Abstract
Patients with chronic kidney disease and severely decreased glomerular filtration rate (GFR) are at high risk for kidney failure, cardiovascular disease (CVD) and death. Accurate estimates of risk and timing of these clinical outcomes could guide patient counseling and therapy. Therefore, we developed models using data of 264,296 individuals in 30 countries participating in the international Chronic Kidney Disease Prognosis Consortium with estimated GFR (eGFR)s under 30 ml/min/1.73m2. Median participant eGFR and urine albumin-to-creatinine ratio were 24 ml/min/1.73m2 and 168 mg/g, respectively. Using competing-risk regression, random-effect meta-analysis, and Markov processes with Monte Carlo simulations, we developed two- and four-year models of the probability and timing of kidney failure requiring kidney replacement therapy (KRT), a non-fatal CVD event, and death according to age, sex, race, eGFR, albumin-to-creatinine ratio, systolic blood pressure, smoking status, diabetes mellitus, and history of CVD. Hypothetically applied to a 60-year-old white male with a history of CVD, a systolic blood pressure of 140 mmHg, an eGFR of 25 ml/min/1.73m2 and a urine albumin-to-creatinine ratio of 1000 mg/g, the four-year model predicted a 17% chance of survival after KRT, a 17% chance of survival after a CVD event, a 4% chance of survival after both, and a 28% chance of death (9% as a first event, and 19% after another CVD event or KRT). Risk predictions for KRT showed good overall agreement with the published Kidney Failure Risk Equation, and both models were well calibrated with observed risk. Thus, commonly-measured clinical characteristics can predict the timing and occurrence of clinical outcomes in patients with severely decreased GFR.
Keywords: chronic kidney disease, albuminuria, cardiovascular disease
INTRODUCTION
Chronic kidney disease (CKD) is an increasingly common global problem.1 In the developing world, disease burden is shifting from communicable to non-communicable etiologies, and the prevalence of CKD has grown with the rise of hypertension, obesity, and diabetes.2 In the developed world, the prevalence of CKD has increased with improvements in life expectancy.3 The implications of CKD include morbidity, mortality, and reduction in quality of life at the individual level and high costs at the societal level.4 Patients with severely decreased glomerular filtration rate (GFR) in particular are at high risk for adverse outcomes, including kidney failure, cardiovascular disease (CVD) events, and death.5,6 Accurate prediction of whether and when clinical outcomes will occur in patients with severely decreased GFR (eGFR <30 ml/min/1.73 m2; subsequently designated as Stage G4+) will help target efforts to treat and prevent worsening of disease.
There are a few existing tools to predict the onset of kidney failure treated with KRT (also frequently referred to as end-stage kidney disease), one of the most costly outcomes of CKD.4 Tangri and colleagues developed an absolute risk prediction tool for patients with CKD stage G3-G5 (eGFR <60 ml/min/1.73 m2) in two Canadian cohorts,7 and this Kidney Failure Risk Equation (KFRE) was subsequently validated in 31 global cohorts.8 Other prediction tools exist, but they have not undergone robust validation.9–11 None of the tools was developed specifically for a population with Stage G4+ CKD, nor do they predict other potentially more common events, such as pre-KRT death and non-fatal CVD events.
Using 29 cohorts of patients with Stage G4+ CKD participating in the international CKD Prognosis Consortium, we simultaneously assessed the risks of KRT, non-fatal CVD events, and death, applying competing risk, meta-analysis with random effects models, and Markov process methodology. The goal was to develop a 2- and 4-year calculator to predict both the probability and the order of clinical events according to nine demographic and clinical characteristics.
RESULTS
Baseline characteristics
In total, there were 264,296 participants with eGFR <30 ml/min/1.73 m2 from 29 cohorts in 30 countries for use in model development (Table 1; Supplemental Table 1). Twenty cohorts had data on non-fatal CVD events, KRT, and death; nine had data on KRT and death only. Average age ranged from 47 years (Nanjing CKD, China) to 82 years (Parcours de Soins des Personnes Agées [PSPA], France). The distribution of eGFR and albuminuria varied by cohort, but was often skewed toward the higher end of eGFR <30 ml/min/1.73 m2 and the lower end of urine albumin-to-creatinine ratio (ACR) (Supplemental Figure 1–2). Cause of CKD included diabetes, hypertension, glomerulonephritis, polycystic kidney disease, and interstitial nephritis, but was unknown in the majority of cohorts (Supplemental Table 2).
Table 1.
Outcomes and baseline characteristics of included cohorts
| study | N | Death | Kidney failure treated with KRT |
CVD event after baseline |
Mean follow- up, yrs |
Age, yrs | Systolic blood pressure, mmHg |
eGFR, ml/min/ 1.73 m2 |
Urine ACR, mg/g | Male sex |
Black race |
Histor y of CVD |
Diabetes mellitus |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| AASK (USA) | 622 | 135 | 286 | 38 | 4 (3) | 56 (12) | 135 (21) | 25 (4) | 130 (34, 488) | 60% | 100% | 53% | 1% |
| BC CKD (Canada) | 9672 | 4717 | 3036 | NA | 5 (3) | 71 (13) | 137 (23) | 24 (5) | 225 (42, 1233) | 55% | 0.41% | 16% | 50% |
| CanPREDDICT (Canada) | 1739 | 452 | 435 | 334 | 3 (2) | 69 (13) | 134 (20) | 23 (5) | 188 (37, 929) | 62% | 1.6% | 38% | 52% |
| CCF (USA) | 9256 | 3000 | 1115 | NA | 2 (1) | 73 (13) | 130 (22) | 24 (5) | 51 (13, 346) | 46% | 17% | 24% | 30% |
| CRIB (UK) | 315 | 133 | 185 | NA | 6 (3) | 62 (14) | 152 (23) | 18 (7) | 589 (118, 1345) | 61% | 5.1% | 45% | 17% |
| CRIC (USA) | 1764 | 473 | 834 | 475 | 5 (3) | 60 (11) | 131 (24) | 25 (4) | 267 (48, 1066) | 54% | 45% | 45% | 60% |
| CRISIS (UK) | 1717 | 710 | 461 | NA | 3 (3) | 66 (14) | 140 (22) | 20 (6) | 150 (55, 466) | 62% | 0.64% | 48% | 36% |
| GCKD (Germany) | 504 | 34 | 33 | 34 | 2 (0) | 64 (11) | 140 (22) | 26 (4) | 130 (23, 877) | 61% | 0% | 43% | 44% |
| Geisinger (USA) | 19293 | 10039 | 1802 | 6292 | 4 (4) | 73 (14) | 127 (22) | 24 (5) | 48 (15, 232) | 41% | 0.99% | 56% | 43% |
| GLOMMS2 (UK) | 6384 | 3283 | 265 | NA | 3 (2) | 79 (11) | 25 (5) | 44 (10, 189) | 38% | <5%† | 26% | 12% | |
| Gonryo (Japan) | 729 | 57 | 354 | 48 | 2 (2) | 67 (13) | 135 (17) | 19 (7) | 666 (318, 1401) | 59% | 0% | 27% | 38% |
| Hong Kong CKD (China) | 502 | 191 | 270 | NA | 6 (3) | 61 (12) | 138 (19) | 17 (7) | 60 (21, 150) | 56% | 0% | 27% | 46% |
| ICES-KDT (Canada) | 79272 | 42006 | 9240 | 25993 | 4 (3) | 76 (13) | 25 (5) | 53 (13, 360) | 43% | <5%† | 34% | 48% | |
| Maccabi (Israel) | 12576 | 7531 | 1693 | 3480 | 4 (3) | 76 (13) | 135 (22) | 25 (5) | 70 (10, 301) | 49% | 0% | 64% | 46% |
| MASTERPLAN (Netherlands) | 437 | 93 | 142 | 32 | 4 (1) | 61 (12) | 138 (22) | 24 (5) | 185 (53, 666) | 69% | 0% | 32% | 32% |
| MDRD (USA) | 851 | 474 | 724 | NA | 14 (7) | 51 (13) | 134 (19) | 22 (6) | 335 (64, 1002) | 60% | 10% | 17% | 9% |
| Nanjing CKD (China) | 1584 | 116 | 1003 | 108 | 4 (3) | 47 (14) | 141 (22) | 21 (6) | 1008 (550, 1839) | 54% | 0% | 12% | 21% |
| NephroTest (France) | 740 | 213 | 372 | NA | 6 (4) | 61 (14) | 139 (22) | 22 (6) | 277 (69, 820) | 67% | 11% | 24% | 36% |
| NRHP-URU (Uruguay) | 2090 | 658 | 512 | 385 | 3 (2) | 72 (13) | 135 (22) | 21 (5) | 83 (0, 655) | 49% | 0.14% | 36% | 32% |
| NZDCS (New Zealand) | 1372 | 919 | 438 | 620 | 6 (3) | 71 (12) | 138 (21) | 23 (6) | 13 (2, 93) | 43% | 0.073% | 47% | 100% |
| PSP CKD (UK) | 3522 | 1251 | 141 | 688 | 2 (1) | 80 (12) | 131 (19) | 24 (5) | 48 (18, 151) | 43% | 0.51% | 47% | 30% |
| PSPA (France) | 573 | 437 | 294 | NA | 3 (2) | 82 (5) | 145 (22) | 13 (4) | 463 (174, 1015) | 57% | 0% | 55% | 39% |
| RCAV (USA) | 78114 | 30012 | 4148 | 21672 | 3 (2) | 69 (11) | 125 (24) | 24 (5) | 38 (10, 220) | 97% | 21.6% | 61% | 58% |
| RENAAL (Multi*) | 1078 | 234 | 327 | 400 | 3 (1) | 60 (7) | 151 (21) | 26 (3) | 1604 (690, 3133) | 59% | 12.5% | 28% | 100% |
| SCREAM (Sweden) | 18486 | 12370 | 1132 | 7882 | 3 (2) | 70 (12) | 25 (5) | 112 (27, 787) | 45% | <5%† | 54% | 25% | |
| SMART (Netherlands) | 137 | 79 | 31 | 29 | 6 (4) | 65 (11) | 152 (25) | 21 (8) | 187 (47, 523) | 70% | 0% | 52% | 29% |
| SRR CKD (Sweden) | 2555 | 778 | 770 | 932 | 3 (2) | 69 (14) | 142 (23) | 21 (6) | 211 (43, 953) | 66% | <5%† | 33% | 38% |
| Sunnybrook (Canada) | 1592 | 636 | 362 | 533 | 3 (2) | 72 (14) | 136 (22) | 23 (6) | 236 (62, 807) | 54% | 0% | 17% | 41% |
| West of Scotland CKD (UK) | 6820 | 2954 | 1136 | 419 | 5 (3) | 68 (13) | 143 (24) | 24 (6) | 151 (34, 800) | 49% | 0.088%† | 25% | 21% |
| Total | 264296 | 123985 | 31541 | 70394 |
Blue color indicates administrative cohort (general population cohort from a clinical or health system database; covariates and outcomes generally from ICD codes); green color indicates referred CKD cohort (similar design to Administrative, but restricted to patients under the care of a nephrologist or in a CKD registry); red color indicates CKD research cohort (designed as a research study with planned study visits and active outcome ascertainment – patients may be similar to those in the referred CKD type).
RENAAL contains participants from 28 countries; Argentina, Austria, Brazil, Canada, Chile, China, Costa Rica, Czech Republic, Denmark, France, Germany, Hungary, Israel, Italy, Japan, Malaysia, Mexico, Netherlands, New Zealand, Peru, Portugal, Russia, Singapore, Spain, Slovakia, United Kingdom, United States of America, Venezuela
See analytic notes in Appendix 1 for the cohort in regards to the race variable.
Abbreviations: N, number of participants; KRT, kidney replacement therapy; CVD, cardiovascular disease; yrs, years; eGFR, estimated glomerular filtration rate; ACR: urine albumin-to-creatinine ratio. Numbers represent mean (standard deviation) or median (1st quartile, 3rd quartile).
Rates, risk factors, and adjusted absolute risk of adverse outcomes
Overall, there were 123,985 deaths, 31,541 events of kidney failure treated with KRT, and 70,394 CVD events identified over a mean follow-up of 3.5 years. Events were categorized not only by their occurrence but also by their timing relative to KRT and CVD events, and modeled as a function of age, sex, race, history of CVD, current smoking status, systolic blood pressure, diabetes mellitus, eGFR, and urine ACR (Figure 1). Strong risk factors for developing KRT as a first event included younger age, black race, higher systolic blood pressure, lower eGFR, and higher urine ACR (Supplemental Table 3). In contrast, strong risk factors for developing a CVD event prior to KRT included older age, previous history of CVD, and diabetes. Older age and smoking were the strongest risk factors for death prior to KRT or CVD. There was some quantitative heterogeneity across cohorts, but risk associations were qualitatively consistent (Supplemental Figure 3–4). The adjusted cumulative incidence of each event over time varied across cohorts, particularly by cohort type (Figure 2). The adjusted absolute risk of KRT as a first event was generally highest among the CKD research cohorts, whereas the risk of a CVD event or death prior to KRT was highest among the administrative cohorts. Second and third events were quantified in a similar manner (Supplemental Table 4–6 and Supplemental Figure 5–7).
Figure 1.
Diagram of states and transitions included in the 5-state Markov model
States are shown in gray ovals and include CKD Stage G4+, CVD, KRT without CVD, KRT & CVD, and death. The state transition probabilities are denoted by P1 through P8, where P is a function of age, x, and time, where x is a vector of covariates. This vector includes baseline sex, race, history of cardiovascular disease, current smoking, systolic blood pressure, diabetes status, albuminuria (P1-P4 and P6), eGFR (baseline for P1-P3, time-updated for P4 and P6), and transplantation status (for P5, P7, and P8). The probabilities of remaining in a state are denoted by P0, and P9-P11.
Figure 2.
Adjusted* cumulative incidence of (A) kidney failure requiring kidney replacement therapy, (B) cardiovascular event, and (C) death as first event from Markov model. Color coding of the lines is described in panel D. The black bold line indicates the equal weighted mean.
*Adjusted to age 60 years, half male, non-black, half history of CVD, half smoker, systolic blood pressure 140 mmHg, half diabetes, eGFR 25 ml/min/1.73 m2and urine ACR 100 mg/g. Grey shaded cohorts in panel D do not have cardiovascular events and are not included in panels A-C.
Risk prediction model: 2- and 4-year outcomes
Risk factors and adjusted absolute risk were combined using a Markov process and simulations to create a prediction model for the probability and timing of clinical events, and approximated using a multinomial model (median R2=0.99; Supplemental Table 7; http://ckdpcrisk.org/lowgfrevents/). In hypothetical scenarios, the probability of adverse events increased with longer follow-up and higher albuminuria. For example, a 60-year-old white man with a history of CVD, systolic blood pressure of 140 mmHg, eGFR of 25 ml/min/1.73 m2 and urine ACR of 30 mg/g but no current smoking or diabetes mellitus was predicted to have a 74% chance of remaining event-free at 2 years, along with a 9% chance of death and a 5% chance of KRT (Figure 3A). In contrast, a similar scenario but for urine ACR of 1000 mg/g and assessment at 4 years resulted in a prediction of event-free survival of 34%, with a 28% chance of death, a 17% chance of survival with KRT, a 17% chance of survival with CVD, and a 4% chance of both (Figure 3B). Other scenarios that dramatically affected the probability of adverse events included lower eGFR (higher risk of KRT), the presence of diabetes mellitus (higher risk of CVD events) and older age (higher risk of death) (Supplemental Figures 8–13).
Figure 3.
The probability and timing of clinical events at 2 and 4 years with increasing level of albuminuria. Top panel shows 2 years and urine ACR 30 mg/g, bottom panel shows 4 years and urine ACR 1000 mg/g. In these models, the scenario was set at age 60 years, male, white, with a history of cardiovascular disease, not a current smoker, systolic blood pressure of 140 mmHg, no diabetes, and an eGFR of 25 ml/min/1.73m2.
Comparison with the KFRE and observed risk of kidney failure treated with KRT
We compared absolute risk projections from the developed risk prediction model with the previously developed, 2-year, 4-variable KFRE for a set of scenarios holding constant the overlapping risk characteristics (age, sex, eGFR, and albuminuria) but varying others that were included only in our model (race, systolic blood pressure, diabetes mellitus, smoking status, and history of CVD), demonstrating good agreement (within-cohort R2 ranging from 0.89 to 0.97; median within-study C-statistic of 0.814 (range, 0.680 to 0.972) and 0.817 (range, 0.666 to 0.929) respectively; Supplemental Figures 14–19). Calibration to observed risk using clinically relevant categories was also good for both the developed risk prediction model as well as the KFRE (Supplemental Figure 20). Within cohorts, the prevalence of 2-year predicted KRT risk >40% (potentially an actionable threshold) was approximately 10% in most cohorts using the KFRE and slightly higher using our Markov model, but many participants had >50% predicted probability of remaining event-free at 4-years (Supplemental Figure 21–22).
Alternative risk prediction model: Assessment of variation by cohort type
Alternate versions of the risk model that incorporated adjusted absolute risk estimates from the three types of cohort (CKD research, administrative, and referred CKD) showed an approximately two-fold variation in the predicted risk of clinical events between models (Supplemental Figure 23; Figure 4). The predicted probability of KRT was higher and of death was lower in the CKD research cohort-based prediction model compared to the overall prediction model; the opposite was true in the administrative cohort-based prediction model (Supplemental Figures 24–26). Variation in adjusted absolute risk over time was less consistent by region or cause of disease (Supplemental Figures 27–28).
Figure 4.
Markov model predicted 2-year survival without kidney failure treated with kidney replacement therapy or cardiovascular events for a range of scenarios (varying systolic blood pressure, race, diabetes, history of cardiovascular disease, and smoking status) for a 60-year old man, comparing estimates using overall mean with cohort type-specific means for the baseline hazards and sub-hazards.
Alternative risk prediction model: Three-state Markov model
In sensitivity analysis, we compared the predicted probability of KRT and death from the 5-state Markov model derived from data from the 20 cohorts with all three outcomes to the predicted probability of KRT and death from a 3-state Markov model (CKD G4+, KRT, and death) derived from data from all 29 cohorts with available KRT and death (Supplemental Figure 29). Risk projections were similar (Supplemental Figure 30).
DISCUSSION
In this global consortium of 264,296 patients with eGFR <30 ml/min/1.73 m2, we developed and tested a model to predict the absolute risk and relative order of KRT, non-fatal CVD events, and death in a 2- and 4-year period. The risk calculator has been made publicly available (http://ckdpcrisk.org/lowgfrevents/) and may aid in patient counseling, including referral recommendations for transplantation or vascular access surgery. With the caveat that many of our cohorts represent incident CKD G4+ patients, we found that occurrence of events (KRT, non-fatal CVD events, or death) was not uniformly high, with nearly 50% of the participants expected to be event-free at the end of four years.
Our study provides evidence that clinical characteristics at eGFR <30 ml/min/1.73m2 have strong relationships with subsequent events, even in individuals with severely decreased GFR.12 Both lower eGFR and higher albuminuria were strong risk factors for kidney failure treated with KRT. However, the absolute risks varied substantially according to age, with the predicted 4-year risk of KRT declining from 33% in a 35-year-old to 5% in an 85-year old in a scenario with baseline eGFR of 25 ml/min/1.73 m2 and urine ACR of 100 mg/g. Not surprisingly, a history of CVD was an exceptionally strong risk factor for the occurrence of a CVD event among patients with Stage G4+ CKD, supporting the potential importance of cardiovascular risk factor reduction in these patients despite their advanced kidney disease. Interestingly, in the included cohorts, many of the participants were predicted to remain event-free in the subsequent four years. This may be in part due to selection: there was a significant subset of participants with relatively low albuminuria, which may or may not be generalizable to the greater population.
A well-validated risk equation for kidney failure requiring KRT in CKD G3+ already exists.7,8 Our study adds to the literature by simultaneously accounting for and estimating rates of competing events, particularly death. We confirm the accuracy of the KFRE in a population with lower GFR than the cohorts in which it was originally developed and validated, and we compare the KFRE to our own model, finding similar results. We might suggest that health providers and systems use the KFRE in persons with GFR <60 ml/min/1.73 m2, in whom kidney failure treated with KRT is the primary event of interest, and when a limited number of covariates are available. For patients with eGFR <30 ml/min/1.73m2 in whom there is interest in incident CVD events or death, or the sequence of such events in relation to KRT, we would suggest our newly developed equation, which uses additional covariates to produce a more refined estimate.
Strengths of our study include a very large number of patients with CKD Stage G4+ from a broad range of countries. Models were developed and rigorously tested with many different sensitivity analyses. Prediction tools incorporated nine different clinical and demographic variables and explained approximately 40-fold of the variation in explained risk, but there remained approximately 5-fold variation between cohorts that was unexplained. Type of cohort did seem to be an important contributor in risk variation, with research cohorts having markedly higher KRT risk than others, even adjusted for baseline covariates. In those cohorts for which we had data, cause of CKD was not a major contributor to variation between cohorts. Although region has previously been found to play an important role in KRT risk, we did not see consistent differences between North America and non-North American regions, perhaps due to the relatively small number of cohorts from each region.8 We had limited data with which to evaluate whether differences in therapeutic interventions such as renin-angiotensin system inhibition or statin use might partially explain variation across cohorts.13
As with all models, there were certain assumptions. Relative risks were modeled as constant over time. The cumulative incidence of competing events was scaled to the cumulative incidence of the composite event derived from Cox regression. Initiation of KRT, which we and others used as an operational definition for the major adverse kidney disease endpoint, is a treatment decision which may be influenced by factors other than kidney function. For example, our observation of older age conferring lower risk for KRT may reflect preferences for conservative care rather than a slower progression of CKD or fewer symptoms. Fatal CVD events were simply counted as death, and not CVD.
In conclusion, our model predicts the occurrence and order of non-fatal CVD events, kidney failure treated with KRT, and death in patients with eGFR <30 ml/min/1.73 m2, based on parameters that are readily available in routine clinical practice. This tool may be a useful supplement to existing risk calculators when refined estimates that take into account competing events and patterns of events are required. Additional work is needed to further characterize sources of unexplained variation between cohorts, with the ultimate goal of identifying treatment strategies and practice patterns that can prevent or forestall adverse outcomes in patients with severely decreased GFR.
METHODS
Study population
Cohorts were identified from the CKD Prognosis Consortium as well as through an open call by the Kidney Disease: Improving Global Outcomes (KDIGO). The CKD-PC has been described previously and in more detail in Appendices 1–2.14–17 Cohorts were considered eligible for the current study if they contained at least 500 patients with eGFR <30 ml/min/1.73 m2, data on albuminuria, and at least 50 events each of kidney failure requiring KRT and death. There were 29 cohorts included in analyses using a 3-state Markov model, and 20 cohorts using a 5-state Markov model. Thirteen cohorts were classified as CKD research cohorts (designed as a research study with planned study visits and active outcome ascertainment), five were classified as administrative cohorts (captured from a clinical or health system database covering an entire patient population), and 11 were classified as referred CKD cohorts (similar design as the administrative cohorts, but restricted to patients under the care of a nephrologist or in a CKD registry). This study was approved by the institutional review board at the Johns Hopkins Bloomberg School of Public Health (Baltimore, Maryland, USA).
Covariates & Outcomes
Serum creatinine was standardized to isotope dilution mass spectrometry and converted to eGFR using the CKD-EPI 2009 creatinine equation.18 Measures of albuminuria included the urine albumin-to-creatinine ratio, urine albumin excretion rate, urine protein-to-creatinine ratio, with conversion to ACR as needed.7 In analyses, urine ACR was log-transformed and scaled to ln(10), so that coefficients were interpreted reflecting differences per 10-fold higher ACR. Diabetes was defined by individual cohorts as fasting glucose ≥7.0 mmol/L (126 mg/dL), non-fasting glucose ≥11.1 mmol/L (200 mg/dL), hemoglobin A1c ≥6.5%, use of glucose lowering drugs, or self-reported diabetes. History of CVD was defined as a history of myocardial infarction, coronary revascularization, heart failure, or stroke. Smoking was categorized as current smoker versus former/never smoker. Systolic blood pressure was reported by the cohorts and treated as a linear spline in regression models, with a knot at 140 mmHg. Cause of disease was classified by individual cohorts. Outcomes included KRT, cardiovascular events, and death, and were defined using cohort-specific definitions (Appendix 1). Missing covariates (except age, sex, race, and eGFR) were estimated using multiple imputation chained equations.19–21
Competing risk analyses
The associations between baseline covariates and first outcome were determined using competing risk regression using the method of Fine and Gray and treating first KRT, first nonfatal CVD event, and pre-KRT, pre-cardiovascular death as competing events.22 This was repeated for all participants who reached KRT first, treating post-KRT CVD events and death as competing events, and for all participants who had a CVD event first, treating post-CVD event KRT and death as competing events. Only the first non-fatal CVD event after the onset of eGFR <30 ml/min/1.73 m2 was captured. At each step, a composite endpoint was also evaluated in the same manner using Cox regression. For the first event, the composite endpoint consisted of first KRT, first CVD event or death pre-KRT and pre-CVD event. For the second event after KRT, the composite endpoint consisted of post-KRT CVD events or post-KRT death. For the second event after a CVD event, the composite endpoint consisted of post-CVD event KRT or post-CVD event death. For the outcome of death after a participant had developed both KRT and a CVD event, Cox regression was used to estimate associations, since there was no competing event.
Meta-analysis and estimation of baseline sub-hazards
Fine and Gray sub-hazard ratios derived in each cohort were pooled using random-effects meta-analysis. Heterogeneity was evaluated using forest plots and I2 statistics.23–25 Cohort-specific adjusted baseline sub-hazards were estimated in each cohort using competing risk regression, holding sub-hazard ratios constant and equal to the meta-analyzed sub-hazard ratios, but allowing the baseline sub-hazard to vary between cohorts. The baseline sub-hazards were then used to calculate the adjusted cumulative incidence of each event over time. Baseline sub-hazards (i.e., the adjusted absolute risk over time) were displayed graphically to evaluate heterogeneity and summarized as the equal-weighted mean over cohorts and, for cohort type-specific analysis, the equal-weighted mean within cohort type. Note that, for the composite endpoints as well as event of death after KRT and a CVD event (where there is no competing event), Cox regression and baseline hazards were used, but the procedure was otherwise the same. A Weibull model was then fit on the equal-weighted mean adjusted sub-hazard (or hazard), in order to allow a smooth, parametric estimate for use in the Markov process.
Markov process and simulations of absolute risks
The combination of parameters from the Weibull model and the meta-analyzed sub-hazard ratios were used to predict time-varying state transition probabilities (e.g., the probability of moving from CKD G4+ to first KRT) for a given set of baseline covariates (age, sex, race, history of CVD, smoking status, diabetes mellitus, systolic blood pressure, eGFR, and urine ACR). In order to ensure that the probabilities for each state summed to 1, we scaled the cumulative probability of the events (for the first state transition, first KRT, first CVD event, and first death) to the cumulative probability of a composite endpoint ascertained using Cox proportional hazards model, as done previously.26 These state transition probabilities thus varied by time, age, and baseline covariates and were incorporated in a heterogeneous Markov process using a cycle length of 1 month and a time horizon of 2- and 4-years. Outcomes were estimated using 10,000 simulations for each scenario, where a scenario corresponds to a set of covariates. In other words, each iteration corresponds to one hypothetical person with the given set of covariates, and variation in the results of the iteration represent the stochastic natures in which persons traverse the Markov model. In order to assess the sensitivity of risk prediction to cohort type, we repeated the procedures using cohort-type specific parameters for the baseline sub-hazards. We also repeated analyses in a 3-state model (CKD G4+, KRT, and death; Supplemental Figure 29) to compare the results. To evaluate sources of unexplained variation, we examined the distribution of cumulative incidence of events by type of cohort, region, and prevalence of different causes of CKD.
Development of a web calculator
In order to implement the Markov process as a web tool, we developed an estimating equation on simulated estimates for 3,702 baseline scenarios (every combination of age (35, 45, 55, 60, 65, 75, 80 and 85 years), sex, race (black and non-black), diabetes status, history of CVD status, smoking status (current smoker and never/former smoker), systolic blood pressure (180 and 140 mmHg), eGFR (15 and 25 ml/min/1.73 m2), and ACR (30, 100 and 1000 mg/g)). To do this, we fit multinomial models and weighted by the inverse probability of each outcome from simulations (e.g., KRT only, CVD only, death only, KRT followed by CVD, CVD followed by KRT, etc.). Multinomial models incorporated all the available covariates and two-way interactions significant for any of the outcomes. Calibration of the multinomial model to the simulated outcomes was assessed using R2 and root-mean-squared errors for each outcome. Functional forms of covariates were the same as those used in the competing risk regression. For the purposes of this manuscript, predicted probabilities in the figures and text stem from the multinomial model.
Comparison of our developed risk model with the KFRE and observed risk
We compared absolute risk estimates of the probability of KRT from our newly developed risk model to that calculated in the absence of competing events using the previously published KFRE.7,8 To do this, we held shared variables constant (age, sex, eGFR, and ACR) and varied the covariates unique to our model (race, systolic blood pressure, diabetes mellitus, smoking status, and history of CVD), and we assessed R2 within cohorts. We also compared risk predictions from our developed risk model as well as the KFRE to observed KRT risk. To do this, we classified predicted risk categories into <20% 20–40%, >40% probability of KRT in the subsequent two years (clinically meaningful thresholds) using our developed risk model and the KFRE, and then we plotted the mean risk estimate against the observed risk within each category by cohort. Discrimination was assessed using the C-statistic.
All analyses were done in Stata 14 MP (College Station, TX).
Supplementary Material
Appendix 1. Data analysis overview and analytic notes for some of individual studies
Appendix 2. Acronyms or abbreviations for studies included in the current report and their key references linked to the Web references
Appendix 3. Acknowledgements and funding for collaborating cohorts
Supplemental Table 1. Underlying data selection by cohort
Supplemental Table 2. Cause of chronic kidney disease
Supplemental Table 3. Random effects meta-analysis of subhazard ratios for KRT, non-fatal CVD, or death as the first event after eGFR <30 ml/min/1.73 m2, 20 cohorts with information on all three outcomes
Supplemental Table 4. Random effects meta-analysis of subhazard ratios for non-fatal CVD or death as the second event after KRT, 20 cohorts with information on all three outcomes.
Supplemental Table 5. Random effects meta-analysis of subhazard ratios for KRT or death as the second event after non-fatal CVD, 20 cohorts with information on all three outcomes.
Supplemental Table 6. Random effects meta-analysis of subhazard ratios for death as the third event after KRT and CVD, 20 cohorts with information on all three outcomes.
Supplemental Table 7. Multinomial coefficients from the model approximating the Markov chain results, 2 (a) and 4 (b) years
Supplemental Figure 1. Distribution of eGFR by cohort, color-coded by cohort type.
Supplemental Figure 2. Distribution of ACR by
Supplemental Figure 3. Forest plot of the subhazard ratio associated with 5 ml/min/1.73 m2 lower eGFR for risk of first KRT, across 20 cohorts with data on KRT, CVD and death.
Supplemental Figure 4. Forest plot of the subhazard ratio associated with 10-fold higher urine albumin-to-creatinine ratio for risk of first KRT, across 20 cohorts with data on KRT, CVD, and death.
Supplemental Figure 5. Adjusted* cumulative incidence of (A) CVD after first KRT and (B) death after first KRT from Markov model. Color coding of the lines is described in panel C.
Supplemental Figure 6. Adjusted* cumulative incidence of (A) KRT after first non-fatal CVD and (B) death after first non-fatal CVD from Markov model. Color coding of the lines is described in panel C.
Supplemental Figure 7. Adjusted* cumulative incidence of (A) death after having experienced both KRT and CVD from Markov model. Color coding of the lines is described in panel B.
Supplemental Figure 8. The probability and timing of clinical outcomes at 2 and 4 years with increasing level of ACR.
Supplemental Figure 9. The probability and timing of clinical outcomes at 4 years, varied by diabetes status.
Supplemental Figure 10. The probability and timing of clinical outcomes at 4 years, varied by SBP level.
Supplemental Figure 11. The probability and timing of clinical outcomes at 4 years, varied by history of CVD.
Supplemental Figure 12. The probability and timing of clinical outcomes at 4 years, varied by age.
Supplemental Figure 13. The probability and timing of clinical outcomes at 4 years, with further subdivision by age.
Supplemental Figure 14. Absolute risk projections for KRT at 2-years for a set of scenarios for a 35-year old man (A) or woman (B).
Supplemental Figure 15. Absolute risk projections for KRT at 2-years for a set of scenarios for a 45-year old man (A) or woman (B).
Supplemental Figure 16. Absolute risk projections for KRT at 2-years for a set of scenarios for a 55-year old man (A) or woman (B).
Supplemental Figure 17. Absolute risk projections for KRT at 2-years for a set of scenarios for a 65-year old man (A) or woman (B).
Supplemental Figure 18. Absolute risk projections for KRT at 2-years for a set of scenarios for a 75-year old man (A) or woman (B).
Supplemental Figure 19. Absolute risk projections for KRT at 2-years for a set of scenarios for an 85-year old man (A) or woman (B).
Supplemental Figure 20. Calibration of the predicted versus observed 2-year risk of KRT for the (A) kidney failure risk equation (KFRE) and (B) MCMC model
Supplemental Figure 21. Prevalence of 2-year predicted KRT risk for the (A) kidney failure risk equation (KFRE) and (B) MCMC model.
Supplemental Figure 22. Distribution of 4-year event free survival
Supplemental Figure 23. Equal-weighted mean of the adjusted cumulative incidence curves for first KRT (A), first CVD (B), and death as a first event (C) within the three types of cohort (CKD research, administrative, and referred CKD).
Supplemental Figure 24. The probability and timing of clinical outcomes at 4 years overall and by type of cohort.
Supplemental Figure 25. The probability and timing of clinical outcomes at 4 years overall and by type of cohort.
Supplemental Figure 26. The probability and timing of clinical outcomes at 4 years overall and by type of cohort.
Supplemental Figure 27. Cumulative incidence of first KRT (A), first CVD (B), and death as a first event (C) by region.
Supplemental Figure 28. Adjusted absolute risk of KRT by cause of CKD in cohorts with available data (Supplemental Table 2).
Supplemental Figure 29. Diagram of states and transitions included in the 3-state Markov model.
Supplemental Figure 30. Absolute KRT (A) or death (B) risk projections at 2-years for a set of scenarios for a 60-year old man from a 5-state Markov model compared to a 3-state Markov model.
Acknowledgments
This project was funded by the Kidney Disease: Improving Global Outcomes Foundation. The CKD-PC Data Coordinating Center is funded in part by a program grant from the US National Kidney Foundation, the Kidney Disease: Improving Global Outcomes Foundation, and the National Institute of Diabetes and Digestive and Kidney Diseases (R01DK100446-01). 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 national institutes of health and medical research councils as well as foundations and industry sponsors listed in appendix 3. The funders had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.
Footnotes
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
CKD-PC investigators/collaborators (study acronyms/abbreviations are listed in appendix 2):
AASK: Brad Astor, Lawrence Appel; BC CKD: Adeera Levin, Ognjenka Djurdjev; CanPREDDICT: Adeera Levin, Mila Tang, Ognjenka Djurdjev; CCF: Sankar D Navaneethan, Stacey E Jolly, Jesse D Schold, Joseph V Nally Jr; CRIB: David C Wheeler, Jonathan Emberson, John Townend, Martin Landray; CRIC: Harold I Feldman, Chi-yuan Hsu, James Lash, Lawrence Appel; CRISIS: Philip A Kalra, James Ritchie, Raman Maharajan, Rachel Middleton, Donal J O’Donoghue; GCKD: Kai-Uwe Eckardt, Markus P Schneider, Anna Köttgen, Florian Kronenberg, Barbara Bärthlein; Geisinger: Alex R Chang, Jamie Green, H Lester Kirchner, Kevin Ho; GLOMMS2: Angharad Marks, Corri Black, Gordon Prescott, Nick Fluck; Gonryo: Masaaki Nakayama, Mariko Miyazaki, Tae Yamamoto, Gen Yamada; Hong Kong CKD: Angela Yee Moon Wang, Sharon Cheung, Sharon Wong, Jessie Chu, Henry Wu; ICES-KDT: Amit X Garg, Eric McArthur, Danielle M Nash; Maccabi: Varda Shalev, Gabriel Chodick; MASTERPLAN: Peter J Blankestijn, Jack Wetzels, Arjan van Zuilen, Jan van den Brand; MDRD: Andrew S Levey, Lesley A Inker, Mark Sarnak, Hocine Tighiouart; Nanjing CKD: Haitao Zhang; NephroTest: Benedicte Stengel, Marie Metzger, Martin Flamant, Pascal Houillier, Jean-Philippe Haymann; NRHP-URU: Pablo Rios, Nelson Mazzuchi, Liliana Gadola, Verónica Lamadrid, Laura Sola; NZDCS: John Collins, C Raina Elley, Timothy Kenealy; PSPA: Olivier Moranne, Cecile Couchoud, Cecile Vigneau; PSP CKD: Nigel J Brunskill, Rupert Major, David Shepherd, James Medcalf; RCAV: Csaba P Kovesdy, Kamyar Kalantar-Zadeh, Miklos Z Molnar, Keiichi Sumida, Praveen Potukuchi; RENAAL: Hiddo JL Heerspink, Dick de Zeeuw, Barry Brenner; SCREAM: Juan Jesus Carrero, Alessandro Gasparini, Abdul Rashid Qureshi, Carl Gustaf Elinder; SMART: Frank LJ Visseren, Yolanda van der Graaf; SRR CKD: Marie Evans, Maria Stendahl, Staffan Schön, Mårten Segelmark, Karl-Göran Prütz; Sunnybrook: David Naimark, Navdeep Tangri; West of Scotland CKD: Patrick B Mark, Jamie P Traynor, Colin C Geddes, Peter C Thomson
CKD-PC Steering Committee: Alex R Chang, Josef Coresh (Chair), Ron T Gansevoort, Morgan E. Grams, Anna Köttgen, Andrew S Levey, Kunihiro Matsushita, Mark Woodward, Luxia Zhang
CKD-PC Data Coordinating Center: Shoshana H Ballew (Assistant Project Director), Jingsha Chen (Programmer), Josef Coresh (Principal Investigator), Morgan E Grams (Director of Nephrology Initiatives), Lucia Kwak (Programmer), Kunihiro Matsushita (Director), Yingying Sang (Lead Programmer), Aditya Surapaneni (Programmer), Mark Woodward (Senior Statistician)
KDIGO Controversies Conference on Prognosis and Optimal Management of Patients with Advanced CKD: Kai-Uwe Eckardt (Conference Co-Chair), Brenda Hemmelgarn (Conference Co-Chair), David C Wheeler (KDIGO Co-Chair), Wolfgang Winkelmayer (KDIGO Co-Chair), John Davis (CEO), Danielle Green (Managing Director), Michael Cheung (Chief Scientific Officer), Tanya Green (Communications Director), Melissa McMahan (Programs Director)
Conflict of Interest Disclosures: JJC received consulting fees from Astellas, lecture fees from Abbott, and grant support from AstraZeneca and ViforPharma donated to Karolinska Institutet. HJLH received consulting fees from Abbvie, AstraZeneca, Boehringer Ingelheim, Fresenius, Janssen, and Merck; lecture fees from AstraZeneca and Boehringer Ingelheim; and grant support from AstraZeneca and Boehringer Ingelheim. KH received lecture fees and is under negotiation for grant funding from Sanofi-Genzyme and has royalties from Partner’s Healthcare patent #5,356,775. SDN received grant support from the National Institutes of Health. BS received consulting fees from Merck Sharp & Dohme and grant support from Amgen, Baxter, Lilly, Fresenius, GlaxoSmithKline, Merck Sharp & Dohme, and Otsuka. AY-MW received lecture fees from Sanofi and travel support from Sanofi and Fresenius Medical Care. AK received grant support from AstraZeneca. ASL received grant support from the National Kidney Foundation, National Institutes of Health, AstraZeneca, and Pharmalink, and holds a provisional patent along with JC (JC, Lesley Inker, and ASL) filed August 15, 2014 –“Precise estimation of glomerular filtration rate from multiple biomarkers” PCT/ US2015/044567. The technology is not licensed in whole or in part to any company. Tufts Medical Center, John Hopkins University, and Metabolon Inc. have a collaboration agreement to develop a product to estimate glomerular filtration rate from a panel of markers. MW received consulting fees from Amgen. JC received grant support from the National Institutes of Health and National Kidney Foundation. All the other authors declared no competing interests.
Some of the data reported here have been supplied by the United States Renal Data System (USRDS). 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.
References
- 1.Levin A, Tonelli M, Bonventre J, et al. Global kidney health 2017 and beyond: a roadmap for closing gaps in care, research, and policy. Lancet. 2017 doi: 10.1016/S0140-6736(17)30788-2. [DOI] [PubMed] [Google Scholar]
- 2.Lozano R, Naghavi M, Foreman K, et al. Global and regional mortality from 235 causes of death for 20 age groups in 1990 and 2010: a systematic analysis for the Global Burden of Disease Study 2010. Lancet. 2012;380:2095–2128. doi: 10.1016/S0140-6736(12)61728-0. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Grams ME, Chow EK, Segev DL, et al. Lifetime incidence of CKD stages 3–5 in the United States. Am J Kidney Dis. 2013;62:245–252. doi: 10.1053/j.ajkd.2013.03.009. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.National Institutes of Health NIoDaDaKD, editor. United States Renal Data System: 2016 USRDS annual data report: Epidemiology of kidney disease in the United States. Bethesda, MD: 2016. [Google Scholar]
- 5.Astor BC, Matsushita K, Gansevoort RT, et al. Lower estimated glomerular filtration rate and higher albuminuria are associated with mortality and end-stage renal disease. A collaborative meta-analysis of kidney disease population cohorts. Kidney Int. 2011;79:1331–1340. doi: 10.1038/ki.2010.550. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Chronic Kidney Disease Prognosis C. Matsushita K, van der Velde M, et al. Association of estimated glomerular filtration rate and albuminuria with all-cause and cardiovascular mortality in general population cohorts: a collaborative meta-analysis. Lancet. 2010;375:2073–2081. doi: 10.1016/S0140-6736(10)60674-5. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Tangri N, Stevens LA, Griffith J, et al. A predictive model for progression of chronic kidney disease to kidney failure. JAMA. 2011;305:1553–1559. doi: 10.1001/jama.2011.451. [DOI] [PubMed] [Google Scholar]
- 8.Tangri N, Grams ME, Levey AS, et al. Multinational Assessment of Accuracy of Equations for Predicting Risk of Kidney Failure: A Meta-analysis. JAMA. 2016;315:164–174. doi: 10.1001/jama.2015.18202. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Schroeder EB, Yang X, Thorp ML, et al. Predicting 5-Year Risk of RRT in Stage 3 or 4 CKD: Development and External Validation. Clin J Am Soc Nephrol. 2017;12:87–94. doi: 10.2215/CJN.01290216. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Weiss JW, Platt RW, Thorp ML, et al. Predicting mortality in older adults with kidney disease: a pragmatic prediction model. J Am Geriatr Soc. 2015;63:508–515. doi: 10.1111/jgs.13257. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Johnson ES, Thorp ML, Yang X, et al. Predicting renal replacement therapy and mortality in CKD. Am J Kidney Dis. 2007;50:559–565. doi: 10.1053/j.ajkd.2007.07.006. [DOI] [PubMed] [Google Scholar]
- 12.Evans M, Grams ME, Sang Y, et al. Risk factors for prognosis in patients with severely decreased GFR. Kidney Int Rep. doi: 10.1016/j.ekir.2018.01.002. https://doi.org/10.1016/j.ekir.2018.01.002. [DOI] [PMC free article] [PubMed]
- 13.Schmidt M, Mansfield KE, Bhaskaran K, et al. Adherence to guidelines for creatinine and potassium monitoring and discontinuation following renin-angiotensin system blockade: a UK general practice-based cohort study. BMJ Open. 2017;7:e012818. doi: 10.1136/bmjopen-2016-012818. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Matsushita K, Coresh J, Sang Y, et al. Estimated glomerular filtration rate and albuminuria for prediction of cardiovascular outcomes: a collaborative meta-analysis of individual participant data. The lancet Diabetes & endocrinology. 2015;3:514–525. doi: 10.1016/S2213-8587(15)00040-6. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Coresh J, Turin TC, Matsushita K, et al. Decline in estimated glomerular filtration rate and subsequent risk of end-stage renal disease and mortality. JAMA. 2014;311:2518–2531. doi: 10.1001/jama.2014.6634. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Kovesdy CP, Coresh J, Ballew SH, et al. Past Decline Versus Current eGFR and Subsequent ESRD Risk. J Am Soc Nephrol. 2015 doi: 10.1681/ASN.2015060687. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Matsushita K, Ballew SH, Astor BC, et al. Cohort profile: the chronic kidney disease prognosis consortium. Int J Epidemiol. 2013;42:1660–1668. doi: 10.1093/ije/dys173. [DOI] [PubMed] [Google Scholar]
- 18.Levey AS, Stevens LA, Schmid CH, et al. A new equation to estimate glomerular filtration rate. Ann Intern Med. 2009;150:604–612. doi: 10.7326/0003-4819-150-9-200905050-00006. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Moons KG, Donders RA, Stijnen T, et al. Using the outcome for imputation of missing predictor values was preferred. J Clin Epidemiol. 2006;59:1092–1101. doi: 10.1016/j.jclinepi.2006.01.009. [DOI] [PubMed] [Google Scholar]
- 20.White IR, Royston P. Imputing missing covariate values for the Cox model. Stat Med. 2009;28:1982–1998. doi: 10.1002/sim.3618. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.White IR, Royston P, Wood AM. Multiple imputation using chained equations: Issues and guidance for practice. Stat Med. 2011;30:377–399. doi: 10.1002/sim.4067. [DOI] [PubMed] [Google Scholar]
- 22.Fine JP, Gray RJ. A Proportional Hazards Model for the Subdistribution of a Competing Risk. J Am Stat Assoc. 1999;94:496–509. [Google Scholar]
- 23.Borenstein M, Hedges LV, Higgins JP, et al. Introduction to meta-analysis. John Wiley & Sons; Chichester, U.K: 2009. [Google Scholar]
- 24.Higgins JP, Thompson SG, Deeks JJ, et al. Measuring inconsistency in meta-analyses. BMJ. 2003;327:557–560. doi: 10.1136/bmj.327.7414.557. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Riley RD, Higgins JP, Deeks JJ. Interpretation of random effects meta-analyses. BMJ. 2011;342:d549. doi: 10.1136/bmj.d549. [DOI] [PubMed] [Google Scholar]
- 26.Grams ME, Yang W, Rebholz CM, et al. Risks of Adverse Events in Advanced CKD: The Chronic Renal Insufficiency Cohort (CRIC) Study. Am J Kidney Dis. 2017 doi: 10.1053/j.ajkd.2017.01.050. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Appendix 1. Data analysis overview and analytic notes for some of individual studies
Appendix 2. Acronyms or abbreviations for studies included in the current report and their key references linked to the Web references
Appendix 3. Acknowledgements and funding for collaborating cohorts
Supplemental Table 1. Underlying data selection by cohort
Supplemental Table 2. Cause of chronic kidney disease
Supplemental Table 3. Random effects meta-analysis of subhazard ratios for KRT, non-fatal CVD, or death as the first event after eGFR <30 ml/min/1.73 m2, 20 cohorts with information on all three outcomes
Supplemental Table 4. Random effects meta-analysis of subhazard ratios for non-fatal CVD or death as the second event after KRT, 20 cohorts with information on all three outcomes.
Supplemental Table 5. Random effects meta-analysis of subhazard ratios for KRT or death as the second event after non-fatal CVD, 20 cohorts with information on all three outcomes.
Supplemental Table 6. Random effects meta-analysis of subhazard ratios for death as the third event after KRT and CVD, 20 cohorts with information on all three outcomes.
Supplemental Table 7. Multinomial coefficients from the model approximating the Markov chain results, 2 (a) and 4 (b) years
Supplemental Figure 1. Distribution of eGFR by cohort, color-coded by cohort type.
Supplemental Figure 2. Distribution of ACR by
Supplemental Figure 3. Forest plot of the subhazard ratio associated with 5 ml/min/1.73 m2 lower eGFR for risk of first KRT, across 20 cohorts with data on KRT, CVD and death.
Supplemental Figure 4. Forest plot of the subhazard ratio associated with 10-fold higher urine albumin-to-creatinine ratio for risk of first KRT, across 20 cohorts with data on KRT, CVD, and death.
Supplemental Figure 5. Adjusted* cumulative incidence of (A) CVD after first KRT and (B) death after first KRT from Markov model. Color coding of the lines is described in panel C.
Supplemental Figure 6. Adjusted* cumulative incidence of (A) KRT after first non-fatal CVD and (B) death after first non-fatal CVD from Markov model. Color coding of the lines is described in panel C.
Supplemental Figure 7. Adjusted* cumulative incidence of (A) death after having experienced both KRT and CVD from Markov model. Color coding of the lines is described in panel B.
Supplemental Figure 8. The probability and timing of clinical outcomes at 2 and 4 years with increasing level of ACR.
Supplemental Figure 9. The probability and timing of clinical outcomes at 4 years, varied by diabetes status.
Supplemental Figure 10. The probability and timing of clinical outcomes at 4 years, varied by SBP level.
Supplemental Figure 11. The probability and timing of clinical outcomes at 4 years, varied by history of CVD.
Supplemental Figure 12. The probability and timing of clinical outcomes at 4 years, varied by age.
Supplemental Figure 13. The probability and timing of clinical outcomes at 4 years, with further subdivision by age.
Supplemental Figure 14. Absolute risk projections for KRT at 2-years for a set of scenarios for a 35-year old man (A) or woman (B).
Supplemental Figure 15. Absolute risk projections for KRT at 2-years for a set of scenarios for a 45-year old man (A) or woman (B).
Supplemental Figure 16. Absolute risk projections for KRT at 2-years for a set of scenarios for a 55-year old man (A) or woman (B).
Supplemental Figure 17. Absolute risk projections for KRT at 2-years for a set of scenarios for a 65-year old man (A) or woman (B).
Supplemental Figure 18. Absolute risk projections for KRT at 2-years for a set of scenarios for a 75-year old man (A) or woman (B).
Supplemental Figure 19. Absolute risk projections for KRT at 2-years for a set of scenarios for an 85-year old man (A) or woman (B).
Supplemental Figure 20. Calibration of the predicted versus observed 2-year risk of KRT for the (A) kidney failure risk equation (KFRE) and (B) MCMC model
Supplemental Figure 21. Prevalence of 2-year predicted KRT risk for the (A) kidney failure risk equation (KFRE) and (B) MCMC model.
Supplemental Figure 22. Distribution of 4-year event free survival
Supplemental Figure 23. Equal-weighted mean of the adjusted cumulative incidence curves for first KRT (A), first CVD (B), and death as a first event (C) within the three types of cohort (CKD research, administrative, and referred CKD).
Supplemental Figure 24. The probability and timing of clinical outcomes at 4 years overall and by type of cohort.
Supplemental Figure 25. The probability and timing of clinical outcomes at 4 years overall and by type of cohort.
Supplemental Figure 26. The probability and timing of clinical outcomes at 4 years overall and by type of cohort.
Supplemental Figure 27. Cumulative incidence of first KRT (A), first CVD (B), and death as a first event (C) by region.
Supplemental Figure 28. Adjusted absolute risk of KRT by cause of CKD in cohorts with available data (Supplemental Table 2).
Supplemental Figure 29. Diagram of states and transitions included in the 3-state Markov model.
Supplemental Figure 30. Absolute KRT (A) or death (B) risk projections at 2-years for a set of scenarios for a 60-year old man from a 5-state Markov model compared to a 3-state Markov model.