Development and validation of a chronic kidney disease progression model using patient-level simulations

Abstract

Chronic disease progression models are available for several highly prevalent conditions. For chronic kidney disease (CKD), the scope of existing progression models is limited to the risk of kidney failure and major cardiovascular (CV) events. The aim of this project was to develop a comprehensive CKD progression model (CKD-PM) that simulates the risk of CKD progression and a broad range of complications in patients with CKD. A series of literature reviews informed the selection of risk factors and identified existing risk equations/algorithms for kidney replacement therapy (KRT), CV events, other CKD-related complications, and mortality. Risk equations and transition probabilities were primarily sourced from publications produced by large US and international CKD registries. A patient-level, state-transition model was developed with health states defined by the Kidney Disease Improving Global Outcomes categories. Model validation was performed by comparing predicted outcomes with observed outcomes in the source cohorts used in model development (internal validation) and other cohorts (external validation). The CKD-PM demonstrated satisfactory modeling properties. Accurate prediction of all-cause and CV mortality was achieved without calibration, while prediction of CV events through CKD-specific equations required implementation of a calibration factor to balance time-dependent versus baseline risk. Predicted annual changes in estimated glomerular filtration rate (eGFR) and urine albumin-creatinine ratio were acceptable in comparison to external values. A flexible eGFR threshold for KRT equations enabled accurate prediction of these events. This CKD-PM demonstrated reliable modeling properties. Both internal and external validation revealed robust outcomes.

Keywords: Chronic kidney disease; disease progression model; microsimulation; validation.

Figure 1.

CKD progression model diagram. BMD: bone and mineral disorders; BMI: body mass index; CKD: chronic kidney disease; CVD: cardiovascular disease; CVM: cardiovascular mortality; eGFR: estimated glomerular filtration rate; ESKD: end-stage kidney disease; HbA1c, glycated hemoglobin; HS: health state; HTN: hypertension; KDIGO: kidney disease improving global outcomes; KRT: kidney with replacement therapy; uACR: urine albumin-creatinine ratio.

Figure 2.

Data sources used for model validation, categorized according to endpoint. ASCVD: atherosclerotic cardiovascular disease; CV: cardiovascular; CVD: cardiovascular disease; eGFR: estimated glomerular filtration rate; ESKD: end-stage kidney disease; HF: heart failure; KRT: kidney replacement therapy; uACR: urine albumin-creatinine ratio.

Figure 3.

External validation of all-cause mortality (composite of cardiovascular, renal, and nonspecific death): predicted vs. observed rate. 1 ARIC, 2 AusDiab, 3 CHS, 4 COBRA, 5 Framingham, 6 Gubbio, 7 HUNT, 8 MESA, 9 NHANES III, 10 PREVEND, 11 Rancho Bernardo, 12 REGARDS, 13 ULSAM (source Matsushita, et al. 2010). Mid-points G1 90, G2 75, G3a 52.5, G3b 37.5, G4 22.5, G5 7.5 ml/min/1.73 m², A1 15, A2 165, A3 1500 mg/g. ACM: all-cause mortality; CVM: cardiovascular mortality; KRT: kidney replacement therapy. Baseline characteristics used to populate the model and generate the predictions by cohort are taken from Matsushita, et al., 2010 [41].

Figure 4.

Internal (A) and external (B) validation of cardiovascular mortality modeled using CKD-Patch: predicted vs. observed rate [43]. Panel A: 1 NHANEScon, 2 China NS, 3 COBRA, 4 Framingham, 5 Takahata, 6 AusDiab, 7 MESA, 8 BIS, 9 PREVEND, 10 Gubbio, 11 ARIC, 12 Rancho Bernardo, 13 NHANESIII, 14 ULSAM (source Matsushita, et al. 2020). CVM: cardiovascular mortality; CKD: chronic kidney disease. Panel B: 1 UK Biobank, 2 GCKD, 3 REGARDS, 4 Sunnybrook, 5 ZODIAC, 6 SMART, 7 CARDIA, 8 SEED, 9 ADVANCE CVM: cardiovascular mortality; CKD: chronic kidney disease. Baseline characteristics used to populate the model and generate the predictions by cohort are taken from Matsushita, et al., 2020 [43].

Figure 5.

Internal (A) and external (B) validation of atherosclerotic CVD modeled using CKD-Patch: predicted vs. observed rate. Panel A: 1 OLDW cohort 3, 2 OLDW cohort 10, 3 Mt Sinai BioMe, 4 Maccabi, 5 AusDiab, 6 Geisinger, 7 PREVEND, 8 Framingham, 9 Tromso, 10 MESA, 11 ARIC, 12 Rancho Bernardo, 13 BIS, 14 ULSAM (source Matsushita 2020, development dataset) bsl, baseline; CKD, chronic kidney disease; CVD, cardiovascular disease. Panel B: 1 OLDW cohort 27, 2 CARDIA, 3 OLDW cohort 17, 4 JHS, 5 UK Biobank, 6 GCKD, 7 NEFRONA, 8 CRIC, 9 SMART, 10 SEED, 11 RENAAL, 12 REGARDS, 13 LCC, 14 ADVANCE (source Matsushita 2020, validation dataset) ASCVD, atherosclerotic cardiovascular disease; bsl, baseline; CVM, cardiovascular mortality; CKD, chronic kidney disease; CVD, cardiovascular disease. Baseline characteristics used to populate the model and generate the predictions by cohort are taken from Matsushita, et al., 2020 [43].

Figure 6.

External validation of kidney replacement therapy event rate, using North American (A) and non-North American (B) cohorts: predicted vs. observed rate. Panel A: 1 AASK, 2 ARIC, 3 BCCKD, 4 CCFACR, 5 CCFDIP, 6 CRIC, 7 Geisinger, 8 ICES-KDT, 9 KEEP, 10 KPNW, 11 MDRD, 12 Mt Sinai BioMe, 13 REGARDS, 14 Sunnybrook, 15 VACKD (source Tangri, et al., 2016, North American cohorts). Panel B: 1 CRIB, 2 GCKD, 3 GLOMMS-1, 4 Gonryo, 5 HUNT, 6 Maccabi, 7 MASTERPLAN, 8 MMKD, 9 NephroTest, 10 NZDCS, 11 Okinawa83, 12 Okinawa93, 13 RENAAL, 14 Severance, 15 SRRCKD (source Tangri, et al., 2016, Non-North America cohorts) Baseline characteristics used to populate the model and generate the predictions by cohort are taken from Tangri, et al., 2016 [57]. Pima cohort was identified as an outlier (kidney replacement therapy rate 168.3 per 1000-person-years) and excluded from the set of North American cohorts.

Figure 7.

Model-predicted examples of patient-level progression to kidney failure, starting from health states G1-2 (A), G3a (B), G3b (C), and G4-5 (D). Panel A: These 6 patients do not reach the stage of KRT within 15 years. Based on the average eGFR slope, we can identify ‘slow progressors’ (patients a, b) and ‘fast progressors’ (patients e, f). Patients c and d are not considered ‘fast progressors’; nevertheless, patient d worsens from G2 category at baseline to G4 by year 15. All patients evolve to macroalbuminuria (A3) by year 15. eGFR, estimated glomerular filtration rate; T2D, type 2 diabetes; uACR, urine albumin-creatinine ratio. Panel B: Slow progressor: patient g. Fast progressors, not receiving KRT by year 15: patient i (death in year 11), and patient k (death in year 12). Fast progressor, requiring KRT: patient h (year 14). Not considered a ‘fast progressor’ but requiring KRT: patient l (year 11). Not considered a ‘slow progressor’ given accelerated worsening in last years: patient j. eGFR, estimated glomerular filtration rate; T2D, type 2 diabetes; uACR, urine albumin-creatinine ratio. Panel C: Patient m: fast progressor, dialysis in year 10, death in year 11. Patient o: dialysis in year 7, successful KRT in year 8. Patient q: starts dialysis in year 7, successful KRT in year 14. Patient n: fast progressor, death in year 8.Patient p: slow progressor, no KRT by year 15. Patient r: starts dialysis in year 9, death in year 10.eGFR, estimated glomerular filtration rate; KRT, kidney replacement therapy; T2D, type 2 diabetes; uACR, urine albumin-creatinine ratio. Panel D: Patient s: fast progressor, successful KRT in year 5, death in year 12. Patient u: starts dialysis in year 7, until death in year 13. Patient w: slow progressor, no KRT, death in year 9. Patient t: starts dialysis in year 3, until death in year 11. Patient v: fast progressor, starts dialysis in year 3, KRT in year 13. Patient x: declines to G5 within 5 years, no KRT, death in year 8. eGFR, estimated glomerular filtration rate; KRT, kidney replacement therapy; T2D, type 2 diabetes; uACR, urine albumin-creatinine ratio. Baseline characteristics used to populate the model and generate the predictions stratified by estimated glomerular filtration rate group are taken from Grams, et al., 2021 [47].

Figure 7.

Model-predicted examples of patient-level progression to kidney failure, starting from health states G1-2 (A), G3a (B), G3b (C), and G4-5 (D). Panel A: These 6 patients do not reach the stage of KRT within 15 years. Based on the average eGFR slope, we can identify ‘slow progressors’ (patients a, b) and ‘fast progressors’ (patients e, f). Patients c and d are not considered ‘fast progressors’; nevertheless, patient d worsens from G2 category at baseline to G4 by year 15. All patients evolve to macroalbuminuria (A3) by year 15. eGFR, estimated glomerular filtration rate; T2D, type 2 diabetes; uACR, urine albumin-creatinine ratio. Panel B: Slow progressor: patient g. Fast progressors, not receiving KRT by year 15: patient i (death in year 11), and patient k (death in year 12). Fast progressor, requiring KRT: patient h (year 14). Not considered a ‘fast progressor’ but requiring KRT: patient l (year 11). Not considered a ‘slow progressor’ given accelerated worsening in last years: patient j. eGFR, estimated glomerular filtration rate; T2D, type 2 diabetes; uACR, urine albumin-creatinine ratio. Panel C: Patient m: fast progressor, dialysis in year 10, death in year 11. Patient o: dialysis in year 7, successful KRT in year 8. Patient q: starts dialysis in year 7, successful KRT in year 14. Patient n: fast progressor, death in year 8.Patient p: slow progressor, no KRT by year 15. Patient r: starts dialysis in year 9, death in year 10.eGFR, estimated glomerular filtration rate; KRT, kidney replacement therapy; T2D, type 2 diabetes; uACR, urine albumin-creatinine ratio. Panel D: Patient s: fast progressor, successful KRT in year 5, death in year 12. Patient u: starts dialysis in year 7, until death in year 13. Patient w: slow progressor, no KRT, death in year 9. Patient t: starts dialysis in year 3, until death in year 11. Patient v: fast progressor, starts dialysis in year 3, KRT in year 13. Patient x: declines to G5 within 5 years, no KRT, death in year 8. eGFR, estimated glomerular filtration rate; KRT, kidney replacement therapy; T2D, type 2 diabetes; uACR, urine albumin-creatinine ratio. Baseline characteristics used to populate the model and generate the predictions stratified by estimated glomerular filtration rate group are taken from Grams, et al., 2021 [47].

Figure 7.

Model-predicted examples of patient-level progression to kidney failure, starting from health states G1-2 (A), G3a (B), G3b (C), and G4-5 (D). Panel A: These 6 patients do not reach the stage of KRT within 15 years. Based on the average eGFR slope, we can identify ‘slow progressors’ (patients a, b) and ‘fast progressors’ (patients e, f). Patients c and d are not considered ‘fast progressors’; nevertheless, patient d worsens from G2 category at baseline to G4 by year 15. All patients evolve to macroalbuminuria (A3) by year 15. eGFR, estimated glomerular filtration rate; T2D, type 2 diabetes; uACR, urine albumin-creatinine ratio. Panel B: Slow progressor: patient g. Fast progressors, not receiving KRT by year 15: patient i (death in year 11), and patient k (death in year 12). Fast progressor, requiring KRT: patient h (year 14). Not considered a ‘fast progressor’ but requiring KRT: patient l (year 11). Not considered a ‘slow progressor’ given accelerated worsening in last years: patient j. eGFR, estimated glomerular filtration rate; T2D, type 2 diabetes; uACR, urine albumin-creatinine ratio. Panel C: Patient m: fast progressor, dialysis in year 10, death in year 11. Patient o: dialysis in year 7, successful KRT in year 8. Patient q: starts dialysis in year 7, successful KRT in year 14. Patient n: fast progressor, death in year 8.Patient p: slow progressor, no KRT by year 15. Patient r: starts dialysis in year 9, death in year 10.eGFR, estimated glomerular filtration rate; KRT, kidney replacement therapy; T2D, type 2 diabetes; uACR, urine albumin-creatinine ratio. Panel D: Patient s: fast progressor, successful KRT in year 5, death in year 12. Patient u: starts dialysis in year 7, until death in year 13. Patient w: slow progressor, no KRT, death in year 9. Patient t: starts dialysis in year 3, until death in year 11. Patient v: fast progressor, starts dialysis in year 3, KRT in year 13. Patient x: declines to G5 within 5 years, no KRT, death in year 8. eGFR, estimated glomerular filtration rate; KRT, kidney replacement therapy; T2D, type 2 diabetes; uACR, urine albumin-creatinine ratio. Baseline characteristics used to populate the model and generate the predictions stratified by estimated glomerular filtration rate group are taken from Grams, et al., 2021 [47].

Figure 7.

Model-predicted examples of patient-level progression to kidney failure, starting from health states G1-2 (A), G3a (B), G3b (C), and G4-5 (D). Panel A: These 6 patients do not reach the stage of KRT within 15 years. Based on the average eGFR slope, we can identify ‘slow progressors’ (patients a, b) and ‘fast progressors’ (patients e, f). Patients c and d are not considered ‘fast progressors’; nevertheless, patient d worsens from G2 category at baseline to G4 by year 15. All patients evolve to macroalbuminuria (A3) by year 15. eGFR, estimated glomerular filtration rate; T2D, type 2 diabetes; uACR, urine albumin-creatinine ratio. Panel B: Slow progressor: patient g. Fast progressors, not receiving KRT by year 15: patient i (death in year 11), and patient k (death in year 12). Fast progressor, requiring KRT: patient h (year 14). Not considered a ‘fast progressor’ but requiring KRT: patient l (year 11). Not considered a ‘slow progressor’ given accelerated worsening in last years: patient j. eGFR, estimated glomerular filtration rate; T2D, type 2 diabetes; uACR, urine albumin-creatinine ratio. Panel C: Patient m: fast progressor, dialysis in year 10, death in year 11. Patient o: dialysis in year 7, successful KRT in year 8. Patient q: starts dialysis in year 7, successful KRT in year 14. Patient n: fast progressor, death in year 8.Patient p: slow progressor, no KRT by year 15. Patient r: starts dialysis in year 9, death in year 10.eGFR, estimated glomerular filtration rate; KRT, kidney replacement therapy; T2D, type 2 diabetes; uACR, urine albumin-creatinine ratio. Panel D: Patient s: fast progressor, successful KRT in year 5, death in year 12. Patient u: starts dialysis in year 7, until death in year 13. Patient w: slow progressor, no KRT, death in year 9. Patient t: starts dialysis in year 3, until death in year 11. Patient v: fast progressor, starts dialysis in year 3, KRT in year 13. Patient x: declines to G5 within 5 years, no KRT, death in year 8. eGFR, estimated glomerular filtration rate; KRT, kidney replacement therapy; T2D, type 2 diabetes; uACR, urine albumin-creatinine ratio. Baseline characteristics used to populate the model and generate the predictions stratified by estimated glomerular filtration rate group are taken from Grams, et al., 2021 [47].