To predict adverse kidney outcomes for use in optimizing medical management and clinical trial design.
In this meta-analysis of individual participant data, 43 cohorts (N = 1,621,817) from research studies, electronic medical records, and clinical trials with global representation were separated into development and validation cohorts. Models were developed and validated within strata of diabetes mellitus (presence or absence) and estimated glomerular filtration rate (eGFR; ≥60 or <60 mL/min/1.73 m2) to predict a composite of ≥40% decline in eGFR or kidney failure (i.e., receipt of kidney replacement therapy) over 2–3 years.
There were 17,399 and 24,591 events in development and validation cohorts, respectively. Models predicting ≥40% eGFR decline or kidney failure incorporated age, sex, eGFR, albuminuria, systolic blood pressure, antihypertensive medication use, history of heart failure, coronary heart disease, atrial fibrillation, smoking status, and BMI, and, in those with diabetes, hemoglobin A1c, insulin use, and oral diabetes medication use. The median C-statistic was 0.774 (interquartile range [IQR] = 0.753, 0.782) in the diabetes and higher-eGFR validation cohorts; 0.769 (IQR = 0.758, 0.808) in the diabetes and lower-eGFR validation cohorts; 0.740 (IQR = 0.717, 0.763) in the no diabetes and higher-eGFR validation cohorts; and 0.750 (IQR = 0.731, 0.785) in the no diabetes and lower-eGFR validation cohorts. Incorporating the previous 2-year eGFR slope minimally improved model performance, and then only in the higher-eGFR cohorts.
Novel prediction equations for a decline of ≥40% in eGFR can be applied successfully for use in the general population in persons with and without diabetes with higher or lower eGFR.
Chronic kidney disease (CKD) afflicts nearly 10% of the world’s population and 25% of the population with diabetes mellitus (1,2). Advanced CKD is largely irreversible; thus, early intervention is critical for reducing CKD progression and CKD-associated morbidity and mortality. The armamentarium of therapeutic options for preventing adverse kidney outcomes has greatly expanded over the past 5 years to include renin–angiotensin system inhibitors, sodium–glucose cotransporter-2 inhibitors, glucagon-like peptide-1 agonists, and selective mineralocorticoid receptor antagonists (3–8). When used early in the course of disease, these agents have the potential to prevent kidney failure, whereas in patients with advanced CKD, effective therapy may only function to delay the onset. As such, optimal medical management requires early identification of patients at high risk of decline in estimated glomerular filtration rate (eGFR) (9–13).
Accurate prediction of the risk of CKD progression can also inform clinical trial design and enrollment. For patients with eGFR <60 mL/min/1.73 m2, Tangri et al. previously developed a kidney failure risk equation (KFRE) that uses demographic and laboratory data to predict the progression of CKD to kidney failure (i.e., receipt of kidney replacement therapy), the major outcome in clinical trials in advanced CKD (14–19). In patients with eGFR ≥60 mL/min/1.73 m2, the short- and intermediate-term risks of kidney failure are very low, and clinical trials often evaluate treatment effects on 40% decline in eGFR, the major surrogate outcome for kidney failure accepted by the US Food and Drug Administration and the European Medicines Agency (6,20). However, there are no widely used prediction models for 40% decline in eGFR in the general population.
To inform risk prediction of early adverse kidney outcomes, we conducted a multinational observational study of >1 million patients in 43 cohorts. We focused on the general population, including but not limited to patients with early CKD (preserved eGFR ≥60 mL/min/1.73 m2 but urine albumin to creatinine ratio [ACR] >30 mg/g) and high cardiovascular risk. Our goal was to develop and externally validate prediction models for the composite outcome of 40% decline in eGFR or kidney failure, using variables that are readily available in the electronic medical record, with a focus on the population with diabetes.
Research Design and Methods
Study Population
Included cohorts were drawn from the CKD Prognosis Consortium, a global consortium of cohorts with data on kidney function and outcomes and at least 1,000 participants (www.ckdpc.org) (21). For the present study, cohorts were required to have measures of creatinine and albuminuria at baseline and at least 2 years of observation thereafter. In total, 43 cohorts had adequate data and all agreed to participate. The period of observation ranged from 2000 and 2019, and data from 23 countries were included. For the purpose of equation development, we divided cohorts into development and validation subsets, with development occurring in cohorts able to send individual participant data to the Data Coordinating Center as well as a random selection of 50% of the cohorts from Optum Labs Data Warehouse (OLDW), and validation occurring in the remaining cohorts. The OLDW is a longitudinal, real-world data asset with de-identified administrative claims and electronic health record data (22). This study was approved for use of de-identified data by the institutional review board at the Johns Hopkins Bloomberg School of Public Health, Baltimore, Maryland. The need for informed consent was waived by the institutional review board.
Procedures
In all cohorts, eGFR was estimated using the CKD Epidemiology Collaboration 2021 equation (23) and serum or plasma creatinine levels. Other key variables included demographics and urine ACR. For participants with measured urine protein to creatinine ratio but not ACR, values were converted to ACR using the unadjusted conversion equation (Supplementary Appendix 1) (24). For patients without diabetes, we also allowed urine dipstick protein categories and similarly converted these values to ACR. Other variables tested for inclusion were hypertension, systolic blood pressure, antihypertension medications, history of heart failure, history of coronary heart disease, history of atrial fibrillation, smoking status, BMI, as well as prior eGFR slope. Prior eGFR slope was estimated using all available creatinine measures and linear regression over the previous 2 years at the individual level and categorized as below −3 mL/min/1.73 m2 per year, between −3 and −1 mL/min/1.73 m2 per year, −1 to 1 mL/min/1.73 m2 per year (reference category), and >1 mL/min/1.73 m2 per year. For patients with diabetes, we also considered hemoglobin A1c, insulin medication use, and oral diabetes medications.
Outcomes
The primary outcome was the composite of decline in eGFR ≥40% or kidney failure. This outcome was chosen because of its status as an accepted surrogate kidney end point by the European Medicines Agency and Food and Drug Administration (12). In sensitivity analyses, we also evaluated a composite of decline in eGFR ≥30% or kidney failure and a composite of decline in eGFR ≥50% or kidney failure.
Predicting ≥40% Decline in eGFR or Kidney Failure in the General Population
We developed an equation to predict the composite of ≥40% decline in eGFR or kidney failure over 2–3 years, the typical time frame of a clinical trial. To incorporate research cohorts that had different time windows between repeated creatinine measurements, we allowed the follow-up to range from 1.5 years to 3.5 years. Our first model (model 1) incorporated only the four variables that had previously been selected for use in the KFRE (namely, age, sex, eGFR, and ACR) (14). We then evaluated the addition of previously identified indicators associated with eGFR decline: systolic blood pressure, antihypertensive medication, their interaction term, history of heart failure, history of coronary heart disease, history of atrial fibrillation, smoking status, and BMI (model 2). Finally, we tested the addition of prior eGFR slope (model 3).
Each model was developed by fitting logistic regression of the composite outcome on covariates in each development cohort and then summarizing via random-effects meta-analysis using the restricted maximum likelihood for estimation and inputs of point estimates for each cohort. To assess performance, we estimated discrimination in each validation cohort using Harrell C statistic (25) and then summarized as the median and 25th–75th percentile across cohorts. We assessed model improvement between models 2 and 3 by 1) running each model on data from the subset of patients with nonmissing, previous 2-year eGFR slope; 2) estimating change in C-statistics between the two models; and 3) meta-analyzing change in the C statistic in the same manner.
We evaluated calibration by plotting deciles of predicted versus observed risk. We also evaluated risk gradients, estimating the relative risk of events in the top decile compared with those in the lowest decile. In the case of fewer than five events in the lowest decile, that decile was iteratively combined with the adjacent decile until there were at least five events in the combined categories (26). In sensitivity analyses, we also evaluated the risk relationships when modeled in multinomial logistic regression, capturing death as a competing outcome. We also evaluated more specific data inputs for antihypertension medication (separate indicators for renin-angiotensin-aldosterone system blockers versus other antihypertension medications) and oral hypoglycemic medications (separate indicators for sodium–glucose cotransporter-2 inhibitors and GLP1RA versus other oral hypoglycemic medications) in the OLDW cohorts. These modifications did not improve the overall C statistic (data not shown). Thus, we maintained the simpler model 2 and model 3 for ease of implementation.
Analyses were done in Stata, version 16 (StataCorp) using complete case analysis and the R package “mvmeta” for the multivariate meta-analysis of all odds ratios. Statistical significance was determined using a two-sided test with a threshold P value of <0.05.
Results
Baseline Characteristics
In total, there were 707,929 participants in 20 development cohorts with 17,399 cases of ≥40% decline in eGFR or kidney failure over an average of 3 years (range, 1.5–3.5), split into strata by presence of diabetes and baseline eGFR (Table 1, Supplementary Tables 1–4). The average age of participants was 57 years, 58% were women, the mean eGFR was 86 mL/min/1.73 m2, and median urine ACR was 15 mg/g. There were 913,888 participants in 23 validation cohorts with 24,591 cases with ≥40% decline in eGFR or kidney failure. The average age, eGFR, and urine ACR were similar to those of the development cohorts, but the proportion of women was lower, owing to the inclusion of the cohort from the Veterans Administration.
Summary characteristics of cohorts used in model development and validation for prediction of the composite outcome of ≥40% decline in eGFR or kidney failure
. | Population with eGFR ≥60 mL/min/1.73 m2 . | Population with eGFR <60 mL/min/1.73 m2 . | ||||||
---|---|---|---|---|---|---|---|---|
. | Population without diabetes . | Population with diabetes . | Population without diabetes . | Population with diabetes . | ||||
. | Development . | Validation . | Development . | Validation . | Development . | Validation . | Development . | Validation . |
No. of cohorts | 19 | 18 | 20 | 20 | 19 | 21 | 20 | 21 |
No. of participants | 492,669 | 556,014 | 126,638 | 244,476 | 58,094 | 64,183 | 30,530 | 49,215 |
eGFR decline, n (%) | ||||||||
30 | 14,997 (3) | 17,389 (3) | 9,249 (7) | 19,012 (8) | 6,413 (11) | 6,947 (11) | 5,693 (19) | 9,176 (19) |
40 | 6,355 (1) | 6,643 (1) | 4,183 (3) | 8,642 (4) | 3,516 (6) | 3,815 (6) | 3,345 (11) | 5,491 (11) |
50 | 2,923 (1) | 2,866 (1) | 2,038 (2) | 4,139 (2) | 1,967 (3) | 2,166 (3) | 2,015 (7) | 3,428 (7) |
Age, years | 54 (15) | 55 (15) | 59 (13) | 61 (12) | 71 (12) | 72 (11) | 70 (10) | 71 (10) |
Female sex, % | 60 | 60 | 48 | 36 | 62 | 62 | 55 | 46 |
eGFR, mL/min/1.73 m2 | 92 (16) | 92 (17) | 90 (16) | 89 (16) | 47 (11) | 47 (10) | 46 (11) | 47 (11) |
ACR and protein to creatinine ratio available, %* | 7.2 | 9.1 | 100 | 100 | 20 | 19 | 100 | 100 |
ACR, median (IQI) | 9 (5, 21) | 9 (5, 20) | 12 (6, 31) | 12 (6, 31) | 19 (9, 98) | 23 (8, 104) | 27 (10, 109) | 28 (10, 124) |
Dipstick proteinuria + or higher, % | 6.9 | 9.1 | NA | NA | 17 | 18 | NA | NA |
Hypertension, % | 42 | 48 | 78 | 81 | 76 | 84 | 93 | 95 |
Systolic blood pressure, mmHg | 125 (16) | 126 (17) | 130 (16) | 131 (16) | 130 (18) | 130 (18) | 132 (18) | 132 (18) |
Antihypertensive medication use, % | 27 | 28 | 47 | 38 | 53 | 54 | 65 | 59 |
Heart failure, % | 2.2 | 2.8 | 5.4 | 6.2 | 12 | 14 | 18 | 19 |
Coronary heart disease, % | 9.2 | 12 | 19 | 25 | 26 | 31 | 38 | 42 |
Atrial fibrillation, % | 3.8 | 4.6 | 5.8 | 6.3 | 14 | 16 | 15 | 16 |
Current smoker, % | 5.6 | 9.3 | 8.1 | 16 | 5.2 | 7.8 | 6.3 | 11 |
Former smoker, % | 13 | 15 | 20 | 23 | 20 | 22 | 26 | 27 |
BMI, kg/m2 | 29 (7) | 30 (17) | 34 (8) | 33 (7) | 29 (6) | 29 (6) | 33 (7) | 33 (7) |
Hemoglobin A1c (SD), % | NA | NA | 7.5 (1.7) | 7.4 (1.6) | NA | NA | 7.2 (1.5) | 7.3 (1.4) |
Oral glucose-lowering medication use, % | NA | NA | 48 | 49 | NA | NA | 39 | 40 |
Insulin use, % | NA | NA | 21 | 19 | NA | NA | 27 | 25 |
Prior 2-year eGFR slope <−3 mL, % | 29.8 | 29.4 | 31.7 | 31.1 | 52.0 | 49.4 | 53.9 | 55.8 |
Prior 2-year eGFR slope between −3 and −1 mL, % | 16.4 | 15.3 | 17.8 | 18.8 | 16.6 | 17.1 | 15.9 | 15.5 |
Prior 2-year eGFR slope between −1 and 1 mL, % | 22.5 | 22.5 | 20.4 | 20.9 | 14.6 | 15.5 | 13.8 | 13.4 |
Prior 2-year eGFR slope ≥1 mL, % | 31.2 | 32.9 | 30.1 | 29.1 | 16.9 | 17.9 | 16.4 | 15.4 |
. | Population with eGFR ≥60 mL/min/1.73 m2 . | Population with eGFR <60 mL/min/1.73 m2 . | ||||||
---|---|---|---|---|---|---|---|---|
. | Population without diabetes . | Population with diabetes . | Population without diabetes . | Population with diabetes . | ||||
. | Development . | Validation . | Development . | Validation . | Development . | Validation . | Development . | Validation . |
No. of cohorts | 19 | 18 | 20 | 20 | 19 | 21 | 20 | 21 |
No. of participants | 492,669 | 556,014 | 126,638 | 244,476 | 58,094 | 64,183 | 30,530 | 49,215 |
eGFR decline, n (%) | ||||||||
30 | 14,997 (3) | 17,389 (3) | 9,249 (7) | 19,012 (8) | 6,413 (11) | 6,947 (11) | 5,693 (19) | 9,176 (19) |
40 | 6,355 (1) | 6,643 (1) | 4,183 (3) | 8,642 (4) | 3,516 (6) | 3,815 (6) | 3,345 (11) | 5,491 (11) |
50 | 2,923 (1) | 2,866 (1) | 2,038 (2) | 4,139 (2) | 1,967 (3) | 2,166 (3) | 2,015 (7) | 3,428 (7) |
Age, years | 54 (15) | 55 (15) | 59 (13) | 61 (12) | 71 (12) | 72 (11) | 70 (10) | 71 (10) |
Female sex, % | 60 | 60 | 48 | 36 | 62 | 62 | 55 | 46 |
eGFR, mL/min/1.73 m2 | 92 (16) | 92 (17) | 90 (16) | 89 (16) | 47 (11) | 47 (10) | 46 (11) | 47 (11) |
ACR and protein to creatinine ratio available, %* | 7.2 | 9.1 | 100 | 100 | 20 | 19 | 100 | 100 |
ACR, median (IQI) | 9 (5, 21) | 9 (5, 20) | 12 (6, 31) | 12 (6, 31) | 19 (9, 98) | 23 (8, 104) | 27 (10, 109) | 28 (10, 124) |
Dipstick proteinuria + or higher, % | 6.9 | 9.1 | NA | NA | 17 | 18 | NA | NA |
Hypertension, % | 42 | 48 | 78 | 81 | 76 | 84 | 93 | 95 |
Systolic blood pressure, mmHg | 125 (16) | 126 (17) | 130 (16) | 131 (16) | 130 (18) | 130 (18) | 132 (18) | 132 (18) |
Antihypertensive medication use, % | 27 | 28 | 47 | 38 | 53 | 54 | 65 | 59 |
Heart failure, % | 2.2 | 2.8 | 5.4 | 6.2 | 12 | 14 | 18 | 19 |
Coronary heart disease, % | 9.2 | 12 | 19 | 25 | 26 | 31 | 38 | 42 |
Atrial fibrillation, % | 3.8 | 4.6 | 5.8 | 6.3 | 14 | 16 | 15 | 16 |
Current smoker, % | 5.6 | 9.3 | 8.1 | 16 | 5.2 | 7.8 | 6.3 | 11 |
Former smoker, % | 13 | 15 | 20 | 23 | 20 | 22 | 26 | 27 |
BMI, kg/m2 | 29 (7) | 30 (17) | 34 (8) | 33 (7) | 29 (6) | 29 (6) | 33 (7) | 33 (7) |
Hemoglobin A1c (SD), % | NA | NA | 7.5 (1.7) | 7.4 (1.6) | NA | NA | 7.2 (1.5) | 7.3 (1.4) |
Oral glucose-lowering medication use, % | NA | NA | 48 | 49 | NA | NA | 39 | 40 |
Insulin use, % | NA | NA | 21 | 19 | NA | NA | 27 | 25 |
Prior 2-year eGFR slope <−3 mL, % | 29.8 | 29.4 | 31.7 | 31.1 | 52.0 | 49.4 | 53.9 | 55.8 |
Prior 2-year eGFR slope between −3 and −1 mL, % | 16.4 | 15.3 | 17.8 | 18.8 | 16.6 | 17.1 | 15.9 | 15.5 |
Prior 2-year eGFR slope between −1 and 1 mL, % | 22.5 | 22.5 | 20.4 | 20.9 | 14.6 | 15.5 | 13.8 | 13.4 |
Prior 2-year eGFR slope ≥1 mL, % | 31.2 | 32.9 | 30.1 | 29.1 | 16.9 | 17.9 | 16.4 | 15.4 |
Data are reported as mean (SD) except where noted otherwise. IQI, interquartile interval; NA, not applicable because the risk factor was not included in risk models.
Protein to creatinine ratio was converted to ACR; dipstick was only converted to ACR in population without diabetes
Development and Validation of a Model for the Composite End Point of ≥40% Decline in eGFR or Kidney Failure in Cohorts With eGFR ≥60 mL/min/1.73 m2
The risk prediction models for ≥40% decline in eGFR or kidney failure developed in participants with eGFR ≥60 mL/min/1.73 m2 with only four variables (model 1: age, sex, eGFR, and urine ACR) had a median C statistic (25th, 75th percentile of cohorts) of 0.704 (0.681, 0.738) in the validation cohorts of participants without diabetes and 0.750 (0.719, 0.758) in the validation cohorts of participants with diabetes (Table 2, model 1). Coefficients varied between those without and with diabetes: older age was strongly associated with ≥40% decline in eGFR or kidney failure in the population without diabetes, but less so in the population with diabetes, and female sex and lower eGFR were risk factors only among participants with diabetes. Higher urine ACR was a consistent risk factor across groups.
Models predicting the composite outcome of ≥40% decline in eGFR or kidney failure in the population with eGFR ≥60 mL/min/1.73 m2 and performance in the development and validation cohorts
. | No Diabetes . | Diabetes . | ||||
---|---|---|---|---|---|---|
. | Model 1 . | Model 2 . | Model 3 . | Model 1 . | Model 2 . | Model 3 . |
Age, per 10 years | 1.59 (1.50, 1.69) | 1.45 (1.36, 1.54) | 1.40 (1.29, 1.52) | 1.15 (1.11, 1.20) | 1.16 (1.10, 1.22) | 1.13 (1.06, 1.22) |
Male sex | 0.97 (0.88, 1.07) | 0.87 (0.79, 0.95) | 0.79 (0.69, 0.89) | 0.80 (0.74, 0.87) | 0.78 (0.71, 0.86) | 0.77 (0.68, 0.86) |
eGFR, 5 mL/min/1.73 m2 | 1.02 (1.00, 1.05) | 1.03 (1.02, 1.05) | 1.02 (1.00, 1.05) | 0.93 (0.91, 0.95) | 0.95 (0.92, 0.97) | 0.95 (0.93, 0.98) |
lnACR* | 1.59 (1.50, 1.68) | 1.52 (1.44, 1.61) | 1.46 (1.36, 1.56) | 1.64 (1.60, 1.68) | 1.51 (1.45, 1.56) | 1.48 (1.42, 1.54) |
SBP, per 20 mmHg | 1.36 (1.28, 1.44) | 1.33 (1.21, 1.46) | 1.16 (1.04, 1.30) | 1.17 (1.02, 1.34) | ||
Antihypertensive medication use | 1.30 (1.12, 1.51) | 1.39 (1.19, 1.64) | 1.33 (1.21, 1.46) | 1.32 (1.17, 1.49) | ||
SBP × HTN medications | 0.89 (0.83, 0.96) | 0.88 (0.77, 1.00) | 0.97 (0.86, 1.09) | 0.90 (0.77, 1.05) | ||
History of HF | 2.87 (2.48, 3.32) | 2.78 (2.32, 3.33) | 2.52 (2.17, 2.92) | 2.66 (2.22, 3.18) | ||
History of CHD | 1.51 (1.36, 1.67) | 1.59 (1.37, 1.83) | 1.24 (1.10, 1.41) | 1.14 (0.98, 1.33) | ||
History of Afib | 1.12 (0.91, 1.38) | 1.12 (0.89, 1.43) | 1.36 (1.04, 1.79) | 1.51 (1.15, 2.00) | ||
Current smoker | 1.46 (1.20, 1.79) | 1.46 (1.15, 1.84) | 1.13 (0.98, 1.30) | 1.19 (1.00, 1.41) | ||
Former smoker | 1.20 (1.10, 1.31) | 1.21 (1.06, 1.37) | 1.08 (0.96, 1.22) | 1.05 (0.92, 1.20) | ||
BMI, per 5 kg/m2 | 1.04 (1.01, 1.08) | 1.04 (1.00, 1.09) | 1.03 (1.00, 1.06) | 1.02 (0.98, 1.06) | ||
HbA1c, mmol/L | 1.10 (1.07, 1.14) | 1.25 (1.04, 1.51) | ||||
Oral antidiabetes medication | 0.94 (0.83, 1.06) | 1.03 (0.82, 1.29) | ||||
Insulin | 1.27 (1.08, 1.49) | 1.47 (1.24, 1.73) | ||||
Slope†, mL | ||||||
<−3 | 1.25 (1.03, 1.53) | 1.09 (1.05, 1.13) | ||||
−3 ≤ slope < −1 | 1.13 (0.93, 1.38) | 0.93 (0.80, 1.08) | ||||
≥1 | 1.69 (1.45, 1.98) | 1.21 (0.99, 1.49) | ||||
Development population, N | 456,129 | 456,129 | 181,619 | 123,201 | 123,201 | 78,285 |
Median C statistic (IQR) | 0.715 (0.679, 0.741) | 0.740 (0.702, 0.776) | 0.739 (0.703, 0.761) | 0.730 (0.698, 0.737) | 0.759 (0.738, 0.780) | 0.751 (0.717, 0.766) |
Change in C statistic from previous model/column (using same N)‡ | 0.029 (0.021, 0.038) | 0.007 (0.003, 0.010) | 0.035 (0.029, 0.041) | 0.004 (0.002, 0.006) | ||
Validation population, N | 550,179 | 550,179 | 236,284 | 238,440 | 238,440 | 142,673 |
Median C statistic (IQR) | 0.704 (0.681, 0.738) | 0.740 (0.717, 0.763) | 0.743 (0.708, 0.758) | 0.750 (0.719, 0.758) | 0.774 (0.753, 0.782) | 0.766 (0.747, 0.796) |
Change in C statistic from previous model/column (using same N)‡ | 0.031 (0.026, 0.036) | 0.008 (0.004, 0.012) | 0.028 (0.022, 0.034) | 0.003 (0.001, 0.005) |
. | No Diabetes . | Diabetes . | ||||
---|---|---|---|---|---|---|
. | Model 1 . | Model 2 . | Model 3 . | Model 1 . | Model 2 . | Model 3 . |
Age, per 10 years | 1.59 (1.50, 1.69) | 1.45 (1.36, 1.54) | 1.40 (1.29, 1.52) | 1.15 (1.11, 1.20) | 1.16 (1.10, 1.22) | 1.13 (1.06, 1.22) |
Male sex | 0.97 (0.88, 1.07) | 0.87 (0.79, 0.95) | 0.79 (0.69, 0.89) | 0.80 (0.74, 0.87) | 0.78 (0.71, 0.86) | 0.77 (0.68, 0.86) |
eGFR, 5 mL/min/1.73 m2 | 1.02 (1.00, 1.05) | 1.03 (1.02, 1.05) | 1.02 (1.00, 1.05) | 0.93 (0.91, 0.95) | 0.95 (0.92, 0.97) | 0.95 (0.93, 0.98) |
lnACR* | 1.59 (1.50, 1.68) | 1.52 (1.44, 1.61) | 1.46 (1.36, 1.56) | 1.64 (1.60, 1.68) | 1.51 (1.45, 1.56) | 1.48 (1.42, 1.54) |
SBP, per 20 mmHg | 1.36 (1.28, 1.44) | 1.33 (1.21, 1.46) | 1.16 (1.04, 1.30) | 1.17 (1.02, 1.34) | ||
Antihypertensive medication use | 1.30 (1.12, 1.51) | 1.39 (1.19, 1.64) | 1.33 (1.21, 1.46) | 1.32 (1.17, 1.49) | ||
SBP × HTN medications | 0.89 (0.83, 0.96) | 0.88 (0.77, 1.00) | 0.97 (0.86, 1.09) | 0.90 (0.77, 1.05) | ||
History of HF | 2.87 (2.48, 3.32) | 2.78 (2.32, 3.33) | 2.52 (2.17, 2.92) | 2.66 (2.22, 3.18) | ||
History of CHD | 1.51 (1.36, 1.67) | 1.59 (1.37, 1.83) | 1.24 (1.10, 1.41) | 1.14 (0.98, 1.33) | ||
History of Afib | 1.12 (0.91, 1.38) | 1.12 (0.89, 1.43) | 1.36 (1.04, 1.79) | 1.51 (1.15, 2.00) | ||
Current smoker | 1.46 (1.20, 1.79) | 1.46 (1.15, 1.84) | 1.13 (0.98, 1.30) | 1.19 (1.00, 1.41) | ||
Former smoker | 1.20 (1.10, 1.31) | 1.21 (1.06, 1.37) | 1.08 (0.96, 1.22) | 1.05 (0.92, 1.20) | ||
BMI, per 5 kg/m2 | 1.04 (1.01, 1.08) | 1.04 (1.00, 1.09) | 1.03 (1.00, 1.06) | 1.02 (0.98, 1.06) | ||
HbA1c, mmol/L | 1.10 (1.07, 1.14) | 1.25 (1.04, 1.51) | ||||
Oral antidiabetes medication | 0.94 (0.83, 1.06) | 1.03 (0.82, 1.29) | ||||
Insulin | 1.27 (1.08, 1.49) | 1.47 (1.24, 1.73) | ||||
Slope†, mL | ||||||
<−3 | 1.25 (1.03, 1.53) | 1.09 (1.05, 1.13) | ||||
−3 ≤ slope < −1 | 1.13 (0.93, 1.38) | 0.93 (0.80, 1.08) | ||||
≥1 | 1.69 (1.45, 1.98) | 1.21 (0.99, 1.49) | ||||
Development population, N | 456,129 | 456,129 | 181,619 | 123,201 | 123,201 | 78,285 |
Median C statistic (IQR) | 0.715 (0.679, 0.741) | 0.740 (0.702, 0.776) | 0.739 (0.703, 0.761) | 0.730 (0.698, 0.737) | 0.759 (0.738, 0.780) | 0.751 (0.717, 0.766) |
Change in C statistic from previous model/column (using same N)‡ | 0.029 (0.021, 0.038) | 0.007 (0.003, 0.010) | 0.035 (0.029, 0.041) | 0.004 (0.002, 0.006) | ||
Validation population, N | 550,179 | 550,179 | 236,284 | 238,440 | 238,440 | 142,673 |
Median C statistic (IQR) | 0.704 (0.681, 0.738) | 0.740 (0.717, 0.763) | 0.743 (0.708, 0.758) | 0.750 (0.719, 0.758) | 0.774 (0.753, 0.782) | 0.766 (0.747, 0.796) |
Change in C statistic from previous model/column (using same N)‡ | 0.031 (0.026, 0.036) | 0.008 (0.004, 0.012) | 0.028 (0.022, 0.034) | 0.003 (0.001, 0.005) |
Data in parentheses are 95% CIs, unless otherwise indicated. Afib, atrial fibrillation; CHD, coronary heart disease; HF, heart failure; HTN, hypertension; SBP, systolic blood pressure.
lnACR was converted by urine dipstick protein, only in the no diabetes models, using a published equation: lnACR = 2.4738 + 0.7539 × (if trace) + 1.7243 × (if +) + 3.3475 × (if ++) + 4.6399 × (if more than ++).
Reference: −1 mL/min/1.73 m2/year ≤ slope < 1 mL/min/1.73 m2/year
Change in model 2 C statistic is from model 1, run with the same sample size. Change in model 3 C statistic is from model 2, rerun with the smaller sample size of model 3 (24).
Incorporating additional variables in model 2 revealed strong associations with 40% decline in eGFR or kidney failure, particularly for systolic blood pressure, a history of heart failure, and smoking status (Table 2, model 2). Hemoglobin A1c and insulin use were also associated with the outcome in the population with diabetes. The median C statistic values (25th, 75th percentile of cohorts) for model 2 were 0.740 (0.717, 0.763) and 0.774 (0.753, 0.782) in the validation cohorts for people without and with diabetes, respectively (Supplementary Table 5). There was good calibration in the populations without and with diabetes (Fig. 1A and B). The observed risk in the top versus the bottom decile was greater with model 2 than in model 1 (median [interquartile range (IQR)] cohort risk gradient: 17.4 (14.0, 19.5) vs. 13.1 (10.6, 14.4) in those without diabetes; 16.4 (14.3, 21.3) vs. 13.2 (11.4, 15.0) in those with diabetes). Adding prior eGFR slope improved discrimination by only a modest amount in the validation cohorts among participants without and with diabetes (Table 2, model 3).
Calibration of the new equations to predict 40% decline in eGFR over 2–3 years in validation cohorts (A) with eGFR ≥60 mL/min/1.73 m2 and no diabetes; (B) with eGFR ≥60 mL/min/1.73 m2 and diabetes; (C) with eGFR <60 mL/min/1.73 m2 and no diabetes; and (D) with eGFR <60 mL/min/1.73 m2 and diabetes. Color key: light gray, <100 events; gray, 100 to ∼199 events; dark gray, 200 to ∼399 events; black, ≥400 events.
Calibration of the new equations to predict 40% decline in eGFR over 2–3 years in validation cohorts (A) with eGFR ≥60 mL/min/1.73 m2 and no diabetes; (B) with eGFR ≥60 mL/min/1.73 m2 and diabetes; (C) with eGFR <60 mL/min/1.73 m2 and no diabetes; and (D) with eGFR <60 mL/min/1.73 m2 and diabetes. Color key: light gray, <100 events; gray, 100 to ∼199 events; dark gray, 200 to ∼399 events; black, ≥400 events.
Development and Validation of a Model for the Composite End Point of ≥40% Decline in eGFR or Kidney Failure in Cohorts With eGFR <60 mL/min/1.73 m2
The risk prediction models for ≥40% decline in eGFR or kidney failure developed in participants with eGFR <60 mL/min/1.73 m2 with only four variables (model 1: age, sex, eGFR, and urine ACR) had a median C statistic (25th, 75th percentile of cohorts) of 0.712 (0.677, 0.772) in the validation cohorts of participants without diabetes and 0.760 (0.731, 0.799) in the validation cohorts of participants with diabetes (Table 3, model 1). Coefficients varied between those without and with diabetes in eGFR <60 mL/min/1.73 m2: older age was protective for ≥40% decline in eGFR or kidney failure in those with diabetes, but not in people without diabetes, and male sex was a risk factor in patients without diabetes but protective in those with diabetes. Higher urine ACR was a consistent predictor of higher risk of adverse outcomes.
Models predicting the composite outcome of ≥40% decline in eGFR or kidney failure in cohorts with eGFR <60 mL/min/1.73 m2
. | No Diabetes . | Diabetes . | ||||
---|---|---|---|---|---|---|
. | Model 1 . | Model 2 . | Model 3 . | Model 1 . | Model 2 . | Model 3 . |
Age, per 10 years | 0.97 (0.93, 1.01) | 0.92 (0.87, 0.98) | 0.86 (0.79, 0.93) | 0.86 (0.80, 0.92) | 0.84 (0.78, 0.91) | 0.84 (0.77, 0.92) |
Male sex | 1.11 (1.02, 1.20) | 1.06 (0.96, 1.18) | 1.04 (0.91, 1.19) | 0.89 (0.79, 0.99) | 0.86 (0.77, 0.97) | 0.88 (0.76, 1.01) |
eGFR, 5 mL/min/1.73 m2 | 0.83 (0.80, 0.86) | 0.85 (0.82, 0.87) | 0.83 (0.80, 0.86) | 0.91 (0.88, 0.94) | 0.93 (0.89, 0.96) | 0.92 (0.88, 0.96) |
lnACR* | 1.51 (1.46, 1.56) | 1.48 (1.43, 1.53) | 1.45 (1.39, 1.51) | 1.67 (1.59, 1.74) | 1.59 (1.51, 1.68) | 1.56 (1.48, 1.65) |
SBP, per 20 mmHg | 1.27 (1.18, 1.37) | 1.34 (1.18, 1.52) | 1.23 (1.12, 1.35) | 1.24 (1.11, 1.40) | ||
Antihypertensive medication use | 1.08 (0.95, 1.24) | 1.21 (1.05, 1.40) | 1.18 (1.02, 1.36) | 1.23 (1.03, 1.47) | ||
SBP × HTN medications | 0.98 (0.89, 1.07) | 0.99 (0.84, 1.16) | 0.95 (0.85, 1.05) | 0.95 (0.83, 1.08) | ||
History of HF | 1.63 (1.43, 1.86) | 1.70 (1.45, 1.99) | 1.52 (1.33, 1.75) | 1.62 (1.38, 1.91) | ||
History of CHD | 1.26 (1.13, 1.41) | 1.27 (1.10, 1.46) | 1.24 (1.09, 1.42) | 1.15 (0.98, 1.36) | ||
History of Afib | 1.08 (0.88, 1.31) | 1.15 (0.89, 1.47) | 1.05 (0.86, 1.27) | 0.96 (0.77, 1.19) | ||
Current smoker | 1.34 (1.08, 1.66) | 1.27 (0.93, 1.74) | 0.97 (0.76, 1.23) | 0.97 (0.73, 1.29) | ||
Former smoker | 1.19 (1.06, 1.34) | 1.17 (1.00, 1.37) | 1.15 (1.02, 1.30) | 1.15 (1.01, 1.31) | ||
BMI, per 5 kg/m2 | 0.98 (0.93, 1.02) | 0.95 (0.90, 1.00) | 1.03 (0.99, 1.06) | 1.05 (1.01, 1.10) | ||
HbA1c, mmol | 1.00 (0.96, 1.04) | 0.99 (0.79, 1.25) | ||||
Oral antidiabetes medication | 0.88 (0.76, 1.02) | 1.08 (0.82, 1.41) | ||||
Insulin | 1.10 (0.95, 1.28) | 1.24 (0.93, 1.64) | ||||
Slope†, mL | ||||||
<−3 | 0.93 (0.75, 1.15) | 0.98 (0.93, 1.03) | ||||
−3 ≤ slope < −1 | 0.97 (0.77, 1.22) | 0.95 (0.80, 1.13) | ||||
≥1 | 1.42 (1.12, 1.79) | 1.17 (0.97, 1.41) | ||||
Development population, N | 50,567 | 50,567 | 29,595 | 29,145 | 29,145 | 21,591 |
Median C statistic (IQR) | 0.702 (0.692, 0.725) | 0.735 (0.717, 0.764) | 0.739 (0.716, 0.762) | 0.763 (0.720, 0.788) | 0.787 (0.738, 0.805) | 0.775 (0.731, 0.787) |
Change in C statistic from previous model/column (using same N)‡ | 0.024 (0.018, 0.030) | 0.004 (0.001, 0.007) | 0.017 (0.013, 0.021) | 0.002 (0.001, 0.003) | ||
Validation population, N | 63,717 | 63,717 | 39,015 | 48,041 | 48,041 | 34,350 |
Median C statistic (IQR) | 0.712 (0.677, 0.772) | 0.750 (0.731, 0.785) | 0.743 (0.706, 0.793) | 0.760 (0.731, 0.799) | 0.769 (0.758, 0.808) | 0.766 (0.756, 0.808) |
Change in C statistic from previous model/column (using same N)‡ | 0.025 (0.015, 0.036) | 0.002 (−0.002, 0.006) | 0.012 (0.007, 0.018) | 0.001 (−0.000, 0.002) |
. | No Diabetes . | Diabetes . | ||||
---|---|---|---|---|---|---|
. | Model 1 . | Model 2 . | Model 3 . | Model 1 . | Model 2 . | Model 3 . |
Age, per 10 years | 0.97 (0.93, 1.01) | 0.92 (0.87, 0.98) | 0.86 (0.79, 0.93) | 0.86 (0.80, 0.92) | 0.84 (0.78, 0.91) | 0.84 (0.77, 0.92) |
Male sex | 1.11 (1.02, 1.20) | 1.06 (0.96, 1.18) | 1.04 (0.91, 1.19) | 0.89 (0.79, 0.99) | 0.86 (0.77, 0.97) | 0.88 (0.76, 1.01) |
eGFR, 5 mL/min/1.73 m2 | 0.83 (0.80, 0.86) | 0.85 (0.82, 0.87) | 0.83 (0.80, 0.86) | 0.91 (0.88, 0.94) | 0.93 (0.89, 0.96) | 0.92 (0.88, 0.96) |
lnACR* | 1.51 (1.46, 1.56) | 1.48 (1.43, 1.53) | 1.45 (1.39, 1.51) | 1.67 (1.59, 1.74) | 1.59 (1.51, 1.68) | 1.56 (1.48, 1.65) |
SBP, per 20 mmHg | 1.27 (1.18, 1.37) | 1.34 (1.18, 1.52) | 1.23 (1.12, 1.35) | 1.24 (1.11, 1.40) | ||
Antihypertensive medication use | 1.08 (0.95, 1.24) | 1.21 (1.05, 1.40) | 1.18 (1.02, 1.36) | 1.23 (1.03, 1.47) | ||
SBP × HTN medications | 0.98 (0.89, 1.07) | 0.99 (0.84, 1.16) | 0.95 (0.85, 1.05) | 0.95 (0.83, 1.08) | ||
History of HF | 1.63 (1.43, 1.86) | 1.70 (1.45, 1.99) | 1.52 (1.33, 1.75) | 1.62 (1.38, 1.91) | ||
History of CHD | 1.26 (1.13, 1.41) | 1.27 (1.10, 1.46) | 1.24 (1.09, 1.42) | 1.15 (0.98, 1.36) | ||
History of Afib | 1.08 (0.88, 1.31) | 1.15 (0.89, 1.47) | 1.05 (0.86, 1.27) | 0.96 (0.77, 1.19) | ||
Current smoker | 1.34 (1.08, 1.66) | 1.27 (0.93, 1.74) | 0.97 (0.76, 1.23) | 0.97 (0.73, 1.29) | ||
Former smoker | 1.19 (1.06, 1.34) | 1.17 (1.00, 1.37) | 1.15 (1.02, 1.30) | 1.15 (1.01, 1.31) | ||
BMI, per 5 kg/m2 | 0.98 (0.93, 1.02) | 0.95 (0.90, 1.00) | 1.03 (0.99, 1.06) | 1.05 (1.01, 1.10) | ||
HbA1c, mmol | 1.00 (0.96, 1.04) | 0.99 (0.79, 1.25) | ||||
Oral antidiabetes medication | 0.88 (0.76, 1.02) | 1.08 (0.82, 1.41) | ||||
Insulin | 1.10 (0.95, 1.28) | 1.24 (0.93, 1.64) | ||||
Slope†, mL | ||||||
<−3 | 0.93 (0.75, 1.15) | 0.98 (0.93, 1.03) | ||||
−3 ≤ slope < −1 | 0.97 (0.77, 1.22) | 0.95 (0.80, 1.13) | ||||
≥1 | 1.42 (1.12, 1.79) | 1.17 (0.97, 1.41) | ||||
Development population, N | 50,567 | 50,567 | 29,595 | 29,145 | 29,145 | 21,591 |
Median C statistic (IQR) | 0.702 (0.692, 0.725) | 0.735 (0.717, 0.764) | 0.739 (0.716, 0.762) | 0.763 (0.720, 0.788) | 0.787 (0.738, 0.805) | 0.775 (0.731, 0.787) |
Change in C statistic from previous model/column (using same N)‡ | 0.024 (0.018, 0.030) | 0.004 (0.001, 0.007) | 0.017 (0.013, 0.021) | 0.002 (0.001, 0.003) | ||
Validation population, N | 63,717 | 63,717 | 39,015 | 48,041 | 48,041 | 34,350 |
Median C statistic (IQR) | 0.712 (0.677, 0.772) | 0.750 (0.731, 0.785) | 0.743 (0.706, 0.793) | 0.760 (0.731, 0.799) | 0.769 (0.758, 0.808) | 0.766 (0.756, 0.808) |
Change in C statistic from previous model/column (using same N)‡ | 0.025 (0.015, 0.036) | 0.002 (−0.002, 0.006) | 0.012 (0.007, 0.018) | 0.001 (−0.000, 0.002) |
Data in parentheses are 95% CIs, unless otherwise indicated
lnACR was converted by urine dipstick protein, only in the no diabetes models, using a published equation as: lnACR = 2.4738 + 0.7539 × (if trace) + 1.7243 × (if +) + 3.3475 × (if ++) + 4.6399 × (if more than ++).
Reference: −1 mL/min/1.73 m2/year ≤ slope < 1 mL/min/1.73 m2/year.
Change in model 2 C statistic is from model 1, run with the same sample size. Change in model 3 C statistic is from model 2, rerun with the smaller sample size of model 3 (24).
Systolic blood pressure, a history of heart failure, and smoking status were again strong risk factors in people with eGFR <60 mL/min/1.73 m2 (Table 3, model 2). The median (25th, 75th percentile of cohorts) C-statistic for Model 2 was 0.750 (0.731, 0.785) and 0.769 (0.758,0.808) in the validation cohorts for people without and with diabetes, respectively. Calibration is shown in Fig. 1C and D. The observed risk in the top vs. bottom decile was greater with model 2 compared with model 1 (median [IQR] cohort risk gradient: 11.6 [10.2, 13.6] vs. 8.9 [8.3, 10.5]) in those without diabetes, 19.5 [16.2, 19.6] vs. 17.3 [16.2, 19.6] in those with diabetes). Adding the prior eGFR slope did not significantly improve discrimination in the validation cohorts in either the population without or with diabetes (Table 3, model 3).
Sensitivity Analyses
Model performance was similar when estimated using hypertension as a categorical variable (instead of systolic blood pressure and antihypertension medications) in both eGFR <60 mL/min/1.73 m2 and eGFR ≥60 mL/min/1.73 m2 (Supplementary Tables 6 and 7). Risk factors were similar when modeling the outcome as 30% decline or kidney failure, and as 50% decline or kidney failure, although absolute risks were higher for the former and lower for the latter (Supplementary Tables 8–11). The C statistic values for the 50% decline were generally higher, consistent with the relative rarity of the event. Finally, when multinomial models were used to capture the competing event of death, risk relationships were fairly consistent, with risk estimates, on average, 1.5%, 0.9%, 2.0%, and 2.7% lower when using the competing risk model in eGFR ≥60 without diabetes, eGFR ≥60 with diabetes, eGFR <60 without diabetes, and eGFR <60 with diabetes, respectively (Supplementary Fig. 1; Supplementary Table 12).
Conclusions
In this multinational collaborative meta-analysis including >1 million individuals across 43 cohorts, we developed new models that predict ≥40% decline in eGFR or kidney failure for use in the general population. The risk tools are publicly available (www.ckdpcrisk.org/gfrdecline40) and may be useful in medical management and in clinical trial design (9–13). We chose a ≥40% decline in eGFR or kidney failure as an outcome because it is accepted as a valid surrogate end point by regulatory bodies, but there was consistency in risk relationships when eGFR declines of ≥30% or ≥50% decline were examined. These models, when used in primary care settings, can identify patients at high risk of CKD progression even when eGFR is preserved. Additional work should evaluate how these and other equations developed for this population, such as our models to predict incidence of eGFR <60 mL/min/1.73 m2 and incidence of albuminuria in people with and without diabetes mellitus (27), could inform clinical care and trial recruitment.
Equations predicting ≥40% decline in eGFR or kidney failure may be useful if implemented in clinical practice before significant eGFR decline has occurred. Despite decades of research and health policy work, CKD remains largely unrecognized until advanced stages (eGFR <30 mL/min/1.73 m2) (28), when the window for intervention and kidney failure prevention is lost. Even patients at high risk of kidney failure remain largely unaware of their diagnosis and prognosis, and they continue to receive suboptimal care (29,30). Multiple disease-modifying therapies are available for patients: both SGLT2I and newer mineralocorticoid receptor antagonists improve CKD progression outcomes when given with renin–angiotensin system inhibitors. Thus, the new equation can identify high-risk patients who would benefit from an early triple-therapy approach to prevent kidney failure over their lifetime (5–8). Conversely, the potential side effects of triple therapy may outweigh the potential benefit in some low-risk individuals with CKD stage G1-G2 and emphasize the need for risk stratification in early disease. When combined with risk-prediction tools for cardiovascular disease, prediction tools for kidney outcomes can help personalize treatment choices, nominating specific medication regimens over others that may be less useful in a given patient.
Predicting 40% decline in eGFR emphasizes the importance of risk factors for CKD progression over current eGFR. eGFR itself is critical but is a poor treatment target because glomerular sclerosis is irreversible. In the KFRE of Tangri et al. (14) for patients with eGFR <60 mL/min/1.73 m2, eGFR is the dominant risk factor, with a relative hazard of 0.57 per 5 mL/min/1.73 m2. In our general population models to predict a 40% decline in eGFR, however, eGFR is only modestly or not associated with the outcome (relative hazard 0.83–1.03 per 5 mL/min/1.73 m2). These differences—and the fact that the vast majority of patients with early CKD do not develop kidney failure (31)—suggest that our models fill an important gap for use in general population.
Because we observed that many of the risk factor associations were different within groups categorized by the presence or absence of eGFR <60 mL/min/1.73 m2 and diabetes mellitus, we developed separate risk equations for each. For example, older age was a risk factor for eGFR decline in eGFR ≥60 mL/min/1.73 m2 but not among those with eGFR <60 mL/min/1.73 m2. The most consistent risk factor across models was albuminuria, a potentially modifiable metric of disease activity. Vascular disease, particularly heart failure, and vascular risk factors, particularly higher systolic blood pressure, also consistently heralded a higher risk of 40% decline in eGFR. Interestingly, incorporating prior eGFR slope did not greatly improve any of the risk models’ performance. Because eGFR slopes require additional calculation for use in risk tools, the logistical issues in implementation may not be worth the incremental benefit.
Strengths of this study include its large sample size and diversity in geography, ethnicity, and health system design, providing strong evidence for generalizability. Our equations use readily available inputs for accurate prediction of a 40% decline in eGFR. However, there are some limitations. First, we focused on patients who had measurements for eGFR and albuminuria (allowing quantitative and dipstick proteinuria measures among patients without diabetes). Thus, some results may be biased because of an informative measurement process or inaccuracies when cataloged in the electronic medical record. However, we did not see any differences in accuracy in cohort studies where measurements were part of scheduled study visits. Second, we were unable to test biomarkers such as cystatin C, neutrophil gelatinase-associated lipocalin, kidney injury molecule-1, or tumor necrosis factor–receptor superfamily members 1A and 1B. These tests are not available for most patients; models incorporating these tests would require a change in current practice. Because cystatin C use increases in clinical settings, inclusion or substitution of cystatin C–based eGFR for creatinine-based eGFR should also be tested. Finally, which cutoffs should define low, medium, and high risk and how they best connect to clinical actions remain to be defined.
In conclusion, our equations for predicting 40% decline in kidney function may inform clinical trial design as well as identify individuals at high risk for CKD progression for effective intervention, early in the course of disease. Implementation studies of the equations in health systems are needed.
This article contains supplementary material online at https://doi.org/10.2337/figshare.20061143.
Morgan E. Grams and Nigel J. Brunskill are co-first authors and Angela Yee-Moon Wang and Navdeep Tangri are co-senior authors on this article.
A complete list of investigators for the CKD Prognosis Consortium can be found in the appendix at the end of the article.
Article Information
Funding. The CKD Prognosis Consortium (CKD-PC) Data Coordinating Center is funded in part by a program grant from the US National Kidney Foundation and the National Institute of Diabetes and Digestive and Kidney Diseases (Grant R01DK100446). A variety of sources have supported enrollment and data collection, including laboratory measurements, and follow-up in the collaborating cohorts of the CKD-PC. These funding sources include government agencies such as the NIH and medical research councils, as well as foundations and industry sponsors listed in Supplementary Appendix 3.
The funder of this study had no role in the study design, data collection, analysis, data interpretation, or writing of the report. Some of the data reported here have been supplied by the U.S. Renal Data System. The interpretation and reporting of these data are the responsibility of the authors and in no way should be seen as an official policy or interpretation of the U.S. government.
Author Contributions. M.E.G. and Y.S. had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis. M.E.G., N.J.B., S.H.B., J.C., K.M., A.Y.-M.W., and N.T. were responsible for the study concept and design. M.E.G., S.H.B., Y.S., J.C., K.M., and A.S., along with the CKD-PC investigators/collaborators listed in the Appendix below, were involved in the acquisition of data. M.E.G., N.B., S.H.B., Y.S., J.C., K.M., A.Y.-M.W., and N.T. drafted the manuscript. All the authors contributed to the analysis and interpretation of data and to the critical revision of the manuscript for important intellectual content as well as the final decision to submit for publication. The corresponding author attests that all listed authors meet authorship criteria and that no others meeting the criteria have been omitted. M.E.G. and J.C. are the guarantors of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Data Sharing. CKD-PC has agreed with collaborating cohorts not to share data outside the consortium. Each participating cohort has its own policy for data sharing.
Duality of Interest. All authors will complete the ICMJE uniform disclosure form at www.icmje.org/coi_disclosure.pdf (available on request from the corresponding author). K.M. reports grants from NIDDK during the study, grants and personal fees from Kyowa Kirin, and fees from Akebi outside the submitted work. M.M.S. received speaker fees from AstraZeneca outside the submitted work. H.J.L.H. reports grants and other support from AstraZeneca, grants and other support from Abbvie, grants and other support from Boehringer Ingelheim, other support from CSL Behring, other support from Bayer, other support from Chinook, other support from Gilead, other support from Merck, other support from NovoNordisk, other support from Janssen, other support from Mitsubishi Tanabe, and other support from Retrophin, outside the submitted work. P.A.K. reports grants from Vifor, Astellas, Evotec, Pharmacosmos, and Unicyte; consulting fees from AstraZeneca, Vifor, Unicyte, and UCB; payment or honoraria from Vifor, AstraZeneca, Pfizer, Pharmacosmos, Napp, and Bayer; support for attending meetings and/or travel from Pharmacosmos and Vifor; and participation on Data Safety Monitoring Board or Advisory Board for Sanofi, Vifor, and Novartis; all outside the submitted work. N.S. reports Boehringer Ingleheim and Novo Nordisk grants outside the submitted work. M.W. has worked as a consultant to Amgen and Freeline over the last 3 years outside the submitted work. J.C. reports grants from the National Institute of Health and other support from Healthy.io during the conduct of the study as well as grants from the National Kidney Foundation and other support from SomaLogic outside the submitted work. G.N.N. reports personal fees, nonfinancial support, and other support from Renalytix, personal fees from Daiichi Sankyo, personal fees from Variant Bio, other support from Pensieve Health, and other support from Nexus I Connect and from Data2Wisdom LLC outside the submitted work. No other potential conflicts of interest relevant to this article were reported.
References
Appendix
CKD Prognosis Consortium (CKD-PC) investigators/collaborators:
Action in Diabetes and Vascular Disease: Preterax and Diamicron Modified Release Controlled Evaluation (ADVANCE): John Chalmers, Mark Woodward; Chronic Renal Insufficiency Cohort Study (CRIC): Chi-yuan Hsu, Ana C. Ricardo, Amanda Anderson, Panduranga Rao, and Harold Feldman; Geisinger Health System: Alex R. Chang, Kevin Ho, Jamie Green, H. Lester Kirchner; Genetics of Diabetes Audit and Research in Tayside Scotland (Go-DARTS): Samira Bell, Moneeza Siddiqui, Colin Palmer; Maccabi Health System: Varda Shalev, Gabriel Chodick; NephroTest Study: Benedicte Stengel, Marie Metzger, Martin Flamant, Pascal Houillier, Jean-Philippe Haymann; OLDW: Nikita Stempniewicz, John Cuddeback, Elizabeth Ciemins; Racial and Cardiovascular Risk Anomalies in CKD Cohort (RCAV): Csaba P. Kovesdy, Keiichi Sumida; Stockholm CREAtinine Measurements Cohort (SCREAM): Juan J. Carrero, Marco Trevisan, Carl Gustaf Elinder, Björn Wettermark; Salford Kidney Study (SKS): Philip Kalra, Rajkumar Chinnadurai, James Tollitt, Darren Green.
CKD-PC Steering Committee: Josef Coresh (chair), Shoshana H. Ballew, Alex R. Chang, Ron T. Gansevoort, Morgan E. Grams, Orlando Gutierrez, Tsuneo Konta, Anna Köttgen, Andrew S. Levey, Kunihiro Matsushita, Kevan Polkinghorne, Elke Schäffner, Mark Woodward, Luxia Zhang.
CKD-PC Data Coordinating Center: Shoshana H. Ballew (assistant project director), Jingsha Chen (programmer), Josef Coresh (co-principal investigator), Morgan E. Grams (co-principal investigator, director of Nephrology Initiatives), Kunihiro Matsushita (director), Yingying Sang (lead programmer), Aditya Surapaneni (programmer), Mark Woodward (senior statistician).