	ALL, mostly EUR	343,339	ESKD, AKI	EUR	22,220	AKI adjusted OR (95% CI) for highest vs. lowest 5% GRS groups, 1.524 (1.204–1.931)

AKI, acute kidney disease; CI, confidence interval; CKD, chronic kidney disease; eGFR, estimated glomerular filtration rate; eGFRcre, creatinine-based eGFR; eGFRcys, cystatin C–based eGFR; ESKD; end-stage kidney disease; GWAS, genome-wide association studies; HR, hazard ratio; OR, odds ratio; PRS, polygenic risk scores.

PRS for Kidney-Related Laboratory Measures

Among kidney-related laboratory measures, eGFR and UACR, both incorporated into the Kidney Disease: Improving Global Outcomes classification; and eGFR slope were frequently evaluated.

eGFR

Eighteen studies, published between 2010 and 2024, examined the PRS related to eGFR (Supplementary Table S3). Among these, 15 studies constructed their own scores, whereas 3 applied previously developed scores. Most studies (n = 13) used GWAS summary statistics for eGFR as the derivation dataset; however, 2 studies relied on summary statistics from other traits (CKD/cardiovascular disease or rapid eGFR decline). The eGFR was calculated using creatinine in 8 studies and creatinine and/or cystatin C in 2 studies; however, the basis was not specified in 3 studies.

Regarding the ancestry of the derivation datasets, the majority are based on European cohorts, although scores derived from European and African datasets (reported in 2023) and African and Admixed American datasets (reported in 2024) have also been published. Compared with studies on other outcomes, those that focused on eGFR generally included larger sample sizes. Among the 15 studies that constructed new scores, 10 involved >100,000 individuals, and 1 exceeded 1 million participants. European cohorts were most frequently used, although 4 studies included additional ancestry in their evaluation datasets.

In total, 4 studies used PRS for stratification analyses, whereas the remaining 14 studies evaluated score performance. Of the 14 studies, 12 reported that the scores were informative, for example, showing significant associations with baseline eGFR; whereas no significant association was observed in 1 study, which was a 2021 study evaluating a score derived from 366 trans-ancestry eGFR lead variants in an African population.²¹ Nonetheless, 1 report has demonstrated significant associations in non-European populations, which constructed scores using European and African derivation datasets and validated them in an Admixed American cohort.²² One study evaluated both eGFR derived from creatinine and 1 derived from cystatin C as outcomes and reported similar results for both measures.²³

UACR

Eleven studies published between 2017 and 2025 evaluated the traits related to UACR (Supplementary Table S4). Nine studies used GWAS of UACR or related albuminuria measures (including the presence or absence of albuminuria) as the derivation dataset, whereas 3 earlier reports used the following: (i) GWAS of eGFR and (ii) variants associated with CKD or cardiovascular disease when constructing the scores. Most derivation datasets (n = 7) were based on individuals of European ancestry, whereas 2 studies used multiancestry (“all superpopulation”) datasets and 1 study used a South Asian cohort. In several studies, evaluation cohorts included East Asian, African, and Admixed American populations.

As for model performance, 5 have clearly demonstrated that PRS were significantly associated with UACR or related measures, such as the presence of albuminuria. One study has reported partially supportive findings, whereas the other 4 concluded that the scores were not useful. Three studies reporting no utility used traits of CKD and cardiovascular disease, or eGFR as its derivation dataset.

eGFR Decline

Ten studies published between 2014 and 2025 were identified to evaluate outcomes related to eGFR slope or eGFR decline (including closely related outcomes such as time to kidney failure; Supplementary Table S5). Three of these studies exclusively validated the previously developed scores for CKD, eGFR, and eGFR decline. Among the remaining 7 studies that developed new scores, the earliest 3 used GWAS of CKD or cross-sectional eGFR as derivation datasets; whereas all later studies used GWAS of eGFR slope or decline. The earliest reports were limited to cohorts exclusively of European ancestry, whereas more recent studies have included Admixed American and African populations.

Except for 1 study that used PRS solely for stratification, 9 studies evaluated performance. Of these, 7 studies have reported statistically significant associations between the score and ≥1 relevant outcome. In a 2022 study by Koraishy et al.,²⁴ a PRS comprising > 400,000 variants was constructed using a trans-ancestry derivation dataset of 195,145 individuals with rapid eGFR decline and was validated in 1601 Europeans. The PRS was significantly associated with the binary outcome of rapid decliners (odds ratio: 1.14, 95% confidence interval: 1.01–1.28), whereas the association with the continuous eGFR slope was not significant (β −0.02, 95% confidence interval: −0.06 to 0.02; P = 0.317). Across other studies, significant associations were more frequently observed when using eGFR decline as a binary end point. No study has reported a significant association between the eGFR slope, treated as a quantitative trait, and the score.

PRS for Kidney-Related Conditions

The key outcomes in patients with kidney disease for which the PRS performance has been evaluated include CKD, ESKD, and AKI. Fewer than 10 studies investigated each CKD subgroup, including IgA nephropathy, membranous nephropathy, and diabetic kidney disease (n = 9, 7, and 6, respectively).

CKD

Forty-three studies have reported that the PRS was related to CKD (Supplementary Table S6). Among these, 6 studies applied previously published scores and 1 constructed score by examining known associated variants. The remaining 36 studies developed and reported the PRS generated within their respective analyses, based on summary statistics.

Of these 36 studies, 18 were constructed based on summary statistics for eGFR, whereas 17 were based on summary statistics for CKD or kidney dysfunction, and 1 was based on summary statistics for rapid eGFR decline. The use of CKD-specific summary statistics has recently increased. One study further stratified CKD into clusters and constructed cluster-specific scores,²⁵ whereas another integrated a PRS derived from 35 biomarkers to develop composite scores for CKD and related diseases.²⁶

Regarding the ancestry of the derivation datasets, 18 studies exclusively included European participants, whereas 7 studies incorporated all 5 superpopulations; the remaining reports involved multi-ancestry cohorts. For the evaluation datasets, 18 studies used European-only samples and 3 included all major superpopulations. Thirteen studies employed European-only cohorts for both derivation and evaluation.

Among the 43 studies, 35 assessed the predictive performance. Of these, 34 reported significant associations between the PRS and incident CKD, indicating clinically informative possibility of the scores (Figure 3a and b). In most cases, PRS was significantly associated with CKD onset after adjusting for age, sex, and additional clinical covariates, yielding higher odds ratios in the multivariable models. The AUCs of PRS varied substantially across studies, ranging from approximately 0.5–0.6 to >0.7. Scores with lower AUCs were typically constructed without incorporating clinical information such as age or sex. The predictive performance did not consistently improve in the more recently reported PRS (Figure 3c).

Figure 3.

Performance evaluation of polygenic risk scores (PRS) for chronic kidney disease outcomes. (a) ORs and (b) HRs, including only those calculated using comparable polygenic risk score units and reported by ≥2 independent studies. Effect estimates are presented according to the unit of polygenic risk scores used in each study (e.g., per SD increase, quartile comparisons, or tertile comparisons), and may not be directly comparable across studies. (c) AUC. Underlined and nonunderlined values indicate adjusted and unadjusted (crude) AUCs, respectively. Adjusted models include covariates such as age, sex, and clinical risk factors as defined in each study. CIs are shown where available; point estimates without error bars indicate that CIs were not reported. AUC, area under the receiver operating characteristic curve; CI, confidence interval; HR, hazard ratio; OR, odds ratio; Q1/Q4, quartile 1/4; T1/2/3, tertile 1/2/3.

Nevertheless, relatively few studies have demonstrated improved discrimination when a genetic score is added to models containing clinical risk factors alone (Figure 4a). Only 4 studies27, 28, 29, 30 have reported increases in the AUC or C-statistics with PRS inclusion, whereas 8 studies have indicated no enhancement in such performance. Additional reports have noted that the AUC values for CKD were lower than those for other complex diseases³¹ and that the PRS for CKD did not correlate with disease prevalence.³²

Figure 4.

Discrimination performance of models using clinical variables alone versus models incorporating polygenic risk scores. (a) Overall chronic kidney disease as the target outcome. (b) Chronic kidney disease subgroups as the target outcome. Circles denote the AUC, and triangles denote C-statistics. AUC and C-statistic were treated equivalently as measures of discrimination. Results are shown for models that include only clinical variables alone and for those that incorporate both clinical variables and polygenic risk scores. Colors indicate model type where applicable. CIs are shown where available; point estimates without error bars indicate that CIs were not reported. AUC, area under the receiver operating characteristic curve; CI, confidence intervals; DKD, diabetic kidney disease; IgAN, IgA nephropathy; MN, membranous nephropathy.

In contrast to analyses in which CKD as a whole was treated as the target disease, studies focusing on specific CKD subgroups such as membranous nephropathy, IgA nephropathy, and diabetic kidney disease generally showed higher AUC values (Figure 4b). Moreover, across these subgroups, a consistent trend was observed in which models incorporating PRS showed better predictive performance than those based on clinical factors alone.33, 34, 35, 36, 37 Detailed results for these subgroups are provided in the Supplementary Material and Supplementary Tables S7 to S9.

ESKD

Seven studies evaluated the association of PRS with ESKD outcomes (Supplementary Table S10). Among these studies, none used a GWAS of ESKD as the derivation dataset. Five studies relied on the GWASs of eGFR decline or slope, and 3 used the GWASs of eGFR levels, with some overlap between the categories.

Regarding the ancestry composition of the derivation datasets, 3 studies used cohorts composed exclusively or predominantly of Europeans, 2 included all major superpopulations, and 1 used an East Asian dataset. Four studies were published in 2019, 2021 (n = 2), and 2025 and incorporated the CKDGen consortium as a primary derivation cohort.^27,38, 39, 40 For evaluation purposes, most studies used European cohorts. One study employed an East Asian derivation dataset; however, no independent evaluation dataset was used.⁴¹

Of the 7 studies, 4 reported significant associations between PRS and outcome, all of which were published in 2022 or later and used derivation GWASs focused on eGFR decline or eGFR slope. None of the studies evaluated AUC for ESKD as an outcome.

AKI

Five studies have investigated PRS in relation to AKI as an outcome (Supplementary Table S11). Two studies examined postoperative AKI (following cardiac surgery or surgery in general), whereas the remaining 3 evaluated AKI that was not restricted to the postoperative setting. Regarding the derivation datasets, 2 studies used GWASs of eGFR decline and 2 used GWASs of eGFR with sample sizes > 300,000 and > 1,000,000, respectively. Only 1 recent study published in 2023 used a GWAS of AKI as the derivation dataset; however, the sample size was limited to 478 individuals, including 121 cases of AKI.⁴²

The ancestry of the derivation cohorts was either European or included all major superpopulations, whereas the evaluation cohorts were exclusively European (1 study from the United States did not report ancestry). Among the 5 studies, only 2 have reported significant associations between PRS and AKI.^5,34 Neither study focusing on postoperative AKI has demonstrated a significant association.

Others

Studies not belonging to the aforementioned categories were grouped as “others” (Supplementary Table S12). This category included studies related to all-cause mortality, peritoneal dialysis, kidney transplantation, nephrolithiasis, lupus nephritis, congenital anomalies of the kidney and urinary tract, nephrotic syndrome, and preeclampsia.

Discussion

In this study, we systematically identified and summarized all prepublished and published PRS studies in the field of nephrology, detailing the underlying derivation datasets, score construction methods, evaluation cohorts, and reported performance metrics. We further categorized the literature into kidney-related laboratory measures, kidney-related outcomes, kidney diseases, and other conditions; and synthesized findings for key outcomes such as eGFR, CKD, and ESKD within each category. Our synthesis highlights several important gaps and challenges that warrant further research.

First, we assessed the range of phenotypes for which PRS analyses were performed in the field of nephrology. Within nephrology, PRS research has most frequently focused on CKD and eGFR (n = 43 and 18, respectively). By contrast, relatively few studies have examined albuminuria (n = 11) despite its importance in CKD staging. Fewer reports have addressed ESKD (n = 7), the most clinically relevant outcome in nephrology, or the eGFR slope (n = 10), which is widely used as a surrogate end point. A notable point is the paucity of GWAS focusing on advanced CKD and ESKD. Without such studies, developing well-calibrated PRS for these outcomes remains challenging. Additional studies targeting these outcomes are warranted to characterize the utility of PRS in kidney disease more comprehensively.

Second, derivation GWAS traits do not always match the outcomes used for PRS evaluation. For ESKD, none of the published PRS were derived from derivation datasets in which ESKD itself was used as a trait. In practice, PRS are often derived from eGFR-based summary statistics even when predicting related but distinct outcomes, potentially leading to discrepancies between the derivation and target phenotypes. Importantly, such mismatches may lead to underestimation of the true predictive performance of PRS. A greater alignment between derivation traits and target phenotypes may improve model performance and enhance biological interpretability.

Third, a major issue identified in the current literature is the substantial imbalance in ancestral representations. Such ancestry imbalances have already been highlighted in scoping reviews of PRS in other disease areas,¹⁹ and the same pattern is evident in nephrology. Most PRS studies have been conducted in populations of European descent, with a smaller but limited number in East Asian populations, whereas research involving African, Hispanic/Latin, or other ancestries remains largely understudied. In addition, PRS developed in one population are often applied to different populations or specific clinical subgroups, which may introduce discrepancies between the derived and target populations. This mismatch can adversely affect predictive performance because of differences in linkage disequilibrium structure, allele frequencies, and environmental or clinical contexts. Several studies have indicated that scores demonstrating good performance in one ancestry are substantially worse than those in others.^34,43,44 These limitations constrain the transferability and clinical validity of polygenic scores across populations and highlight the urgent need for broader ancestral representation. Several strategies, including increasing GWAS sample sizes in underrepresented populations, developing ancestry-specific and cross-ancestry PRS methods,⁴⁵ and prioritizing external validation across diverse populations, may help address this gap.

Fourth, we observed substantial heterogeneity in the PRS construction methods, parameter tuning strategies, and performance evaluations across the studies. Although more advanced methods, such as Bayesian LD-aware approaches and penalized regression frameworks, have been proposed, their application in the current literature remains relatively limited, indicating considerable room for further methodological development. Moreover, parameter-tuning strategies, including grid search or parameter sweep approaches, as well as thresholds for variant inclusion, vary widely across studies. Similarly, performance evaluation metrics and validation frameworks are not standardized, further complicating direct comparisons between studies. This methodological heterogeneity may affect both the reproducibility and comparability of the findings and highlights the need for greater standardization, transparent reporting, and systematic benchmarking of PRS approaches in future research.

In addition, incorporating variants with large effect sizes, such as APOL1 G1 and G2 risk alleles, may further improve the predictive performance in specific populations. However, standardized approaches for directly integrating these high-impact variants into conventional PRS remain underdeveloped. Although some of these variants are relatively common within particular ancestral populations, their biological effects and genetic architectures differ substantially from those assumed in standard polygenic models. Therefore, rather than being directly incorporated into PRS, they may be more appropriately modeled as a distinct layer of genetic risk and combined with polygenic scores within integrated frameworks.^46,47 Further methodological development is required to establish optimal strategies for incorporating these variants into risk-prediction models.

Beyond methodological considerations, a rigorous evaluation of PRS performance requires multivariable modeling that incorporates established clinical predictors. Although some studies have reported relatively low AUC values for CKD (approximately 0.5–0.6), these estimates were largely derived from models that were not adjusted for clinical covariates. Adjustment for conventional risk factors may lead to modestly improved discrimination.

However, statistically significant associations between PRS and kidney-related traits or outcomes after covariate adjustment do not necessarily translate into significant improvements in discriminative performance. When CKD was considered a single heterogeneous disease entity, the incremental contribution of PRS beyond the clinical risk factors may be limited. By contrast, analyses focusing on more specific underlying diseases of CKD have demonstrated a consistent trend toward improved discrimination when PRS was added to clinical models compared with models that included clinical factors alone. Thus, restricting analyses to more specific disease subtypes may enhance the predictive utility of the PRS. Commonly proposed strategies to improve the performance of PRS include increasing the sample size of derivation datasets, integrating multi-ancestry data, refining PRS construction algorithms, and leveraging information from genetically correlated traits.48, 49, 50 These approaches may be beneficial for the development of PRS in kidney disease.

Moreover, although AUC is the most reported metric for evaluating PRS performance, it primarily reflects discrimination and may not fully capture clinical utility. Metrics such as the net reclassification index and decision curve analysis can provide additional insights into the potential impact of the PRS on clinical decision-making.^51,52 However, only a limited number of studies in this review reported such measures, highlighting an important gap in the current literature.

Taken together, the current literature highlights several overarching challenges in the application of the PRS in kidney-related diseases. Key limitations include phenotypic mismatches between the derivation and target traits, substantial ancestry imbalances, and methodological heterogeneity in PRS construction and evaluation. Important knowledge gaps remain, particularly the limited availability of GWAS for advanced CKD and ESKD and the underrepresentation of diverse populations. Collectively, these factors represent major barriers to clinical translation, including limited generalizability and uncertainty regarding clinical utility. Future research should prioritize expanding GWAS to diverse populations and understudied phenotypes; improving phenotype-specific and cross-ancestry PRS methodologies; integrating high-impact variants; and establishing standardized frameworks for model development, validation, and reporting.

These considerations are directly relevant to clinical translation. The current utility of PRS in nephrology appears to be context dependent. For broadly defined outcomes, such as CKD, the incremental predictive value of PRS beyond established clinical risk factors remains modest, and current evidence does not support their use as standalone tools in clinical practice. In contrast, several studies have demonstrated improved predictive performance for more specific disease subtypes such as IgA nephropathy, diabetic kidney disease, and membranous nephropathy, in which PRS may outperform or complement conventional clinical models.33, 34, 35, 36, 37 These findings suggest that the most promising applications of PRS may be in more biologically homogeneous disease subgroups than in the broadly defined CKD. However, these potential clinical implications should be interpreted with caution, because the findings summarized in this review remain exploratory and require further validation in future studies.

Genetic risk is fixed throughout life, whereas clinical risk factors evolve. This characteristic provides a theoretical advantage for early risk stratification, potentially enabling the identification of high-risk individuals before the onset of clinical risk factors such as diabetes, hypertension, or age-related decline in kidney function.

Further refinement of PRS is critical for advancing precision medicine for kidney diseases. New computational approaches continue to emerge^53,54; however, robust evaluation across diverse ancestral backgrounds is indispensable for ensuring equitable and clinically actionable risk predictions. Expanding the representation and optimizing phenotype definitions are essential to realize the full potential of genomic risk stratification in CKD, including its impact on clinical decision-making and patient outcomes.

Several issues should be considered when interpreting the findings of this study. First, there was substantial heterogeneity in terms of the PRS construction, scaling methods, outcome definitions, and covariate adjustment, limiting direct comparability. Second, differences in reporting metrics, including odds ratios, hazard ratios, and AUC or C-statistics, further complicate cross-study comparisons. Third, including both adjusted and unadjusted estimates introduces variability because of differences in confounding controls across studies. These heterogeneities inherently limit the direct quantitative comparison of PRS performance across studies.

Limitations

First, despite conducting a comprehensive scoping review that included both preprint and peer-reviewed articles across 4 major databases, our search strategy may have missed some relevant publications. This review did not include a dedicated search of grey literature sources, which may have led to the omission of relevant unpublished or nonindexed studies and introduced a potential publication bias. Second, because no language-based exclusion criteria were applied, only one of the included studies was written in Chinese. The content was interpreted using an AI-based translation tool; therefore, the accuracy of our understanding of the article could not be fully ensured. Finally, a narrative synthesis represents a secondary analysis that relies on interpretations provided in the original studies rather than on primary data. In addition, as a scoping review, this study was intended to synthesize existing evidence rather than formally evaluate the strength or consistency of the evidence across studies. Consequently, our findings should be regarded as interpretative and considered a heuristic framework rather than definitive evidence.

Conclusion

We identified 104 nephrology-related studies incorporating PRS analyses, with a predominant focus on CKD and eGFR and limited data on other clinically relevant traits, such as ESKD. Substantial ancestry imbalances and phenotypic mismatches remain key challenges. Clinically, the utility of PRS is context dependent. Although their added value for broadly defined CKD is modest, they show greater promise for more specific disease subtypes and for early risk stratification. Although not yet ready for standalone clinical use, the PRS may provide meaningful value when integrated with established clinical risk factors. Continued efforts to improve diversity, phenotype definitions, and methodological standardization are essential for translation into precision nephrology.

Disclosure

All the authors declared no competing interests.

Supplementary Material

Supplementary File (xlsx and PDF)

Table S1. List of the 104 studies and their information (xlsx). Table S2. Ancestry of discovery and validation cohorts for each paper addressing PRS (xlsx). Table S3. Previously reported PRS evaluated with eGFR as the outcome (xlsx). Table S4. Previously reported PRS evaluated with albuminuria as the outcome (xlsx). Table S5. Previously reported PRS were evaluated with eGFR decline as the outcome (xlsx). Table S6. Previously reported PRS evaluated with CKD as the outcome (xlsx). Table S7. Previously reported PRS evaluated with IgA nephropathy as the outcome (xlsx). Table S8. Previously reported PRS evaluated with MN as the outcome (xlsx). Table S9. Previously reported PRS evaluated with diabetic kidney disease as the outcome (xlsx). Table S10. Previously reported PRS evaluated with end-stage kidney disease as the outcome (xlsx). Table S11. Previously reported PRS evaluated with AKI as the outcome (xlsx). Table S12. Previously reported PRS evaluated with other outcomes (xlsx). PRISMA checklist (PDF). Search terms used for the MEDLINE, Embase, Scopus, and Web of Science databases (PDF). Detailed results for subgroups of chronic kidney disease (PDF).