Genetic Risk Prediction for CKD: A Journey of a Thousand Miles

Abstract

Related Article, p. 19 Related Article, p. 19 By targeting interventions to high-risk population subgroups, tools that provide quantitative estimates of the risk for particular clinical outcomes have the potential to improve clinical decision making and reduce morbidity and mortality. The paradigmatic example of a clinical risk score is the Framingham risk score, which incorporates demographic (age, race, sex) and clinical (smoking and diabetes status, serum total cholesterol, and blood pressure values) variables to estimate 10-year risk for myocardial infarction.1Batsis J.A. Lopez-Jimenez F. Cardiovascular risk assessment–from individual risk prediction to estimation of global risk and change in risk in the population.BMC Med. 2010; 8: 29https://doi.org/10.1186/1741-7015-8-29Crossref PubMed Scopus (49) Google Scholar Similarly, several research groups have reported chronic kidney disease (CKD) risk scores.2Bang H. Vupputuri S. Shoham D.A. et al.Screening for Occult Renal Disease (SCORED): a simple prediction model for chronic kidney disease.Arch Intern Med. 2007; 167: 374-381Crossref PubMed Scopus (121) Google Scholar, 3Chien K.L. Lin H.J. Lee B.C. Hsu H.C. Lee Y.T. Chen M.F. A prediction model for the risk of incident chronic kidney disease.Am J Med. 2010; 123: 836-846.e2Abstract Full Text Full Text PDF PubMed Scopus (68) Google Scholar, 4Halbesma N. Jansen D.F. Heymans M.W. Stolk R.P. de Jong P.E. Gansevoort R.T. Development and validation of a general population renal risk score.Clin J Am Soc Nephrol. 2011; 6: 1731-1738Crossref PubMed Scopus (64) Google Scholar, 5Kshirsagar A.V. Bang H. Bomback A.S. et al.A simple algorithm to predict incident kidney disease.Arch Intern Med. 2008; 168: 2466-2473Crossref PubMed Scopus (96) Google Scholar The receiver operating characteristic curve C statistic values ranged from 0.67 to 0.84, although the highest value did not derive from a replication cohort (Table 1). Others have reported risk scores for developing end-stage renal disease,6Hippisley-Cox J. Coupland C. Predicting the risk of chronic kidney disease in men and women in England and Wales: prospective derivation and external validation of the QKidney Scores.BMC Fam Pract. 2010; 11: 49https://doi.org/10.1186/1471-2296-11-49Crossref PubMed Scopus (137) Google Scholar, 7Johnson E.S. Smith D.H. Thorp M.L. Yang X. Juhaeri J. Predicting the risk of end-stage renal disease in the population-based setting: a retrospective case-control study.BMC Nephrol. 2011; 12: 17https://doi.org/10.1186/1471-2369-12-17Crossref PubMed Scopus (19) Google Scholar, 8Keane W.F. Zhang Z. Lyle P.A. et al.Risk scores for predicting outcomes in patients with type 2 diabetes and nephropathy: the RENAAL study.Clin J Am Soc Nephrol. 2006; 1: 761-767Crossref PubMed Scopus (158) Google Scholar and the role of renal risk scores has been reviewed.9Taal M.W. Brenner B.M. Renal risk scores: progress and prospects.Kidney Int. 2008; 73: 1216-1219Crossref PubMed Scopus (93) Google ScholarTable 1Risk Scores for Screening and Prediction of Incident CKD Stage 3StudyDesignDiscovery CohortValidation CohortVariables in the ModelC StatisticaWhere given, the numbers in parentheses indicate the 95% confidence interval.NRI and IDIDiscovery CohortValidation CohortBang et al,2Bang H. Vupputuri S. Shoham D.A. et al.Screening for Occult Renal Disease (SCORED): a simple prediction model for chronic kidney disease.Arch Intern Med. 2007; 167: 374-381Crossref PubMed Scopus (121) Google Scholar 2007PredictionNHANES (12 y)ARICAge, sex, anemia, HTN, DM, CVD, HF, PVD, proteinuria (latter lacking in ARIC)0.88 (0.86-0.90)0.71NRKshirsagar et al,5Kshirsagar A.V. Bang H. Bomback A.S. et al.A simple algorithm to predict incident kidney disease.Arch Intern Med. 2008; 168: 2466-2473Crossref PubMed Scopus (96) Google Scholar 2008Prediction (up to 9 y)ARIC, CVS (2/3 sample)ARIC, CVS (1/3 sample)Age, sex, anemia, HTN, DM, CVD, HF, PVD0.690.68NRChien et al,3Chien K.L. Lin H.J. Lee B.C. Hsu H.C. Lee Y.T. Chen M.F. A prediction model for the risk of incident chronic kidney disease.Am J Med. 2010; 123: 836-846.e2Abstract Full Text Full Text PDF PubMed Scopus (68) Google Scholar 2010PredictionTaiwan University Hospital cohort (4 y)Taiwan community cohort (2 y)Age, BMI, DBP, T2DM, stroke0.768 (0.748-0.798)0.667 (0.631-0.703)NSHalbesma et al,4Halbesma N. Jansen D.F. Heymans M.W. Stolk R.P. de Jong P.E. Gansevoort R.T. Development and validation of a general population renal risk score.Clin J Am Soc Nephrol. 2011; 6: 1731-1738Crossref PubMed Scopus (64) Google Scholar 2011Prediction (median f/u 6.4 y)PREVENDPREVEND (bootstrap resampling)Age, SBP, CRP, albuminuria0.84 (0.82-0.86)0.84NRNote: Three studies defined the outcome as eGFR < 60 mL/min/1.73 m2, while for the study by Halbemsa et al, the outcome was rapid progression (decline rate in the top quintile) combined with eGFR < 60 mL/min/1.73 m2. The receiver operating characteristic C statistic (a C score of 0.5 is equivalent to random selection) is shown, with the comparison between discovery and replication cohorts.Abbreviations: ARIC, Atherosclerosis Risk in Communities; BMI, body mass index; CKD, chronic kidney disease; CRP, C-reactive protein; CVD, cardiovascular disease; CVS, Cardiovascular Health Study; DBP, diastolic blood pressure; DM, diabetes mellitus; HF, heart failure; HTN, hypertension; IDI, integrated discrimination improvement; NHANES, National Heath and Nutrition Examination Survey; NR, not reported; NRI, net reclassification improvement; NS, not significant; PREVEND, Prevention of Renal and Vascular End-stage Disease Study; PVD, peripheral vascular disease; SBP, systolic blood pressure; T2DM, type 2 diabetes mellitus.a Where given, the numbers in parentheses indicate the 95% confidence interval. Open table in a new tab Note: Three studies defined the outcome as eGFR < 60 mL/min/1.73 m2, while for the study by Halbemsa et al, the outcome was rapid progression (decline rate in the top quintile) combined with eGFR < 60 mL/min/1.73 m2. The receiver operating characteristic C statistic (a C score of 0.5 is equivalent to random selection) is shown, with the comparison between discovery and replication cohorts. Abbreviations: ARIC, Atherosclerosis Risk in Communities; BMI, body mass index; CKD, chronic kidney disease; CRP, C-reactive protein; CVD, cardiovascular disease; CVS, Cardiovascular Health Study; DBP, diastolic blood pressure; DM, diabetes mellitus; HF, heart failure; HTN, hypertension; IDI, integrated discrimination improvement; NHANES, National Heath and Nutrition Examination Survey; NR, not reported; NRI, net reclassification improvement; NS, not significant; PREVEND, Prevention of Renal and Vascular End-stage Disease Study; PVD, peripheral vascular disease; SBP, systolic blood pressure; T2DM, type 2 diabetes mellitus. Will genetic risk scores provide additional information to clinical risk scores, by defining the risk profile more precisely, and thus find a place in personalized nephrology care? A journey of a thousand miles begins beneath one's feet, according to Lao-Tzu, a Chinese sage of the 6th century bce. In this issue of the American Journal of Kidney Diseases, O'Seaghdha and colleagues10O'Seaghdha C.M. Yang Q. Wu H. Hwang S.-J. Fox C.S. Performance of a genetic risk score for CKD stage 3 in the general population.Am J Kidney Dis. 2012; 59: 19-24Abstract Full Text Full Text PDF PubMed Scopus (27) Google Scholar have taken a first step toward the goal of assisting clinicians to use genetic information to supplement clinical scoring systems for prediction of CKD risk. The authors used data from the Framingham Heart Study Original and Offspring cohorts, involving 2,489 participants. CKD stage 3 was defined as estimated glomerular filtration rate (eGFR) <60 mL/min/1.73 m2 based on serum creatinine using the 4-variable Modification of Diet in Renal Disease Study equation. Over a mean of 10.8 years of follow-up, 270 cases of CKD stage 3 developed. The authors used a clinical risk score, with the variables age and sex, and developed a genetic risk score, using 16 single-nucleotide polymorphisms, 1 from each of 16 loci that have previously been associated with eGFR <60 mLmin/1.73 m2 in European descent populations.11Kottgen A. Pattaro C. Boger C.A. et al.New loci associated with kidney function and chronic kidney disease.Nat Genet. 2010; 42: 376-384Crossref PubMed Scopus (628) Google Scholar Of these loci, 2 involved nonsynomous variants (ie, were non–codon changing), 2 were upstream of the transcriptional start site, 9 were located within introns, and 3 were intergenic or of uncertain location. None of these variants has been shown to affect function but instead they likely track functional alleles. In their analysis, O'Seaghdha and colleagues found similar genetic risk scores among participants with and without CKD stage 3 (17.5 ± 2.8 vs 17.3 ± 2.6 risk alleles, respectively, out of a maximum 32 risk alleles). Nevertheless, logistic regression analysis showed an age- and sex-adjusted odds ratio of 1.06 (95% confidence interval, 1.01-1;11, P = 0.03) for each risk allele, representing a 6% increase in risk. However, the C statistics for the clinical model (adjusted for age, sex, cohort status, baseline eGFR, hypertension, diabetes, and proteinuria) and the combined clinical/genetic model were 0.780 and 0.781, respectively (P = 0.06). Secondary analysis by age and by diabetes status did not alter the outcome. Interestingly, a genetic risk score with risk alleles weighted by their β (regression) coefficients from previous meta-analyses showed more risk alleles in those who did and did not develop CKD stage 3 (18.2 and 17.7, respectively; P = 0.01), but this more informed genetic risk score still did not increase the C statistic in the combined clinical/genetic risk model. The report by O'Seaghdha and colleagues has a number of limitations, as the authors point out. First, the study involved only European descent participants. As CKD risk is elevated in other ethnic groups, similar studies in diverse populations, using appropriately selected genetic variants, will be important. Second, the 16 genetic loci account for a small portion of the likely genetic contribution to CKD incidence. While certain common risk alleles may contribute to CKD risk, with the MYH9/APOL1 locus being the most salient example,12Genovese G. Friedman D.J. Ross M.D. et al.Association of trypanolytic ApoL1 variants with kidney disease in African Americans.Science. 2010; 329: 841-845Crossref PubMed Scopus (1413) Google Scholar, 13Kopp J.B. Smith M.W. Nelson G.W. et al.MYH9 is a major-effect risk gene for focal segmental glomerulosclerosis.Nat Genet. 2008; 40: 1175-1184Crossref PubMed Scopus (588) Google Scholar it appears likely that rare genetic variants at multiple loci may explain much of the missing heritability of common complex diseases, including CKD. Third, unless markers are in very strong linkage disequilibrium with the true causal variants, it may be difficult to infer mode of inheritance, and the allele model will generally be less accurate in quantifying risk than a genotype model informed by mode of inheritance. In Table 2, the results presented by O'Seaghdha and colleagues for the clinical risk score and the combined clinical risk and genetic risk scores are put in context with genetic risk score studies involving coronary heart disease and type 2 diabetes.10O'Seaghdha C.M. Yang Q. Wu H. Hwang S.-J. Fox C.S. Performance of a genetic risk score for CKD stage 3 in the general population.Am J Kidney Dis. 2012; 59: 19-24Abstract Full Text Full Text PDF PubMed Scopus (27) Google Scholar, 14Meigs J.B. Shrader P. Sullivan L.M. et al.Genotype score in addition to common risk factors for prediction of type 2 diabetes.N Engl J Med. 2008; 359: 2208-2219Crossref PubMed Scopus (606) Google Scholar, 15Paynter N.P. Chasman D.I. Pare G. et al.Association between a literature-based genetic risk score and cardiovascular events in women.JAMA. 2010; 303: 631-637Crossref PubMed Scopus (300) Google Scholar, 16Ripatti S. Tikkanen E. Orho-Melander M. et al.A multilocus genetic risk score for coronary heart disease: case-control and prospective cohort analyses.Lancet. 2010; 376: 1393-1400Abstract Full Text Full Text PDF PubMed Scopus (419) Google Scholar, 17Talmud P.J. Hingorani A.D. Cooper J.A. et al.Utility of genetic and non-genetic risk factors in prediction of type 2 diabetes: Whitehall II prospective cohort study.BMJ. 2010; 340: b4838https://doi.org/10.1136/bmj.b4838Crossref PubMed Scopus (234) Google Scholar In all 5 studies, the addition of the genetic risk score to the clinical risk score failed to improve the C statistic associated or the net reclassification improvement (NRI), although one study found gains in the clinical NRI and the integrated discrimination improvement.Table 2Comparative Performance of Clinical, Genetic, and Combined Risk Scores in Predicting Complex Disease IncidenceDiseaseStudyNo. in Cohort (Cohort Name)Replication Cohort (No.)OutcomeClinical Risk ScoreGenetic Risk ScoreModel Combining Clinical/Genetic Risk ScoresVariablesC StatisticNo. SNPsC StatisticOR (95% CI) per Risk AlleleC StatisticEffect of Adding Genetic Risk ScoreCHDPaynter et al,15Paynter N.P. Chasman D.I. Pare G. et al.Association between a literature-based genetic risk score and cardiovascular events in women.JAMA. 2010; 303: 631-637Crossref PubMed Scopus (300) Google Scholar 201019,313 (Women's Genome Health Study)None10-y MI riskAge0.701120.5171.05 (1.01-1.09)0.705NRI, 0.6%Ripatti et al,16Ripatti S. Tikkanen E. Orho-Melander M. et al.A multilocus genetic risk score for coronary heart disease: case-control and prospective cohort analyses.Lancet. 2010; 376: 1393-1400Abstract Full Text Full Text PDF PubMed Scopus (419) Google Scholar 20103,829 (Finnish case-control study)Finnish/Swedish prospective cohorts (30,720)10-y MI riskAge, sex, LDL, HDL, smoking, BMI, BP, DM0.86013ND0.881NRI, 2.2%; clinical NRI, 9.7%; IDI, 3%T2DMMeigs et al,14Meigs J.B. Shrader P. Sullivan L.M. et al.Genotype score in addition to common risk factors for prediction of type 2 diabetes.N Engl J Med. 2008; 359: 2208-2219Crossref PubMed Scopus (606) Google Scholar 20082,377 (Framingham Offspring Study)NoneIncident DMAge, sex, family history, BMI, fasting glucose, SBP, HDL, TG0.90018ND1.12 (1.07-1.17)0.901NRI, 4%Talmud et al,17Talmud P.J. Hingorani A.D. Cooper J.A. et al.Utility of genetic and non-genetic risk factors in prediction of type 2 diabetes: Whitehall II prospective cohort study.BMJ. 2010; 340: b4838https://doi.org/10.1136/bmj.b4838Crossref PubMed Scopus (234) Google Scholar 20105,355 (UK Whitehall II study)NoneIncident DMSame0.78200.54ND0.78NRI, 0.2%CKDO'Seaghdha et al,10O'Seaghdha C.M. Yang Q. Wu H. Hwang S.-J. Fox C.S. Performance of a genetic risk score for CKD stage 3 in the general population.Am J Kidney Dis. 2012; 59: 19-24Abstract Full Text Full Text PDF PubMed Scopus (27) Google Scholar 20112,489 (Framingham Heart Study)NoneeGFR < 60 after a mean of 10.8 yAge, sex, cohort (Original or Offspring), eGFR, HTN, DM, proteinuria0.78016ND1.06 (1.01-1.11)aAge and sex adjusted.0.781NDNote: For CHD and T2DM, 2 studies were selected that each that provided both clinical and genetic risk scores. For CKD, O'Seaghdha et al represent the only report to date that provides both clinical risk score and genetic risk score. All studies assumed that each allele had an equivalent effect on the primary outcome. The C statistic is shown for the clinical risk score, the genetic risk score (where available), and the combined clinical and genetic risk scores. Evaluating changes in C statistic, genetic risk scores published to date have in general not improved on clinical risk scores in predicting complex disease phenotypes. On the other hand, analytic approaches that took into account reclassification from one clinical category to another reached significance only in the report by Ripatti et al.Abbreviations: BMI, body mass index; BP, blood pressure; CHD, coronary heart disease; CI, confidence interval; CKD, chronic kidney disease; CRP, C-reactive protein; DM, diabetes mellitus; eGFR, estimated glomerular filtration rate; HDL, high-density lipoprotein; HTN, hypertension; IDI, integrated discrimination improvement; LDL, low-density lipoprotein; MI, myocardial infarction; ND, not determined; NRI, net reclassification improvement; OR, odds ratio; SBP, systolic blood pressure; SNPs, single nucleotide polymorphisms; T2DM, type 2 diabetes mellitus; TG, triglycerides; UK, United Kingdom.a Age and sex adjusted. Open table in a new tab Note: For CHD and T2DM, 2 studies were selected that each that provided both clinical and genetic risk scores. For CKD, O'Seaghdha et al represent the only report to date that provides both clinical risk score and genetic risk score. All studies assumed that each allele had an equivalent effect on the primary outcome. The C statistic is shown for the clinical risk score, the genetic risk score (where available), and the combined clinical and genetic risk scores. Evaluating changes in C statistic, genetic risk scores published to date have in general not improved on clinical risk scores in predicting complex disease phenotypes. On the other hand, analytic approaches that took into account reclassification from one clinical category to another reached significance only in the report by Ripatti et al. Abbreviations: BMI, body mass index; BP, blood pressure; CHD, coronary heart disease; CI, confidence interval; CKD, chronic kidney disease; CRP, C-reactive protein; DM, diabetes mellitus; eGFR, estimated glomerular filtration rate; HDL, high-density lipoprotein; HTN, hypertension; IDI, integrated discrimination improvement; LDL, low-density lipoprotein; MI, myocardial infarction; ND, not determined; NRI, net reclassification improvement; OR, odds ratio; SBP, systolic blood pressure; SNPs, single nucleotide polymorphisms; T2DM, type 2 diabetes mellitus; TG, triglycerides; UK, United Kingdom. Genetic risk scores attempt to combine data from multiple genetic variants to provide a risk estimate for complex genetic diseases, which typically involve multiple genetic variants, each of which has low penetrance, and environmental factors. This topic has been the subject of several recent reviews. Jostins and Barrett assessed the performance of genetic risk scores for 18 common diseases (not including CKD),18Jostins L. Barrett J.C. Genetic risk prediction in complex disease.Hum Mol Genet. 2011; 20: R182-R188Crossref PubMed Scopus (134) Google Scholar while Drawz and Sedor discuss the role and assessment of genetic testing in personalized renal medicine.19Drawz P.E. Sedor J.R. The genetics of common kidney disease: a pathway toward clinical relevance.Nat Rev Nephrol. 2011; 7: 458-468Crossref Scopus (18) Google Scholar The Evaluation of Genomic Applications in Practice and Prevention (EGAPP) working group reviewed the data for 29 genes in which variants have been implicated in cardiovascular risk and concluded that there was no evidence of net health benefit (benefits minus costs) to adding any of these genetic tests, singly or as panels, to the Framingham score.20EGAPP Working GroupRecommendations from the EGAPP Working Group: genomic profiling to assess cardiovascular risk to improve cardiovascular health.Genet Med. 2010; 12: 839-843PubMed Google Scholar A consortium meeting at the US Centers for Disease Control and Prevention in 2009 made recommendations for the generation and description of genetic risk scores, and noted that while many genetic risk scores have been published, the performance of most has been poor; exceptions noted include genetic risk scores for age-related macular degeneration, hypertriglyceridemia, and Crohn disease.21Janssens A.C. Ioannidis J.P. van Duijn C.M. Little J. Khoury M.J. Strengthening the reporting of Genetic RIsk Prediction Studies: the GRIPS Statement.PLoS Med. 2011; 8: e1000420Crossref PubMed Scopus (50) Google Scholar, 22Janssens A.C. Ioannidis J.P. Bedrosian S. et al.Strengthening the reporting of genetic risk prediction studies (GRIPS): explanation and elaboration.Eur J Hum Genet. 2011; 19https://doi.org/10.1038/ejhg.2011.27Crossref Scopus (6) Google Scholar Is the improvement in the C statistic the best yardstick to measure the incremental benefit afforded by genetic risk score? Two aspects of model accuracy can be distinguished.23Cook N.R. Use and misuse of the receiver operating characteristic curve in risk prediction.Circulation. 2007; 115: 928-935Crossref PubMed Scopus (1550) Google Scholar First, discrimination is a measure of how well a test distinguishes those with disease and those without disease, and is most commonly applied to assessing the sensitivity, specificity, and predictive value of diagnostic tests. The C statistic, which represents the area under the receiver operating characteristic curve, also is equivalent to the probability that an individual with disease has a higher test score than a healthy individual. Second, calibration is a measure of how well predicted probabilities agree with observed outcomes. It is important that the fit of predicted probabilities and observed outcomes is maintained across various subgroups (eg, deciles of risk); this can be assessed by various statistical tests, including the Hosmer-Lemeshow statistic. Some statistical tests combine elements of calibration and discrimination, such as the likelihood ratio statistic, NRI, clinical NRI, and integrated discrimination improvement; this topic has been the subject of several recent reviews.24Steyerberg EW, Pencina MJ, Lingsma HF, Kattan MW, Vickers AJ, Van Calster B. Assessing the incremental value of diagnostic and prognostic markers: a review and illustration [published online ahead of print July 4, 2011]. Eur J Clin Invest. doi:10.1111/j.1365-2362.2011.02562.x.Google Scholar, 25Pencina M.J. D'Agostino Sr, R.B. Steyerberg E.W. Extensions of net reclassification improvement calculations to measure usefulness of new biomarkers.Stat Med. 2011; 30: 11-21Crossref PubMed Scopus (1753) Google Scholar, 26Romanens M. Ackermann F. Spence J.D. et al.Improvement of cardiovascular risk prediction: time to review current knowledge, debates, and fundamentals on how to assess test characteristics.Eur J Cardiovasc Prev Rehabil. 2010; 17: 18-23Crossref PubMed Scopus (34) Google Scholar, 27Steyerberg E.W. Vickers A.J. Decision curve analysis: a discussion.Med Decis Making. 2008; 28: 146-149Crossref PubMed Scopus (93) Google Scholar As an example, the Framingham risk score classifies an individual's 10-year cardiovascular risk as low (<10%), moderate (10%-20%), and high (>20%), and the clinician could particularly benefit from knowing more about those in the moderate group. The NRI is calculated as follows:[(events reclassified higher−events reclassified lower)total events]+[(nonevents reclassified lower−nonevents reclassified higher)total events]Both elements represent improvement and so the higher the sum, the more the model improvement. The clinical NRI applies NRI analysis to a particular subset, in this Framingham case, the moderate risk group (Table 2). Although O'Seaghdha and colleagues found that the genetic risk score conferred no additional predictive information over a clinical risk score, this suggests that nongenetic factors, most of which are subject to modification by patients, are at present the only basis for kidney disease risk prediction in the general population. Nevertheless, this is the first step of an iterative process in which the growing understanding of the molecular landscape of CKD risk will almost certainly lead to integrated clinical and genetic risk scores. A similar genetic risk profile might be undertaken for diabetic nephropathy, given the growing number of associated single-nucleotide polymorphisms.28McDonough C.W. Palmer N.D. Hicks P.J. et al.A genome-wide association study for diabetic nephropathy genes in African Americans.Kidney Int. 2011; 79: 563-572Crossref PubMed Scopus (119) Google Scholar This work was supported by the Intramural Research Programs of the National Cancer Institute and the National Institute of Diabetes and Digestive and Kidney Diseases. Financial Disclosure: The authors declare that they have no relevant financial interests. Performance of a Genetic Risk Score for CKD Stage 3 in the General PopulationAmerican Journal of Kidney DiseasesVol. 59Issue 1PreviewRecent genome-wide association studies have identified multiple genetic loci that increase the risk of chronic kidney disease (CKD) in the general population. We hypothesized that knowledge of these loci might permit improved CKD risk prediction beyond that provided by traditional phenotypic risk factors. Full-Text PDF