Association of UACR and UPCR with kidney failure: analysis of observational data in patients with rare kidney diseases Authors: Katie Wong, David Pitcher, Jonathan Barratt, Daniel P Gale, RaDaR Consortium Published: 2026 ABSTRACT ABSTRACTBackgroundProteinuria is associated with long-term kidney failure (KF) risk. In trials of CKD and diabetic kidney disease, urine albumin:creatinine ratio (ACR) has been the preferred urinary protein measurement and features in guidelines and risk-prediction models. In studies of rare, glomerular diseases, urine protein: creatinine ratio (PCR) has more often been used, consequently featuring in treatment algorithms for these diseases. While equations estimating ACR from PCR (estACR) have been developed in large CKD cohorts, whether ACR, PCR, non-albumin proteinuria, and estACR differ in their relationships with KF in patients with rare kidney diseases is unknown.MethodsPatients in RaDaR with same-day ACR and PCR measurements were included. Analyses were undertaken for the whole cohort, and by disease subtype and ACR tertile. Calibration plots assessed the performance of converting PCR to ACR. Cox regression models of time to KF or death (adjusting for age, sex, eGFR, disease) were run with four proteinuria measures: ACR, PCR, non-albumin proteinuria and estACR. Model fit was assessed using AIC (Akaike information criterion) and SBC (Schwarz Bayesian information Criterion) values, ROC curves and time-dependent AUC, and concordance compared using Uno’s C statistic.ResultsOut of 4156 patients, 717 (17%) experienced a KF or death event. At lower predicted ACR deciles, conversion of PCR to ACR underestimated true ACR. Very similar AIC and SBC statistics were observed for all proteinuria measurements, including estACR. Time-dependent AUC at 5 years were >0.83 for all models and disease sub-groups. No statistically significant differences in model concordance were seen (Chi2 P > .2 for all comparisons). Within each ACR tertile, AIC, SBC, and AUC values were similar across all proteinuria metrics.ConclusionsIn this cohort of patients with rare kidney diseases, across all levels of proteinuria studied, we identified no statistically significant differences in the performance of ACR, PCR, non-albumin proteinuria, or estACR to predict KF risk.Keywords: glomerular, proteinuria, rare kidney diseases What was known: Urine albumin:creatinine ratio (ACR) has been the preferred urinary protein measurement in clinical guidelines and trials in CKD. Urine albumin:creatinine ratio (ACR) has been the preferred urinary protein measurement in clinical guidelines and trials in CKD. However, for rare glomerular kidney diseases, some of which are characterized by very high levels of proteinuria, urine protein:creatinine ratio (PCR) has more often been used. However, for rare glomerular kidney diseases, some of which are characterized by very high levels of proteinuria, urine protein:creatinine ratio (PCR) has more often been used. This study adds: Whether ACR, PCR, non-albumin proteinuria and ACR estimated from PCR (estACR) are equivalent in predicting kidney failure (KF) events in patients with rare kidney disease is unclear. Whether ACR, PCR, non-albumin proteinuria and ACR estimated from PCR (estACR) are equivalent in predicting kidney failure (KF) events in patients with rare kidney disease is unclear. In this study we identified no statistically significant differences in the performance of ACR, PCR, non-albumin proteinuria, or estACR to predict KF risk, across all disease subtypes and proteinuria levels, in our rare kidney disease cohort. In this study we identified no statistically significant differences in the performance of ACR, PCR, non-albumin proteinuria, or estACR to predict KF risk, across all disease subtypes and proteinuria levels, in our rare kidney disease cohort. Potential impact: These results suggest that estimated ACR can be used interchangeably with measured ACR for patients with rare kidney diseases clinically for risk prediction, and in Cox regression models of time to KF for observational studies or clinical trials across a range of proteinuria levels These results suggest that estimated ACR can be used interchangeably with measured ACR for patients with rare kidney diseases clinically for risk prediction, and in Cox regression models of time to KF for observational studies or clinical trials across a range of proteinuria levels INTRODUCTION Proteinuria has been shown to be an important predictor of risk of chronic kidney disease (CKD) progression and cardiovascular events in unselected cohorts [1, 2]. Urine albumin:creatinine ratio (ACR) has been the preferred urinary protein measurement in clinical guidelines [3], risk-prediction models [4, 5] and therapeutic clinical trials in CKD [6–9], owing to its higher sensitivity and better precision at lower levels of proteinuria, and increased sensitivity in identifying early diabetic kidney disease, a significant contributor to CKD globally [10]. However, many of the studies that have contributed to the decision to use ACR in preference to urine protein:creatinine ratio (PCR) have been conducted in large, population-based CKD cohorts, made up predominantly of patients with diabetic or hypertensive kidney disease. Patients with rare kidney diseases have higher rates of kidney failure (KF) and lower mortality rates compared to patients with general CKD, [11] and individual rare kidney diseases have distinct aetiologies that are often characterized by very high levels of proteinuria, which may make data from previous population-based CKD studies less transferable. In recent years, the US Food and Drug Administration has allowed applications for accelerated approval for rare disease therapies based on the demonstration in clinical trials of short-term reduction of proteinuria as a surrogate end-point reasonably likely to predict true clinical benefit in terms of progression to KF, with full approval granted following confirmatory data showing slowing of eGFR decline over 24 months. Examples of therapies approved by this pathway include sparsentan [12] and atrasentan [13, 14] for IgA nephropathy. In C3 glomerulopathy, both iptacopan and pegcetacoplan recently received full approval following trials that had a proteinuria primary end-point (measured using PCR) at 6 months [15, 16]. However, whether ACR, PCR, or non-albumin proteinuria are equivalent in predicting KF events in patients with rare kidney disease is unclear. In addition, in real-world practice total protein or PCR is often collected, and in retrospective cohort studies only PCR may be available. An ACR to PCR approximate ratio of 0.7 has previously been cited, however, this is complicated by a nonlinear association between the two measurements; at higher levels of proteinuria ~70% of total protein is albumin, whereas at lower levels of proteinuria this can be as low as 20%. The strength of the correlation between ACR and PCR therefore varies at different levels of proteinuria, being strongest at higher levels. Additional variation may also come from patient factors such as sex, kidney function, and ethnicity. An equation has been developed in a large population-based Canadian cohort to estimate ACR from PCR [17], accounting for non-linearity, and patient factors such as age, sex, and presence of hypertension and diabetes. There has been conflicting evidence around how this conversion equation performs in kidney transplant patients, who may have higher frequency of tubular proteinuria [18, 19]. Certain individual rare kidney diseases, such as cystinosis, have a clear tubular pathology leading to predominantly tubular proteinuria, whereas for other disorders such as autosomal dominant polycystic kidney disease (ADPKD), cysts may disrupt both glomerular and tubular architecture leading to a different proteinuria profile compared with more common causes of CKD. Therefore, how accurately this equation performs in patients with rare kidney diseases is uncertain and has not previously been well studied. Furthermore, whether estimated ACR (from PCR) predicts KF with the same accuracy as measured ACR in a rare kidney disease cohort is unclear; previous studies in general CKD cohorts have shown excellent performance in predicting kidney failure for both measured and estimated values [20, 21], however this varied significantly depending on level of proteinuria. We have therefore used same-day paired ACR and PCR data for 4156 patients from the UK National Registry of Rare Kidney Diseases (RaDaR) to address three questions: Do ACR to PCR conversion equations, developed in a large cohorts of patients with unselected CKD differ in their association with KF in a rare kidney disease cohort? Do ACR to PCR conversion equations, developed in a large cohorts of patients with unselected CKD differ in their association with KF in a rare kidney disease cohort? Are ACR and PCR equally predictive of KF in rare kidney diseases, and does measurement of non-albumin proteinuria provide any additional discrimination? Are ACR and PCR equally predictive of KF in rare kidney diseases, and does measurement of non-albumin proteinuria provide any additional discrimination? Are the predictive performances of ACR and estimated ACR (converted from PCR) comparable in this cohort? Are the predictive performances of ACR and estimated ACR (converted from PCR) comparable in this cohort? MATERIALS AND METHODS Data source and linkage The UK National Registry of Rare Kidney Diseases (RaDaR) was established in 2010 to collect longitudinal prospective and retrospective laboratory and clinical data for individuals with >30 rare kidney disease groups, and currently recruits from 108 UK National Health Service (NHS) sites. It receives routine blood and urine test result data via data linkage with the UK Renal Data Collaboration, and validated data on initiation of kidney replacement therapy (KRT) and death via the UK Renal Registry, including data provided by NHS Blood and Transplant. Patients provide written informed consent for analysis and linkage of their data at recruitment. Recruitment characteristics and potential biases have been reported previously [22]. Study population All patients recruited to RaDaR with ACR and PCR measurements on the same day were eligible. Data after KRT initiation (dialysis or kidney transplantation) were excluded. To minimize the effect of initial disease treatment after diagnosis on the association between proteinuria value and kidney outcomes, data within 6 months of diagnosis date, where available, were also excluded. For patients with several pairs of ACR–PCR measurements available, one pair was chosen at random. All analyses have been repeated using the pair of ACR–PCR measurements with the median ACR:PCR ratio for each patient, and are presented as sensitivity analyses in the Supplementary Materials. Distribution of all ACR results in RaDaR, comparing results with and without paired PCR results available are presented in Supplementary Fig. S1. CKD stage was calculated from serum creatinine values within 90 days of concurrent proteinuria measurement data, using the CKD Epidemiology Collaboration equation 2021 without adjustment for race (CKD-EPI 2021). Data were analysed for the whole RaDaR cohort, and by rare kidney disease category: cystic (ADPKD n = 847, autosomal dominant tubulointerstitial kidney disease/nephronophthisis n = 22, autosomal recessive polycystic kidney disease n = 15, HNF1b n = 5), glomerular (Alport syndrome n = 157, idiopathic nephrotic syndrome n = 723, IgA nephropathy n = 721, lupus nephritis n = 41, monoclonal gammopathy of renal significance n = 24, C3 glomerulopathy/immune complex membranoproliferative glomerulonephritis n = 149, membranous nephropathy n = 525, vasculitis n = 615, atypical haemolytic uraemic syndrome n = 9, Fabry n = 7, Shiga toxin-producing E. coli HUS n = 2), tubular (tubulopathies n = 74), metabolic (cystinosis n = 6, cystinuria n = 17, hyperoxaluria n = 5), and other (n = 192). Statistical analyses Calibration plots were used to assess the performance of conversion of PCR to ACR across the range of ACR. An approximate factor of ACR = PCR × 0.7 and the Weaver et al. model [23] using eGFR, age, and sex to predict ACR were both assessed and compared to observed ACR. Survival analyses assessing kidney survival from the date of concurrent proteinuria measurement to KRT initiation or death were used to examine the association of ACR and PCR with long-term kidney outcomes. Patients were stratified into tertiles based on ACR and also PCR, and Kaplan–Meier (KM) survival curves overlaid on top of each other. Owing to the strong associations between CKD stage and proteinuria, and also CKD stage and kidney survival, proteinuria tertiles were calculated within each CKD stage. In addition, to assess whether non-albumin PCR confers any additional association beyond ACR alone, ACR:PCR ratio was calculated to account for potential correlation between ACR and non-albumin PCR. Then within each CKD stage and ACR tertile, patients were further stratified based on ACR:PCR ratio tertile and time to KF estimated using KM methods. Sensitivity analyses were performed using a competing-risks regression based on Fine and Gray’s proportional subdistribution hazards model, with KF as the outcome of interest and death considered a competing risk. Cumulative incidence functions were derived from the fitted model and plotted by ACR and PCR tertile. To further assess whether ACR or PCR has a stronger association with kidney outcomes, Cox regression models adjusting for age, sex, and eGFR were run with four different methods of quantifying proteinuria; ACR, PCR, ACR and non-albumin PCR, and ACR estimated from observed PCR (estACR) using the Weaver et al. model. To assess model fit AIC (Akaike information criterion) and SBC (Schwarz Bayesian information criterion) values were compared, receiver operator curves and time-dependent AUC [24] at 5 years were estimated, and Uno’s C statistic [25] was used to compare concordance of each quantification method across all of follow up with 100 perturbations used to estimate 95% CIs. To assess whether the performance of the models differed by level of proteinuria, model fit and concordance statistics were also calculated stratified by ACR tertile. Finally, each patient had their 5-year KFRE risk score estimated using observed ACR and also estACR (using the Weaver model). Brier scores were calculated at 5 years for each method. Analyses are presented for the whole RaDaR cohort, and for cystic and glomerular sub-groups; stratification by tubular and other sub-groups have not been presented due to small numbers. Ethics RaDaR has ethical approval as a research registry provided by NHS South-West Central Bristol Research ethics committee for England (24/SW/0135 IRAS: 349779) and the Scotland A ethics committee for Scotland (24/SS/0092 IRAS: 351286). This study adheres to the Strengthening the Reporting of Observation Studies in Epidemiology statement. RESULTS In total, 4156 patients were included in the analyses, 45% female, median (IQR) age 53 years (39–66). n = 717 (17%) experienced a KF or death event. n = 889 (21%) patients had cystic, n = 2973 (72%) glomerular n = 74 (2%) tubular, n = 28 (1%) metabolic, and n = 192 (5%) other kidney disease (Table 1). Table 1:Clinical demographics of included patients.All RaDaRCysticGlomerularTubularMetabolicOtherTotal (n)4 1568892 9737428192KF, KRT initiation, or death (n, %)717 (17.3)183 (20.6)510 (17.2)1 (1.4)1 (3.6)22 (11.5)Deaths (n, %)200 (4.8)19 (2.1)176 (5.9)1 (1.4)0 (0.0)4 (2.1)Deaths without KF (n, %)120 (2.9)8 (0.9)110 (3.7)0 (0.0)0 (0.0)2 (1.0)Age, years (median, IQR)53 (39–66)48 (38–58)56 (41–69)38 (28–54)40 (24–65)44 (33–58)Sex, female (n, %)1 883 (45.3)481 (54.1)1226 (41.2)39 (52.7)13 (46.4)124 (64.6)ACR (mg/mmol, median, IQR)-5 (2–12)33 (6–123)6 (1–25)12 (2–51)24 (3–84)PCR (mg/mmol, Median, IQR)-17 (10–32)57 (18–182)24 (11–55)30 (14–139)44 (13–124)CKD stage (n, %) 1982 (23.6)208 (23.4)681 (22.9)36 (48.7)6 (21.4)51 (26.6) 2746 (18.0)154 (17.3)542 (18.2)15 (20.3)7 (25.0)28 (14.6) 3a894 (21.5)173 (19.5)665 (22.4)8 (10.8)7 (25.0)41 (21.4) 3b806 (19.4)152 (17.1)605 (20.4)9 (12.2)4 (14.3)36 (18.8) 4684 (16.5)195 (21.9)446 (15.0)6 (8.1)4 (14.3)33 (17.2) 544 (1.1)7 (0.8)34 (1.1)0 (0.0)0 (0.0)3 (1.6) Clinical demographics of included patients. Estimating ACR from PCR Calibration plots of (i) measured ACR; and ACR predicted from PCR using (ii) the approximate ratio of 0.7, and (iii) Weaver’s equation, across a range of ACR values for all RaDaR patients and on a log scale are presented in Fig. 1a, with predicted ACR divided into deciles. ACR values calculated from PCR using the conversion ratio of 0.7 were higher than measured values of ACR at lower levels of proteinuria. Values of ACR estimated from PCR using the Weaver equation [23] (estACR) conversely underestimated ACR at lower predicted ACR deciles (decile 1: mean measured ACR 1 mg/mmol, estACR 1 mg/mmol, decile 2: mean measured ACR 2 mg/mmol, estACR 1 mg/mmol, decile 3: mean measured ACR 7 mg/mmol, estACR 5 mg/mmol, decile 4: mean measured ACR 13 mg/mmol, estACR 11 mg/mmol). Stratifying by disease group showed a similar pattern of good agreement between the methods for ACR values over 100 mg/mmol, and over-estimation using the 0.7 conversion for ACR values below this. Calibration plot of measured ACR, est(ACR), and approximated ACR = PCR × 0.7, for (a) all patients, (b) cystic, and (c) glomerular kidney disease. Time to event analyses To investigate the strength of association between ACR, PCR, and non-albumin PCR and time to KF or death from concurrent ACR–PCR result, KM analyses, and Cox regression models were performed using each proteinuria measurement. Paired proteinuria results were divided into tertiles (tertile 1 = lowest, tertile 3 = highest). Thresholds for each tertile are presented in Table 2. Patients in the highest ACR tertile (tertile 3) predominantly had primary glomerular pathology (Supplementary Table S1, Supplementary Fig. S2): CKD stage 1 n = 302 (92.4% glomerular disease), stage 2 n = 228 (91.6%), stage 3a n = 270 (90.6%), stage 3b n = 253 (94.1%), stage 4 n = 200 (87.7%), and stage 5 n = 13 (92.3%). Table 2:ACR and PCR tertile thresholds, stratified by CKD stage.ACR tertile thresholds (mg/mmol)PCR tertile thresholds (mg/mmol)CKD stage123123G1<2.32.3–22.0>22.0<12.012.0–40.0>40.0G2<3.03.0–25.7>25.7<12.012.0–48.0>48.0G3a<7.27.2–49.0>49.0<20.020.0–79.0>79.0G3b<10.610.6–73.5>73.5<28.028.0–115.7>115.7G4<22.722.7–112.0>112.0<50.050.0–175.0>175.0G5<43.543.5–171.1>171.1<93.093.0–306.0>306.0 ACR and PCR tertile thresholds, stratified by CKD stage. KM analyses of time to KF or death from paired ACR–PCR date, stratified by ACR and PCR tertile for each CKD stage are presented in Fig. 2, showing no difference in observed outcome when using either ACR or PCR stratification. Similar results were seen when stratifying by rare kidney disease category (Supplementary Figs. S3 and S4, Supplementary Table S2), and when using a competing-risks regression considering death as a competing risk (Supplementary Fig. S5). Time to KF or death (years) from date of concurrent results, stratified by ACR or PCR tertile. Tertile ranges are presented in Table 2. To investigate whether non-albumin proteinuria provides any additional discriminatory information, within each CKD stage and ACR tertile, patients were further stratified by ACR/PCR ratio tertile (tertile 1 = lowest, tertile 3 = highest), with no statistically significant differences in time to KF or death observed (Supplementary Fig. S6, Supplementary Table S3). Adjusted hazards ratios for time to KF or death from Cox regression models using either ACR, PCR, or non-albumin proteinuria measurements for the whole RaDaR cohort were very similar (Table 3). Assessments of accuracy of model fit for each proteinuria measurement showed little difference between models (Table 3): log(ACR) AIC 9440.34, log(PCR) AIC 9423.73, log(ACR and non-albumin proteinuria) AIC 9427.32, and log(ACR) SBC 9586.74, log(PCR) SBC 9570.13, and log(ACR and non-albumin proteinuria) SBC 9578.29. For both AIC and SBC, the model using log(PCR) had marginally better fit. Examining patients with cystic and glomerular diseases separately, results were again very similar across models for both groups, however for patients with cystic diseases the model using log(ACR) had best fit (AIC 1680.40, SBC 1702.87) and for glomerular diseases the model using log(PCR) again had best fit (AIC 6479.89, SBC 6539.17). Table 3:Fit and concordance statistics for Cox models of time from concurrent ACR–PCR values to KF or Death, all adjusting for age, sex, eGFR and rare disease group, by proteinuria measurement used.FitModelUnitAdjusted HRa (95% CI)AICSBCAll RaDaRlog(ACR)log(ACR) = 11.33 (1.26–1.41)9 440.349 586.74log(PCR)log(PCR) = 11.48 (1.38–1.58)9 423.739 570.13log(ACR) and non-albumin proteinurialog(ACR) = 11.19 (1.10–1.29)9 427.329 578.29log(ACR) and non-albumin proteinurialog(non-albumin proteinuria) = 11.24 (1.11–1.37)9 427.329 578.29log(estACR)log(estACR) = 11.30 (1.24–1.37)9 436.259 582.65Glomerular diseaselog(ACR)log(ACR) = 11.35 (1.26–1.44)6 502.076 561.35log(PCR)log(PCR) = 11.51 (1.39–1.63)6 479.896 539.17log(ACR) and non-albumin proteinurialog(ACR) = 11.16 (1.06–1.28)6 487.966 551.48log(ACR) and non-albumin proteinurialog(non-albumin proteinuria) = 11.29 (1.14–1.46)6 487.966 551.48log(estACR)log(estACR) = 11.33 (1.25–1.42)6 491.636 550.92Cystic diseaselog(ACR)log(ACR) = 11.33 (1.19–1.50)1 680.401 702.87log(PCR)log(PCR) = 11.38 (1.18–1.61)1 688.341 710.80log(ACR) and non-albumin proteinurialog(ACR) = 11.36 (1.15–1.59)1 682.311 707.98log(ACR) and non-albumin proteinurialog(non-albumin proteinuria) = 10.96 (0.74–1.24)1 682.311 707.98log(estACR)log(estACR) = 11.24 (1.12–1.37)1 686.551 709.01ConcordanceModelComparator modelUno’s concordance statisticStandard errorChi-square P valueAll RaDaRlog(PCR)log(ACR)−0.00100.0020.6263log(PCR)log(ACR) and non-albumin proteinuria−0.00100.0017.5573log(ACR)log(ACR) and non-albumin proteinuria0.00000.0013.9696log(ACR)log(estACR)0.00210.0026.4161log(ACR) and non-albumin proteinurialog(estACR)0.00210.0023.3573log(PCR)log(estACR)0.00110.0019.5637Glomerular diseaselog(PCR)log(ACR)−0.00050.0024.8289log(PCR)log(ACR) and non-albumin proteinuria−0.00060.0026.8243log(ACR)log(ACR) and non-albumin proteinuria−0.00010.00330.9818log(ACR)log(estACR)0.00360.00370.3235log(ACR) and non-albumin proteinurialog(estACR)0.00370.00400.3584log(PCR)log(estACR)0.00310.00340.3566Cystic diseaselog(PCR)log(ACR)−0.00260.00500.6060log(PCR)log(ACR) and non-albumin proteinuria−0.00260.00540.6296log(ACR)log(ACR) and non-albumin proteinuria0.00000.00190.9966log(ACR)log(estACR)0.00080.00570.8824log(ACR) and non-albumin proteinurialog(estACR)0.00080.00600.8880log(PCR)log(estACR)−0.00180.00210.4058aAdjusted for eGFR, age, sex, rare disease group. SBC, a lower value represents a better fit Fit and concordance statistics for Cox models of time from concurrent ACR–PCR values to KF or Death, all adjusting for age, sex, eGFR and rare disease group, by proteinuria measurement used. aAdjusted for eGFR, age, sex, rare disease group. SBC, a lower value represents a better fit There were no statistically significant differences in Uno’s concordance statistic between any two models (Table 3), suggesting all models had similar discriminative abilities. Results were the same for patients with cystic and glomerular disease. ROC curve results (Fig. 3) assessed at 5 years were similar across models, with time-dependent AUC (SE) for all RaDaR estimated as 0.8535 (0.0081) for log(ACR), 0.8529 (0.0082) for log(PCR), and 0.8533 (0.0082) for log(ACR and non-albumin proteinuria) (Supplementary Table S4). AUC values were generally lower for models for patients with glomerular diseases compared with those with cystic diseases, but were almost identical within rare disease group, and estimated at >0.83 for all models (Supplementary Table S4). ROC curves for each Cox model of time from concurrent ACR–PCR values to KF or death, all adjusting for age, sex, eGFR and rare disease group, by proteinuria measurement used for (a) all RaDaR, (b) cystic, and (c) glomerular patients. When stratifying by ACR tertile (Supplementary Tables S5–S7), AIC and SBC results were again very similar across all proteinuria metrics within each tertile. For all RaDaR patients and for those with glomerular disease, out of log(PCR), log(ACR), and log(ACR) and log(non-albumin proteinuria), the model using log(PCR) had a marginally better fit across all ACR tertiles. For patients with cystic disease, log(ACR) showed the best fit in ACR tertiles 1 and 3 (tertile 1 AIC 134.74, SBC 414.7, tertile 3 AIC 830.17, SBC 848.81), and in ACR tertile 2 log (ACR) and log(non-albumin proteinuria) had the best fit (AIC 392.23, SBC 408.58), however, again results were very similar across all proteinuria measures. No statistically significant differences in Uno’s concordance statistic were seen by the rare disease group or ACR tertile. Across ACR tertiles, AUC results were generally similar. Although the AUCs in tertile 3 were lower for all RaDaR and the cystic sub-group, the magnitudes of the differences were small (Supplementary Table S4, Supplementary Figs. S7–S9). Estimated ACR Models using estimated ACR (estACR) had similar fit and concordance when compared to those using measured ACR and PCR (Table 3), with no statistically significant differences in Uno’s concordance statistic between the model using log(estACR) and models using measured log(ACR) (P = .42), log(PCR) (P = .56) and log(ACR and non-albumin PCR) (P = .36), and time-dependent AUC (SE) estimated as 0.8532 (0.0081). Stratifying by ACR tertile, the model using estACR showed similar fit and concordance to those using measured proteinuria metrics, even having marginally lower AIC and SBC statistics for all RaDaR and patients with glomerular diseases in ACR tertile 1 (All RaDaR: AIC 1063.65, SBC 1135.16, glomerular disease: AIC 724.27, SBC 754.92). Observed and predicted rates of KF within 5 years using the 4-variable Kidney Failure Risk equation were calculated using both measured ACR and estACR. As described previously, especially at higher levels of predicted risk, the KFRE underestimates KF risk in this rare kidney disease cohort [11]. That considered, we observed no difference in the performance of the KFRE using measured ACR and estACR, with Brier Scores (where lower values indicate better accuracy) of 0.171 for measured ACR and 0.170 for estACR across all ACR values (Fig. 4). When stratifying by predicted risk decile (Supplementary Fig. S10), there were no statistically significant differences in Brier Scores for measured ACR vs est(ACR) within each decile. Similar results were seen in the sub-groups of patients with cystic disease (measured ACR 0.146, estACR 0.147) and glomerular disease (measured ACR 0.180, estACR 0.177) (Fig. 4). The performance of ACR approximated by calculating PCR × 0.7 was poorer than both measured and estimated ACR (for all RaDaR, Brier score 0.183). Sensitivity analyses using median, rather than random, paired ACR–PCR measurements showed no differences in results (Supplementary Table 8 and Supplementary Figs. S11–S14). Brier scores for accuracy of KFRE using measured ACR, est(ACR), and approximated (× 0.7) ACR, for all RaDaR patients. DISCUSSION We investigated how equations and approximate ratios designed to enable the calculation of an estimated ACR (estACR) from PCR values perform in a rare kidney disease cohort. We found, consistent with previous studies, that estimation of ACR using the approximate ratio of 0.7 over-estimates ACR at lower levels of proteinuria, and that using the equation developed by Weaver et al. [23] conversely underestimates ACR at lower levels. We also examined the predictive performance in patients with rare kidney diseases of ACR, PCR, non-albumin proteinuria, and estACR for KF or death events, and found no consistent evidence of superiority of any of the tested proteinuria measurements, with very similar AIC and SBC (model fit) statistics, and no statistically significant differences in Uno’s global concordance statistic across all proteinuria measurements. For all RaDaR, the model with the best fit by AIC and SBC used log(PCR), however, differences between models were minimal. Analyses by disease sub-group showed similar results. In the glomerular sub-group, AIC and SBC statistics both suggested that of the measured proteinuria metrics, the Cox model using log(PCR) (AIC 6502.07, SBC 6561.35) had the best fit, which was consistent at all levels of proteinuria. While a previous study in a cohort of 315 ADPKD patients described the superiority of PCR over ACR in predicting KF [20], we found no evidence of better performance of PCR in our cystic sub-group. AIC and SBC statistics suggested a Cox model including log(ACR) had best fit in this sub-group, however, this difference was not present in any other analyses and was not consistent across ACR tertiles. Our results also indicate that estACR had very similar discriminative and predictive ability to measured ACR and PCR, with no statistical difference in model concordance between estACR and other proteinuria measures, and all proteinuria measurements having estimated time-dependent AUC of >0.83. In addition, we observed no difference in the performance of the KFRE using measured ACR and estACR (Brier Score for measured ACR 0.170, estACR 0.171). While a study in a large Korean cohort [20] with general CKD suggested estimated ACR (estimated from the Sumida et al. equation [26]) performed less well compared with measured ACR at lower proteinuria levels, we found model fit and concordance statistics for the Cox regression model using estACR were very similar to those for models using measured proteinuria metrics including in the lowest ACR tertile, even having marginally best fit (by AIC and SBC) for glomerular diseases in ACR tertile 1. These results suggest that estimated ACR can be used interchangeably with measured ACR for patients with rare kidney diseases clinically when using the KFRE for risk prediction, and also in Cox regression models of time to KF for observational studies or clinical trials across a range of proteinuria levels. In real-world practice, measuring albuminuria rather than total protein may not always be feasible, and as total urine protein measurement is often less expensive than urine albumin measurement [27], PCR may be preferred over ACR measurement on economic grounds, especially in low-to-middle income countries. Initiatives to better delineate and understand the association between changes in proteinuria and hard kidney outcomes in specific rare kidney diseases (for instance in IgA nephropathy and FSGS [28]), to establish proteinuria as a validated surrogate end-point enabling accelerated approval for novel therapies, have required international collaboration to identify sufficient patients for analysis. These results indicate that estACR calculated from PCR results can be used with no detectable loss of predictive ability, decreasing barriers to inclusion of data from countries where PCR, but not ACR, measurements may be available. Limitations Limitations of this study include that RaDaR is a UK based registry, and as such recruited patients reflect the socio-economic and ethnic make-up of the UK, receiving the standard of care in a developed country, which may make data presented here less transferable to other populations. While same-day paired results were available across a large range of low to high ACR values, patients with higher ACR results were more likely to have paired PCR samples (Supplementary Fig. S1), and therefore to be included in these analyses. While these analyses may be particularly relevant in patients with higher proteinuria levels, this distribution of data should be considered when interpreting our results. Furthermore, a single, random pair of ACR–PCR values were used per patient to avoid within-patient correlation. While this may have reduced the overall power of these analyses, there were still a large number of patients with data points available. Patients’ results following KRT initiation were excluded, therefore these results may not be applicable to patients who received a kidney transplant, and further work is required to examine the associations between measured proteinuria and kidney outcomes following transplantation. Underlying mechanisms of proteinuria may also be different in tubular disorders as compared to glomerular or cystic conditions. The small numbers of patients with primary tubular diseases included in this study (n = 74) did not allow for meaningful analyses in this sub-group, and therefore limits how these data may inform clinical trials focussing on tubular diseases. Finally, methods for measuring urine albumin and protein may vary, and there is no standardization nationally or internationally; studies where different measurement methods have been used may produce different results. An international standardized assay for urinary albumin will be available in 2027, which will help decrease variation in laboratory results within a patient over time, and also between laboratories. CONCLUSIONS In patients with rare kidney diseases, this study did not identify differences in the performance of ACR, PCR, non-albumin proteinuria, and estimated ACR to predict kidney failure or death events. Supplementary Material Contributor Information Katie Wong, Centre for Kidney and Bladder Health, University College London, UK; National Registry of Rare Kidney Diseases, UK Kidney Association, Bristol, UK. David Pitcher, Centre for Kidney and Bladder Health, University College London, UK; National Registry of Rare Kidney Diseases, UK Kidney Association, Bristol, UK. Jonathan Barratt, Mayer IgA Nephropathy Laboratories, University of Leicester, Leicester, UK. Daniel P Gale, Centre for Kidney and Bladder Health, University College London, UK; National Registry of Rare Kidney Diseases, UK Kidney Association, Bristol, UK. RaDaR Consortium, Multiple Affiliations. AUTHORS’ CONTRIBUTIONS D.P.G. conceived of the study and acquired funding. K.W. and D.P. curated data, performed formal analyses, and accessed and verified the data. D.P.G. and K.W. wrote the initial draft. J.B. and D.P. reviewed and edited the paper. D.P.G. and K.W. had final responsibility to submit for publication. CONFLICT OF INTEREST STATEMENT D.P.G. is supported by the St Peter’s Trust for Kidney, Bladder and Prostate research and has consulted or presented for Alexion, Alnylam, Novartis, Bayer, Britannia, SOBI, Vifor, Judo Bio, Sanofi, Otsuka, Calliditas, and Apellis. J.B. is a consultant for Alebund, Alnylam Pharmaceuticals, Alpine, Argenx, Astellas, BioCryst, Calliditas Therapeutics, Chinook, CSL Vifor, Dimerix, Hi-Bio, Kira, Novartis, Omeros, Otsuka, Q32 Bio, Roche, Sanofi, Takeda, Travere Therapeutics, Vera Therapeutics, and Visterra; and has received research funding from Argenx, Calliditas Therapeutics, Chinook, Galapagos, GlaxoSmithKline, Novartis, Omeros, Travere Therapeutics, and Visterra. All the other authors declared no competing interests. FUNDING This paper was published as part of a supplement financially supported by STADAPHARM GmbH. RaDaR was set up with initial significant grants from UKRI, Kidney Research UK and Kidney Care UK and is now developed, maintained and supported by the UK Kidney Association. DATA AVAILABILITY STATEMENT The RaDaR database is hosted by the UK Renal Registry and its metadata are available via https://rarerenal.org. Individual-level data are not available for export. Proposals to perform analyses using the data for academic, audit, or commercial purposes can be made to the RaDaR Operations Group via https://rarerenal.org. REFERENCES Associated Data Supplementary Materials Data Availability Statement The RaDaR database is hosted by the UK Renal Registry and its metadata are available via https://rarerenal.org. Individual-level data are not available for export. Proposals to perform analyses using the data for academic, audit, or commercial purposes can be made to the RaDaR Operations Group via https://rarerenal.org.