Integrated Use of Autosomal Dominant Polycystic Kidney Disease Prediction Tools for Risk Prognostication

Visual Abstract

Abstract

Key Points

• The Mayo clinic imaging classification and the predicting renal outcome in polycystic kidney disease score are used to assess the risk of progression to kidney failure in autosomal dominant polycystic kidney disease.

• Mayo imaging classification and predicting renal outcome in polycystic kidney disease show little concordance; combined use increased the ability to identify rapid progression especially among intermediate risk patients.

• Accurate risk prediction is key for determining indication for specific treatment.

Background

Autosomal dominant polycystic kidney disease is the most common genetic cause of kidney failure. Specific treatment is indicated on observed or predicted rapid progression. For the latter, risk stratification tools have been developed independently based on either total kidney volume or genotyping as well as clinical variables. This study aimed to improve risk prediction by combining both imaging and clinical-genetic scores.

Methods

We conducted a retrospective multicenter cohort study of 468 patients diagnosed with autosomal dominant polycystic kidney disease. Clinical, imaging, and genetic data were analyzed for risk prediction. We defined rapid disease progression as an eGFR slope ≥3 ml/min per 1.73 m² per year over 2 years, Mayo imaging classification (MIC) 1D–1E, or a predicting renal outcome in polycystic kidney disease (PROPKD) score of ≥7 points. Using MIC, PROPKD, and rare exome variant ensemble learner scores, several combined models were designed to develop a new classification with improved risk stratification. Primary endpoints were the development of advanced CKD stages G4–G5, longitudinal changes in eGFR, and clinical variables such as hypertension or urological events. Statistically, logistic regression, survival, receiver operating characteristic analyses, linear mixed models, and Cox proportional hazards models were used.

Results

PKD1-genotype (P < 0.001), MIC class 1E (P < 0.001), early-onset hypertension (P < 0.001), and early-onset urological events (P = 0.003) correlated best with rapid progression in multivariable analysis. While the MIC showed satisfactory specificity (77%), the PROPKD was more sensitive (59%). Among individuals with an intermediate risk in one of the scores, integration of the other score (combined scoring) allowed for more accurate stratification.

Conclusions

The combined use of both risk scores was associated with higher ability to identify rapid progressors and resulted in a better stratification, notably among intermediate risk patients.

Introduction

Autosomal dominant polycystic kidney disease (ADPKD) is the most common genetic kidney disorder leading to kidney failure^1–4 due to bilateral cystic organ enlargement and subsequent eGFR decline.^1,2,5 To predict disease progression, tools for clinical guidance have been developed. With the advent of the first specific treatment (tolvaptan), timely identification of patients who are prone to rapid progression became most important.^6–8 To date, the main prediction tool used in clinical practice is the Mayo imaging classification (MIC). The classification is based on age-adjusted and height-adjusted total kidney volume (htTKV), which correlates with future eGFR-decline.^9–11 In ADPKD, kidney volumes increase at an average rate of approximately 5% per year and larger kidneys are associated with a more rapid decline in kidney function, highlighting the role of total kidney volume (TKV) as an informative marker of disease severity and progression.^11–14 The typical MIC class 1 discerns five subclasses (A–E) with differential growth rates (1.5%, 1.5%–3%, 3%–4.5%, 4.5%–6%, and 6% per year) yielding differential ages of expected kidney failure. During a 10-year follow-up period, MIC classes were strongly predictive of kidney survival.^11,15 Apart from imaging-based prognostication, genetic and clinical variables have been found to correlate with rapid progression.¹⁶ Dozens of studies report kidney outcomes to be associated with the gene and the type of variant: On average, kidney failure occurs approximately 20 years earlier in PKD1 than in PKD2 patients.² In addition, truncating PKD1 (PKD1-T) variants are associated with more severe kidney disease than nontruncating alterations (PKD1-NT), although a low percentage may still have mild disease on the basis of clinical and imaging analysis.^1,17–19 In 2016, the predicting renal outcome in polycystic kidney disease (PROPKD) score was proposed. Based on a cohort of 1341 patients, the PROPKD score considers both clinical and genetic factors to assign patients into low-risk (0–3 points), intermediate-risk (4–6 points), and high-risk (7–9 points) categories which showed expected kidney failure at 71, 57, and 49 years, respectively.^6,20 The score implements sex (male/female), genotype (PKD1-T/PKD1-NT/PKD2), early-onset (younger than 35 years) hypertension, and early-onset (younger than 35 years) urological events. Although shown to be an excellent predictor of rapid progression in patients with early-onset hypertension or urological events, limitations of the PROPKD score concern individuals younger than 35 years at assessment and families with vast phenotypic variability.⁶ Recently, a European consensus statement on behalf of the European Renal Association (among others) suggests to administer tolvaptan in two scenarios: firstly, observed eGFR-decline exceeding 3 ml/min per 1.73 m² per year over 4 years (gold standard) or secondly, predicted eGFR-decline exceeding 3 ml/min per 1.73 m² per year as assessed by TKV/MIC or PROPKD score.⁸ Therefore, current prediction scores independently define patients with rapid progression in clinical practice.^8,21,22 Whether MIC or PROPKD is favored largely depends on access to tomographic-imaging (MIC) or availability of genetic testing (PROPKD) in respective national health care systems. Therefore, few studies have evaluated the performance of both scores in real-world settings, and there are limited data on score concordance.^23,24 In addition, a detailed evaluation of the combined use of these tools in predicting rapid progression is underexplored. Therefore, the objective of this observational multicenter study was to evaluate concordance, investigate sensitivity/specificity, and test for the additive value of combined scoring. Moreover, the study evaluated the possibility of improving the PROPKD score performance through the addition of more granular genetic information through the rare exome variant ensemble learner (REVEL),^25,26 the current in silico gold standard tool for distinguishing rare pathogenic missense variants from neutral ones.²⁷

Methods

Population

The study cohort was recruited at four tertiary care centers Berlin (N=162), Leipzig (N=95), Bologna (N=117), and Dublin (N=94) between the years of 2016 and 2023. Inclusion criteria were a clinically and genetically confirmed diagnosis of ADPKD with typical MIC (class 1), older than 18 years, and detectable causative gene variant (PKD1 or PKD2). Patients with atypical ADPKD (class 2), variants in disease-causing genes other than PKD1/2, or incomplete clinical data were excluded.^6,28 In total, 264 patients were excluded for not fulfilling all criteria, leaving 468 individuals for analyses. For further details regarding the collected patient data which encompasses genetic and radiological analyses, and other relevant information, please refer to the Supplemental Methods.

Ethics

Written informed consent was obtained from all participants. Prior approval was obtained from the participating institutions and ethics and human research committees including Institutional Review Board protocols from the University of Bologna (010/2021/Oss/AOUBo), University of Leipzig (289/20-ek and 402/16-ek), Charité Universitätsmedizin Berlin (EA4/066/21), and Beaumont Hospital Dublin (REC-19/28). This study was performed in compliance with the ethical guidelines of the 1975 Declaration of Helsinki. Every patient provided written informed consent regarding publication of deidentified genetic data.

Study Endpoints and Outcomes

The diagnosis of arterial hypertension was defined by the treating physician (indication for antihypertensive therapy). Patients' eGFR was calculated from the measured serum creatinine using the non–race-based CKD Epidemiology Collaboration equation.^29–31 The primary endpoint was defined as the development of advanced CKD (aCKD) stages G4–G5 normalized to age at CKD G4 (first diagnosis) or if not available: age at CKD G5, initiation of KRT, or eGFR persistently <15 ml/min per 1.73 m².²⁹ Patients were categorized into rapid and slow progression of disease on the basis of eGFR slope, PROPKD score, and MIC class as follows: Those with an eGFR slope of ≥3 ml/min per 1.73 m² per year over at least 2 years, a high-risk PROPKD score (7–9 points), or MIC class 1D–1E were defined as rapid progression. In an exploratory, nonvalidated approach, the REVEL score, predicting the pathogenicity of PKD1 missense variants of primarily uncertain significance, was used for refinement of the PROPKD (PROPKD-R) score. Those with a REVEL score above the threshold of 0.75 were classified as pathogenic. Subsequently, variants previously classified as PKD1-NT, which contributed two points to the PROPKD score, were reclassified as PKD1-T variants, resulting in four points being allocated to the PROPKD-R score calculation.²⁷ The web application ADPKD-Risk was developed to support the visualization and prognostic risk stratification of ADPKD progression. Vue.js and Chart.js were used to create an interactive platform, where users can enter patient-specific data (age, height, genotype, clinical characteristics, kidney volume, etc.) which the app uses to calculate the MIC and PROPKD score, assign an individual risk score. These data points are displayed on a responsive graph showing the MIC and PROPKD score in context. The web application (https://halbritter-lab.github.io/adpkd-risk/) and its source code are freely available.

Statistical Analyses

Apart from descriptive cohort statistics, concordance between MIC and PROPKD scores were assessed using Kappa correlation statistics. Linear and logistic regression models were used to evaluate the significance of covariates in predicting eGFR decline and htTKV growth rate. P values were derived from the Kruskal-Wallis test for continuous variables and the chi-squared test for categorical variables when comparing baseline characteristics of patients by genotype. Primary endpoint comparison was performed by Kaplan–Meier survival analysis. In all survival analyses, time zero refers to status at birth (age 0). T tests with a significance level of P ≤ 0.05 were used for normally distributed two-sided analyses, and log-rank tests were used to evaluate differences in the risk of aCKD by PROPKD, genotype, and MIC. Variables associated with aCKD were compared using univariate and multivariable Cox proportional hazard, longitudinal mixed-effects regression and receiver operating characteristic (ROC) analysis with concordance statistics. Hazard ratios (HRs) were reported with a 95% confidence interval.

Results

Baseline Characteristics

In total, 468 patients with clinically and genetically diagnosed ADPKD were analyzed. The distribution of sex, age, and genotype was similar between the recruiting centers. The mean age was 50 (±14) years, and the sex distribution comprised 53% of female patients. There were 382 (82%) patients with a family history of PKD; 221 (47%) had hypertension diagnosed before the age of 35, and 157 (34%) experienced a urological event before the age of 35. Genotypes comprised 239 (51%) patients with (likely) pathogenic variants in PKD1-T, 138 (30%) in PKD1-NT, and 91 (19%) in PKD2. During the mean follow-up period of 37 months, 170 (36%) patients developed aCKD (CKD stages G4–G5), meeting the primary clinical endpoint. Imaging data were available for all patients for calculation of htTKV for MIC assessment. The mean TKV was 2126 ml (interquartile range, 800–2917 ml), and the mean htTKV was 1218 ml/m (interquartile range, 484–1619 ml/m). Notably, hypertension and urological events were significantly more frequent in PKD1 patients than in PKD2 patients (chi-squared test P < 0.001, P < 0.001, respectively), which was also the case for patients with a family history of PKD (chi-squared test P = 0.02). Medication use data were available for most of the patients. There were 129 (28%) patients who received tolvaptan over a period of at least 1 year. The first-line antihypertensive therapy was angiotensin-converting enzyme inhibitors or angiotensin receptor blockers in all centers, and 235 (50%) patients received this medication at the time of inclusion into the study. At follow-up, over 90% of the patients received any kind of antihypertensive therapy (Table 1 and Supplemental Figures 1 and 2).

Table 1.

Summary of the baseline characteristics of the study cohort by genotypic group

Baseline Characteristics	Genotypic Groups
	PKD1-T (N=239)	PKD1-NT (N=138)	PKD2 (N=91)	Total (N=468)	P Value
Sex, n (%)
Female	123 (51)	72 (52)	52 (57)	247 (53)	0.65
Male	116 (49)	66 (48)	39 (43)	221 (47)
Age at inclusion, mean (SD)	48 (14)	51 (14)	56 (14)	50 (14)	<0.001a
Age at diagnosis, mean (SD)	31 (13)	36 (15)	42 (15)	35 (15)	<0.001a
aCKD, n (%)	103 (43)	43 (31)	24 (26)	170 (36)	0.006a
Age at aCKD, mean (SD)	49 (9)	54 (10)	65 (7)	53 (10)	<0.001a
eGFR, n	231	136	89	456 (97)
Age at eGFR, mean (SD)	44 (12)	47 (13)	54 (14)	47 (13)	<0.001a
eGFR CKD-EPI (ml/min per 1.73 m2), mean (SD)	47 (35)	53 (33)	54 (30)	50 (33)	0.16
Tolvaptan intake, n (%)	79 (33)	36 (26)	14 (15)	126 (27)	0.005a
Urological event, n (%)	108 (45)	37 (27)	11 (12)	156 (33)	<0.001a
Hypertension, n (%)	133 (56)	70 (51)	18 (20)	221 (47)	<0.001a
Family history, n (%)	184 (77)	122 (88)	76 (84)	382 (82)	0.02a
htTKV, n	235	136	91	462 (99)
Age at TKV/htTKV, mean (SD)	43 (13)	46 (13)	52 (14)	45 (13)	<0.001a
htTKV (ml/m), mean	1275	1160	1392	1264	0.36
MIC, n (%)	239 (100)	138 (100)	91(100)	468 (100)
1A	6 (3)	9 (6)	12 (13)	27 (6)	<0.001a
1B	40 (17)	22 (16)	28 (31)	90 (19)	<0.001a
1C	77 (32)	62 (45)	28 (31)	167 (36)	<0.001a
1D	61 (25)	30 (22)	18 (20)	109 (23)	<0.001a
1E	55 (23)	15 (11)	5 (5)	75 (16)	<0.001a
PROPKD risk score, n (%)	239	138	91	468 (100)
Low risk (0–3)	0 (0)	56 (41)	88 (97)	144 (31)	<0.001a
Intermediate risk (4–6)	108 (45)	69 (50)	3 (3)	180 (38)	<0.001a
High risk (7–9)	131 (55)	13 (9)	0 (0)	144 (31)	<0.001a

Descriptive results for continuous variables are reported as mean (±SD) and for categorical variables as frequency count (percentage). The Bonferroni correction was used to correct for multiple testing. Scores refer to subscore unless otherwise indicated. aCKD, advanced CKD; CKD-EPI, CKD Epidemiology Collaboration; htTKV, height-adjusted total kidney volume; MIC, Mayo imaging classification; PKD1-T, polycystic kidney disease 1 truncating variant; PKD1-NT, polycystic kidney disease 1 nontruncating variant; PKD2, polycystic kidney disease 2; PROPKD, predicting renal outcome in polycystic kidney disease; TKV, total kidney volume.

^aSignificant P values.

Variables Influencing Kidney Survival

The primary endpoint aCKD was reached by 170 patients at a mean age of 53 (±10) years, either at onset of CKD stage G4 (N=68) or stage G5 (N=102). The analysis of sex differences revealed that male patients presented with more severe kidney disease outcomes compared with female patients. Male patients exhibited significantly lower eGFR, higher mean TKV, and a greater prevalence of early-onset hypertension. Furthermore, male patients were more frequently treated with tolvaptan and classified as rapid progressors, while female patients represented a larger proportion of slow progressors. However, survival analysis showed no statistically significant differences, with male patients reaching aCKD at a mean age of 52 (±11) years, 1 year earlier than female patients (53±10; Supplemental Figure 3A and Supplemental Table 1). Thirty-nine percent of PKD1 patients developed aCKD at a mean age of 51 (±9) years compared with 26% PKD2 patients at a mean age of 65 (±7) years (Supplemental Figure 3B). Further analysis also revealed the well-established, yet significant, differences between PKD1 genotypes, with the mean age at aCKD onset being 49 (±9) years for PKD1-T versus 54 (±10) years for PKD1-NT patients. The age at onset of aCKD also varied among the different MIC subclasses, ranging from 56 (±11) years for MIC-1A to 43 (±6) years for MIC-1E. Less than 20% of MIC-1A patients reached the primary endpoint, compared with 51% of MIC-1E patients. When comparing the risk categories for MIC, low-risk patients (MIC-1A and 1B) reached the endpoint at 59 (±10) years, compared with 56 (±9) years for intermediate risk (MIC-1C) and 48 (±9) years for high-risk (MIC-1D and 1E; P < 0.001; Supplemental Figure 3, C and D). Similarly, for the PROPKD risk categories, low-risk patients (score 0–3) reached the primary endpoint at an average age of 62 (±8) years, compared with 52 (±9) years for intermediate-risk (score 4–6) and 48 (±9) years for high-risk patients (score 7–9; P < 0.001; Supplemental Figure 3, E and F).

Observed and Predicted Rapid Progression

To determine the sensitivity and specificity of both prediction tools, the actual eGFR decline of a subgroup of 259 patients with at least 24 months of available kidney function data were assessed. Thereby, these patients were stratified into progression groups: actual slow progressors (<3 ml/min per 1.73 m² per year; n=131) and actual rapid progressors (≥3 ml/min per 1.73 m² per year; n=128) through this observational real-world data (Table 2).⁸ At baseline, both groups had similar mean eGFR levels (67±33 as compared with 68±25 ml/min per 1.73 m²), but rapid progression patients showed higher mean htTKV values (1330 ml/m compared with slow progressors 852 ml/m). There was no difference in the sex distribution within the rapid progression group (51% female, 49% male). However, in the slow progression group, women predominated with 67%. The percentage of early-onset (younger than 35 years) hypertension was significantly higher among rapid progressors (52%) compared with slow progressors (36%). Furthermore, occurrence of an urological event before the age of 35 was reported in 51 (40%) rapid, compared with 29 (22%) slow progression patients. The frequency of PKD1-T variants was higher for rapid versus slow progressors (48% versus 42%). Conversely, the frequency of PKD2 variants was lower in rapid progression patients (19% versus 26%; Table 2).

Table 2.

Summary of the baseline characteristics of the study cohort classified according to the clinical definition of rapid progression (slow progressor: eGFR slope <3 ml/min per 1.73 m² per year after 2 years; rapid progressor: eGFR slope ≥3 ml/min per 1.73 m² per year after 2 years)

Baseline Characteristics	Slow Progressor (n=131)	Rapid Progressor (n=128)	Total (n=259)
Sex, n (%)
Female	88 (67)	65 (51)	153 (59)
Male	43 (33)	63 (49)	106 (41)
aCKD, n (%)	44 (34)	27 (21)	71 (27)
Age at aCKD, mean (SD)	55 (10)	56 (12)	55 (11)
Height, m, mean	1.71	1.72	1.71
Weight, kg (SD)	73 (14)	73 (18)	73 (16)
Age, yr (SD)	50 (15)	51 (11)	50 (14)
Age at eGFR, mean (SD)	44 (13)	45 (13)	44 (13)
eGFR first visit CKD-EPI (ml/min per 1.73 m2), mean (SD)	67 (33)	68 (25)	68 (29)
Age at TKV/htTKV, mean (SD)	45 (13)	46 (13)	45 (13)
htTKV ml/m, mean	852	1330	1086
Hypertension <35 yr, n (%)	47 (36)	67 (52)	114 (44)
Urological-event <35 yr, n (%)	29 (22)	51 (40)	80 (31)
Family history, n (%)	105 (80)	103 (81)	208 (80)
Genotype, n (%)
PKD1-T	55 (42)	61 (48)	116 (45)
PKD1-NT	42 (32)	42 (33)	84 (32)
PKD2	34 (26)	25 (19)	59 (23)
MIC, n (%)
Low risk (1A–1B)	48 (37)	21 (16)	69 (27)
Intermediate risk (1C)	53 (40)	50 (39)	103 (40)
High risk (1D–1E)	30 (23)	57 (45)	87 (33)
PROPKD risk score, n (%)
Low risk (0–3)	56 (43)	38 (29)	94 (37)
Intermediate risk (4–6)	46 (35)	51 (40)	97 (37)
High risk (7–9)	29 (22)	39 (31)	68 (26)

Descriptive results for continuous variables are reported as mean (±SD) and for categorical variables as frequency count (percentage). Scores refer to subscore unless otherwise indicated. aCKD, advanced CKD; CKD-EPI, CKD Epidemiology Collaboration; htTKV, height-adjusted total kidney volume; MIC, Mayo imaging classification; PKD2, polycystic kidney disease 2; PKD1-NT, polycystic kidney disease 1 nontruncating variant; PKD1-T, polycystic kidney disease 1 truncating variant; PROPKD, predicting renal outcome in polycystic kidney disease; TKV, total kidney volume.

Among the rapid progressors, 57 patients (45%) were classified as high-risk MIC-1D or 1E, and 39 (31%) had a high risk PROPKD score. On examining the predictive capability of both tools (MIC and PROPKD) for identifying actual rapid progressors, it was found that the MIC displayed a higher specificity in detecting rapid progression patients (specificity 77% and sensitivity 45%). However, the PROPKD score showed superior sensitivity, correctly identifying 59% of slow progressors with a specificity of 57%. Modifying the PROPKD score by reclassifying missense variants using the REVEL tool was not associated with higher sensitivity, which was confirmed by the ROC analysis (Figure 1). The proportion estimates and the area under the ROC were pooled according to the MIC (0.65) and the PROPKD score (0.60) alone and in a combination of the PROPKD and REVEL scores (termed PROPKD-R score; 0.60), the MIC and PROPKD scores (0.67), and the MIC and PROPKD-R scores (0.67). The combination of MIC and PROPKD resulted in a higher area under the curve (AUC) and improved sensitivity and specificity values. On comparing the AUCs of the models with the DeLong test, no statistical significance was observed between the three individual scores when comparing each score alone or the two combined models with each other. However, there were significant results when comparing each of the three individual scores with any of the two combined models (Figure 1 and Table 2).

Figure 1.

ROC with sensitivity, specificity, and AUROC estimates for each of the different scores alone and in combination. AUROC, area under the receiver operating characteristic; ROC, receiver operating characteristic; AUC, area under the curve.

Improving Known Scores for Better Risk Stratification

For differences in their prognostic ability, we tested the concordance and mutual correlation of the MIC and the PROPKD score. There was only a slight positive concordance with Cohen Kappa of 0.20. Additional correlation tests also showed only slightly positive results, with a Gwet’s AC1 coefficient being of 0.21 (P < 0.001), a Spearman-Rho coefficient of 0.41 (P < 0.001), and a Kendall-Tau-b coefficient of 0.33 (P < 0.001; Figure 2). In total, only 47% of patients (N=220) showed concordance in risk categorization into low, intermediate, and high. Furthermore, 58% of patients (N=273) were classified as intermediate risk by at least one of the scores, and 16% (N=74) were assigned to the intermediate-intermediate risk category (Supplemental Table 2). As a preponderance of the patients fell into the intermediate groups, specifically with regard to PROPKD scoring, we used the REVEL score to differentiate between PKD1 missense variants and thereby modify the PROPKD score (PROPKD-R). However, after their coalescence, concordance between the MIC and PROPKD-R scores did not improve significantly (Supplemental Table 3).

Figure 2.

Correlation analysis of MIC and PROPKD scores. There is a slight positive correlation, with a Spearman-Rho coefficient of 0.43 when looking at the total score points (A) and when just looking at the risk classes (B), with a coefficient of 0.41, within the total cohort. The numbers next to the circles in the legend represent the size of the circles, which correspond to the frequency or count of data points in each category. MIC, Mayo imaging classification; PROPKD, predicting renal outcome in polycystic kidney disease.

For improved individual risk assessment, especially among intermediate-risk patients, the MIC and PROPKD scores were combined in both directions into the following six subcategories: low-low, intermediate-low, low-high, intermediate-intermediate, intermediate-high, and high-high. This approach distinguished significantly between subcategories (Table 3 and Supplemental Figures 4 and 5). Especially for patients classified as intermediate risk by either of the scores, the combined assignment into intermediate/high, intermediate/intermediate, and intermediate/low yielded discriminative age at aCKD-onset (51±8 versus 54±9 versus 58±9 years; Figure 3 and Table 3). Beyond the age at aCKD-onset, other variables such as sex, age at inclusion, age at ADPKD-diagnosis, hypertension-diagnosis or urological events before the age of 35 years, eGFR, htTKV, and the age at first imaging showed statistically significant differences between the six risk groups (Supplemental Figure 5, C and D). Further, this allowed for an enhanced predictive discrimination regarding the risk of experiencing rapid progression. To exclude potential bias of specific treatment, we conducted additional analysis of the combined risk groups excluding patients who received tolvaptan. The distribution into the six risk groups remained balanced, and the groups exhibited the same statistically significant differences from one another. For patients classified as intermediate risk by one of the scores, the combined assignment into intermediate/high, intermediate/intermediate, and intermediate/low still yielded a discriminative age at aCKD onset (Supplemental Figure 6 and Supplemental Table 6).

Table 3.

Summary of the characteristics of the study cohort stratified into six risk combination groups, which were modeled in both directions by the three risk classes of the Mayo imaging classification and the predicting renal outcome in polycystic kidney disease score (low-low/intermediate-low/low-high/intermediate-intermediate/high-intermediate/high-high).

Baseline Characteristics	Risk Combination Groups
	Low-Low (N=59)	Inter-Low (N=94)	Low-High (N=49)	Inter-Inter (N=74)	Inter-High (N=105)	High-High (N=87)	P Value
Sex, n (%)
Female	46 (78)	53 (56)	21 (43)	51 (69)	53 (51)	23 (26)	<0.001a
Male	13 (22)	41 (44)	28 (57)	23 (31)	52 (49)	64 (74)
Age at inclusion, mean (SD)	55 (14)	56 (15)	54 (13)	48 (15)	47 (13)	44 (12)	0.005a
Age at diagnosis, mean (SD)	42 (14)	40 (15)	37 (16)	34 (13)	31 (14)	27 (12)	0.001a
aCKD, n (%)	11 (19)	32 (34)	18 (37)	24 (32)	39 (37)	46 (53)	<0.001a
Age at aCKD, mean (SD)	68 (6)	58 (9)	58 (7)	54 (9)	51 (8)	44 (7)	<0.001a
eGFR, n	58	92	48	71	103	84
Age at eGFR, mean (SD)	53 (14)	52 (13)	50 (11)	44 (14)	44 (12)	41 (9)	<0.001a
eGFR CKD-EPI (ml/min per 1.73 m2), mean (SD)	63 (29)	52 (34)	46 (31)	56 (38)	50 (33)	38 (29)	0.14
Urological event, n (%)	4 (7)	11 (12)	12 (25)	23 (31)	40 (38)	66 (76)	<0.001a
Hypertension, n (%)	6 (10)	15 (16)	18 (37)	34 (46)	67 (64)	81 (93)	<0.001a
Family history, n (%)	48 (81)	80 (85)	36 (74)	66 (89)	87 (83)	65 (75)	0.37
htTKV, n	59	93	49	73	103	85
Age at htTKV, mean (SD)	50 (14)	51 (14)	49 (12)	44 (15)	42 (12)	39 (9)	<0.001a
htTKV (ml/m), mean	426	865	2154	850	1378	1990	0.10

Descriptive results for continuous variables are reported as mean (±SD) and for categorical variables as frequency count (%). aCKD, advanced CKD; CKD-EPI, CKD Epidemiology Collaboration; htTKV, height-adjusted total kidney volume; PROPKD, predicting renal outcome in polycystic kidney disease.

^aSignificant P values.

Figure 3.

Clinical comparison of intermediate risk MIC/PROPKD patients. Renal survival analyses (A) for MIC/PROPKD risk combinations with focus on the intermediate cases only (intermediate/high, intermediate/intermediate, intermediate/low; n=273) are shown here with the mean age for reaching the primary endpoint aCKD for every risk class (B). From the mildest risk class intermediate-low to the most severe class intermediate-high, the mean ages were 58.17, 53.58, and 51.36 years, respectively (P < 0.001). The eGFR measurements at baseline (C) and htTKV values (D) confirm the good discrimination between the three classes, as indicated by the survival analyses. aCKD, advanced CKD; htTKV, height-adjusted total kidney volume.

Univariate and Multivariable Analyses

To investigate which of the clinical factors were decisive in the combined risk stratification, both univariate and multivariable analyses were performed. Several factors, including genotype, MIC and PROPKD risk groups, hypertension diagnosis, occurrence of urological events, and family history of PKD, were associated with the incidence of aCKD in the univariate analysis (Table 4). The risk of reaching aCKD was five times higher in PKD1-T patients and two times higher in PKD1-NT patients than in PKD2 patients HR of 5.40 and 2.21 independent of sex and MIC class. Similar results were observed for MIC classes, with class 1E patients carrying the highest risk of developing aCKD (15-fold) followed by MIC-1D patients with a three-fold risk compared with the reference group MIC-1A (HR, 14.59 and 2.81). The HRs for MIC-1B and 1C patients were nonsignificant. Urological events and hypertension diagnosis before the age of 35 years were found to be valuable clinical markers for the development of aCKD, with a three-fold to four-fold higher risk (HR, 2.83 and 3.72; Table 4A).

Table 4.

Univariate Cox regression analysis between Mayo imaging classification and predicting renal outcome in polycystic kidney disease variables with incidence of advanced CKD as the endpoint (A), multivariable Cox-regression analysis between Mayo imaging classification (B), predicting renal outcome in polycystic kidney disease (C), or predicting renal outcome in polycystic kidney disease and Mayo imaging classification and other variables (D) with incidence of advanced CKD as the endpoint and Harrell C-statistics for concordance probability

Cox-regression Variables	N	HR (95% CI)	P Value	C-Statistics (95% CI)
(A) Univariate
Genotype	468	—	—	0.69 (0.65 to 0.72)
PKD1-T	—	5.40 (3.44 to 8.79)	<0.001a	—
PKD1-NT	—	2.21 (1.40 to 3.72)	0.002a	—
PKD2	—	Ref	Ref	Ref
Male (versus female)	468	1.22 (0.90 to 1.65)	0.20	0.54 (0.50 to 0.59)
Urological event (versus no event)	468	2.83 (2.06 to 3.87)	<0.001a	0.65 (0.61 to 0.69)
Hypertension (versus no HTN)	468	3.72 (2.69 to 5.17)	<0.001a	0.66 (0.63 to 0.70)
MIC	468	—	—	0.72 (0.67 to 0.76)
1A	—	Ref	Ref	Ref
1B	—	1.13 (0.48 to 3.34)	0.80	—
1C	—	1.54 (0.68 to 4.43)	0.36	—
1D	—	2.81 (1.22 to 8.14)	0.03a	—
1E	—	14.59 (6.10 to 43.37)	<0.001a	—
Family history	468	0.99 (0.68 to 1.50)	0.97	0.49 (0.46 to 0.52)
(B) MIC	468	—	—	0.72 (0.68 to 0.77)a
1A	—	Ref	Ref	Ref
1B	—	1.14 (0.48 to 3.37)	0.79	—
1C	—	1.60 (0.70 to 4.60)	0.32	—
1D	—	3.08 (1.32 to 9.02)a	0.02a	—
1E	—	16.03 (6.58 to 48.13)a	<0.001a	—
Male (versus female)	—	0.83 (0.60 to 1.15)	0.26	—
Family history	—	1.09 (0.74 to 1.65)	0.69	—
(C) PROPKD	468	—	—	0.77 (0.74 to 0.81)a
PKD1-T	—	3.63 (2.23 to 6.10)a	<0.001a	—
PKD1-NT	—	1.56 (0.93 to 2.67)	0.01a	—
PKD2	—	Ref	Ref	Ref
Male (versus female)	—	1.12 (0.82 to 1.54)	0.46	—
Urological event (versus no event)	—	2.04 (1.46 to 2.84)a	<0.001a	—
Hypertension (versus no HTN)	—	2.71 (1.93 to 3.81)a	<0.001a	—
Family history	—	1.19 (0.80 to 1.83)	0.41	—
(D) MIC and PROPKD	468	—	—	0.81 (0.78 to 0.85)a
PKD1-T	—	3.91 (2.38 to 6.63)a	<0.001a	—
PKD1-NT	—	1.78 (1.05 to 3.07)a	0.04	—
PKD2	—	Ref	Ref	Ref
Male (versus female)	—	0.82 (0.59 to 1.14)	0.23	—
Urological event (versus no event)	—	1.67 (1.18 to 2.34)a	0.003a	—
Hypertension (versus no HTN)	—	2.47 (1.74 to 3.51)a	<0.001a	—
MIC-1A	—	Ref	Ref	Ref
MIC-1B	—	0.59 (0.24 to 1.77)	0.29	—
MIC-1C	—	0.73 (0.31 to 2.14)	0.51	—
MIC-1D	—	1.26 (0.52 to 3.80)	0.64	—
MIC-1E	—	5.21 (2.00 to 16.38)a	<0.001a	—
Family history	—	1.39 (0.93 to 2.16)	0.12	—

Age scale used for all Cox regression models. The univariable and multivariable models are based on complete cases, N=468. CI, confidence interval; HR, hazard ratio; HTN, hypertension; MIC, Mayo imaging classification; PKD2, polycystic kidney disease 2; PKD1-NT, polycystic kidney disease 1 nontruncating variant; PKD1-T, polycystic kidney disease 1 truncating variant; PROPKD, predicting renal outcome in polycystic kidney disease.

^aSignificant hazard ratios (95% confidence interval) and high C-values.

In the multivariable Cox regression analysis, we tested the influence of the MIC and the PROPKD score alone, combined and coupled with additional variables such as family history of PKD. In the first analysis on MIC, classes 1D and 1E remained statistically significant predictors of aCKD (Table 4B). In the second analysis on PROPKD, PKD1-T, hypertension, and urological events before the age of 35 years remained statistically significant (Table 4C). Both the PROPKD and the MIC demonstrated strong discriminatory power for the endpoint (Supplemental Figure 7). In a third multivariable analysis, the two scores were combined (Table 4D). Once more, an improvement in the C-index (0.81) was observed. The genotypic groups PKD1-T and PKD1-NT compared with PKD2, as well as MIC-1E, urological events, and hypertension, scored significantly better than in the previous approach using MIC or PROPKD variables alone (C-indexes of 0.72 and 0.77, respectively; Table 4). Likelihood ratio tests (G²) of the multivariable Cox analyses were performed to determine which model was most accurate. All models were statistically significant (P < 0.001), confirming their accuracy. The combined model, however, had the highest scores (94.6 for the MIC score alone, 124.7 for the PROPKD score alone, and 184.6 for the combined scores), indicating that a combination of both scores resulted in an additive value (Figure 4A and Table 4). Finally, Cox analysis of the six risk categories in comparison with the reference class low-low showed significantly different HRs (HRs, 2.57, 3.94, 4.34, 8.32, and 27.70 for risk classes 2, 3, 4, 5, and 6, respectively; Figure 4B).

Figure 4.

Forest plots displaying the HRs for the PROPKD score, MIC, additional variables (A), and the six distinct risk combination classes (B). Reference values are MIC-1A, PKD2, and for the final analysis the risk class low-low. HR, hazard ratio.

Discussion

Owing to the vast clinical heterogeneity, identifying those patients who would benefit most from specific treatment (e.g., tolvaptan) is crucial in ADPKD. Therefore, accurate prognostication of rapid disease progression at early stages is required. Over the past years, scoring systems, such as the MIC and the PROPKD score have been developed to allocate treatment on the basis of predicted disease progression. Individually, both scores are currently widely used⁸; however, there is still no consensus on the optimal predictive model for identifying rapid progressors.^8,22,24,32 To date, only two studies have tested combined or additive modeling on the basis of genotype and kidney imaging. In the large consortium for radiologic imaging studies of Polycystic Kidney Disease and the Polycystic Kidney Disease Treatment Network cohort, it was demonstrated that both genetic and MIC parameters were highly predictive of the endpoint kidney failure and eGFR decline. The discriminatory power of MIC classes was found to be slightly superior to that of genotypes. However, the predictive power was further enhanced when both genotype and MIC data were combined, compared with when either was used alone.^10,23

This study provides a comprehensive investigation of determinants that influence the progression. To avoid bias of atypical ADPKD, we only included patients with variants in PKD1 or PKD2. Numerous studies have focused on either imaging-based prediction or genotypes; however, the strength of our study is the integration of both imaging and clinical-genetic data.^{15,20,25,26,28,33}

The first objective was to evaluate and improve the concordance between the MIC and the PROPKD score. Subsequently, we investigated the effectiveness of a combined model using both scores to predict the probability of disease progression to aCKD and kidney failure. Cohen statistical analysis revealed poor concordance and a moderately positive correlation of the MIC and the PROPKD score. Specifically, more than half of the patients identified as rapidly progressing by the MIC score were not identified as such by the PROPKD score. By contrast, 40% of patients considered rapidly progressing according to the PROPKD score were not considered as rapid progressors when applying the MIC. When comparing these results with the clinical definition of rapid progression, the MIC was found to be specific (77%) but not very sensitive (45%). On the other hand, the PROPKD score seemed to be more sensitive (59%), but not very specific (57%). The overall performance of the ROC analyses resulted in satisfactory discrimination with AUCs slightly below 0.7. We determined that low concordance was mainly driven by two influencing factors: the intermediate MIC-1C class and PKD1-NT variants because these categories applied to most patients. Consequently, the intermediate risk groups of the MIC and the PROPKD score are difficult to interpret in terms of the likelihood of experiencing rapid progression. Next, we used REVEL to refine the classification of nontruncating PKD1 missense variants. The aim was to use this in silico tool as a modifier for the PROPKD score, to obtain a more accurate risk stratification for the considerable number of missense variants that fall within an intermediate risk category. However, after reclassification concordance did not improve, and the adjusted PROPKD-R score did not perform better on multivariable analysis. Because the exploratory use of REVEL did not demonstrate clinical utility in our model, it was excluded from further analyses.

By combining both plain scores (PROPKD and MIC) into six risk categories, we observed additional discriminatory power. This suggests that disease variability is best reflected through an aggregate of both clinical-genetic and imaging parameters, beyond the influence of individual factors alone. For example, in the combined high-high risk class, aCKD occurred approximately 4 years earlier than in the totality of PKD1-T patients, 2 years earlier than in patients with MIC-1D/-1E, and 2 years earlier than in individuals belonging to the PROPKD high risk group alone. Through combined scoring, treatment decisions may also be based on a more reliable prognostication, particularly in the intermediate risk classes. For instance, 105 of 273 intermediate-risk patients (MIC-1C or PROPKD 4–6) were assigned to an intermediate-high class, now qualifying as likely rapid progressors with consideration or indication of specific treatment. Among the intermediate-high risk patients, 40% had already received tolvaptan treatment, leaving 60% with a potential indication. Further research is needed to validate our findings and evaluate the clinical utility of a combined scoring system for decision-making in ADPKD. We propose a stepwise approach to facilitate clinical practice implementation and obtain the best possible risk stratification. We show that having both scores at hand is most important for the intermediate group (notably MIC-1C).²³ We recommend prioritizing the gene score and classification into three gene risk classes when kidney volumetry is not available. In addition, if only sonography is available for imaging, height-adjusted kidney length was recently proposed for predictive assessment together with genotypes.²³ However, CT/MRI data are more accurate and should always be consulted for specific treatment indication.^11,34,35 In general, the clinical parameters contained in the PROPKD score, and the family history should be analyzed and used for the most comprehensive assessment.²³

Our study has certain limitations: first, the limited sample size in different subgroups. For example, only 108 patients had kidney function data for a period longer than 4 years, and only 27 patients were classified as MIC-1A, indicating that MIC-1A was underrepresented, likely for being asymptomatic or undiagnosed. Second, follow-up data on eGFR decline and extra kidney involvement (liver cysts, cerebral aneurysms, other causes of CKD deterioration, such as infections or AKI) were partially incomplete because of the retrospective, nonstandardized data collection across centers. Finally, the patient population was predominantly of European descent lacking ethnicity diversity. Given these limitations, our results should be interpreted with caution. We strongly emphasize the need for larger, multicenter studies, and validation in external cohorts to draw more definitive conclusions.

In summary, the emergence of novel ADPKD treatments has made identifying patients at higher risk for kidney failure increasingly imperative. Our study highlights the potential of combining imaging, genetic testing, and clinical variables for the identification of rapid progressors and demonstrates that, specifically, individuals classified as intermediate risk may benefit from a combined assessment of these parameters.

Supplementary Material

cjasn-20-397-s001.pdf ^{(1.4MB, pdf)}

cjasn-20-397-s002.xlsx ^{(57.9KB, xlsx)}

cjasn-20-397-s003.pdf ^{(2.4MB, pdf)}

Disclosures

Disclosure forms, as provided by each author, are available with the online version of the article at http://links.lww.com/CJN/C135.

Funding

J. Halbritter: Deutsche Forschungsgemeinschaft (HA 6908/4-1, 4-1, HA 6908/7-1, and HA 6908/8-1). E.A.E. Elhassan: Royal College of Surgeons in Ireland.

Author Contributions

Conceptualization: Irene Capelli, Elhussein A.E. Elhassan, Jan Halbritter, Constantin Aaron Wolff.

Data curation: Valeria Aiello, Carolin B. Brigl, Elhussein A.E. Elhassan, Jan Halbritter, Sebastian Kursch, Julia S. Münster, Bernt Popp, Nina Rank, Ria Schönauer, Constantin A. Wolff.

Formal analysis: Valeria Aiello, Elhussein A.E. Elhassan, Jan Halbritter, Nina Rank, Constantin A. Wolff.

Funding acquisition: Irene Capelli, Peter J. Conlon, Jan Halbritter.

Investigation: Katherine Benson, Irene Capelli, Francesca Ciurli, Peter J. Conlon, Amalia Conti, Carlotta Cristalli, Jonathan de Fallois, Elhussein A.E. Elhassan, Jan Halbritter, Sebastian Kursch, Gaetano La Manna, Sarah Lerario, Francesca Montanari, Alexandro Paccapelo, Ria Schönauer, Nicola Sciascia, Marco Seri, Constantin A. Wolff.

Methodology: Valeria Aiello, Elhussein A.E. Elhassan, Jan Halbritter, Constantin A. Wolff.

Project administration: Jan Halbritter, Constantin A. Wolff.

Resources: Valeria Aiello, Jan Halbritter, Constantin A. Wolff.

Software: Jan Halbritter, Bernt Popp, Constantin A. Wolff.

Supervision: Irene Capelli, Kai-Uwe Eckardt, Elhussein A.E. Elhassan, Jan Halbritter.

Validation: Valeria Aiello, Elhussein A.E. Elhassan, Jan Halbritter, Bernt Popp, Constantin A. Wolff.

Visualization: Valeria Aiello, Jan Halbritter, Bernt Popp, Constantin A. Wolff.

Writing – original draft: Valeria Aiello, Jan Halbritter, Constantin A. Wolff.

Writing – review & editing: Valeria Aiello, Carolin B. Brigl, Irene Capelli, Peter J. Conlon, Elhussein A.E. Elhassan, Jan Halbritter, Julia S. Münster, Ria Schönauer, Constantin A. Wolff.

Data Sharing Statement

All data are included in the manuscript and/or supporting information. Decline color. No.

Supplemental Material

This article contains the following supplemental material online at http://links.lww.com/CJN/C133, http://links.lww.com/CJN/C134.

Supplemental Methods. More detailed description of the main methods section.

Supplemental Figure 1. Pie charts summarizing the distribution of sex (A), genotype (B), MIC risk classes: low-risk (1A–1B), intermediate-risk (1C), and high-risk (1D–1E), (C) PROPKD risk classes: low-risk (0–3 points), intermediate-risk (4–6 points), and high-risk (7–9 points), and (D) for the total cohort and the primary endpoint advanced CKD (aCKD) meaning CKD G4/G5.

Supplemental Figure 2. Pie charts summarizing MIC class (A), incidence of aCKD at MIC class (B), PROPKD score (C), and incidence of aCKD at PROPKD score (D) distribution for the total cohort and the primary endpoint advanced CKD meaning CKD G4/G5.

Supplemental Figure 3. Unadjusted Kaplan–Meier kidney survival analysis analyzing sex (A), genotype (B), MIC classes (C), MIC risk (D), PROPKD classes, (E) and PROPKD risk (F), with P values derived from log-rank (Mantel-Cox) tests shown. The mean age at aCKD is 49.3, 54.0, and 64.9 years for PKD1-T, PKD1-NT, and PKD2, respectively (n=171, P < 0.001). (A) The mean age at aCKD is 53.3 years for female patients and 52.1 years for male patients (n=171, P = 0.40). (B) The mean age at aCKD is 56.0, 59.6, 56.0, 52.4, and 42.8 years for MIC-1A–1E (n=170, P < 0.001) with <20% of MIC-1A patients experiencing aCKD compared with 51% of MIC-1E patients (C). Those results were also shown for the three MIC risk classes low, intermediate, and high with mean ages at aCKD of 59.0, 56.0, and 48.0 years (D). The mean age at aCKD is 62.4, 52.3, and 47.5 years for the PROPKD risk classes low, intermediate, and high, respectively (n=170, P < 0.001; E and F).

Supplemental Figure 4. Pie charts summarizing the distribution of the six combined MIC/PROPKD risk stratification groups for the total cohort (A) and the primary endpoint advanced CKD (aCKD), meaning CKD G4/G5 (B).

Supplemental Figure 5. Kidney survival analyses (A) for MIC/PROPKD risk combinations with six different groups (ranging from green to red; n=468) are shown here with the mean age the primary endpoint aCKD was reached for every risk class (B). From the mildest risk class (low-low) to the most severe class (high-high), the mean ages were 67.64, 57.75, 58.17, 53.58, 51.36, and 44.20 years, respectively (P < 0.001). The eGFR measurements at baseline (C) and htTKV values (D) for the six different groups are also shown.

Supplemental Figure 6. Kidney survival analyses (A) for MIC/PROPKD risk combinations with six different groups (colored different from green to red; n=339) are shown here. For the purposes of this analysis, any patients treated with tolvaptan for a period exceeding 1 year were excluded. The mean age for reaching the primary endpoint aCKD for every risk class (B) is displayed.

Supplemental Figure 7. Forest plots showing the hazard ratios (HR) for the MIC and other variables (A), and PROPKD variables, and other (B). Reference values are MIC-1A and PKD2.

Supplemental Table 1. Summary of the baseline characteristics of the study cohort with a comparison of the sexes.

Supplemental Table 2. The correlation and concordance analysis (according to Spearman, Kendall-Tau, and Cohen) between the risk classes of MIC and PROPKD is shown here.

Supplemental Table 3. The correlation and concordance analysis (according to Spearman, Kendall-Tau, and Cohen) between MIC and PROPKD, and after the REVEL score was used to reclassify PKD1 nontruncating variants (PROPKD-R), is shown here.

Supplemental Table 4. Summary of the baseline characteristics of the study cohort classified according to the clinical definition of rapid progression (slow progressor: eGFR slope <3 ml/min per 1.73 m² per year after 2 years; rapid progressor: eGFR slope ≥3 ml/min per 1.73 m² per year after 2 years).

Supplemental Table 5. The descriptive statistical analysis of the six MIC/PROPKD risk combinations, including Age at enrollment, Age at ADPKD diagnosis, Age at primary endpoint aCKD, and eGFR and htTKV values, sorted by risk group, starting with low-low (1), intermediate-low (2), low-high (3), intermediate-intermediate (4), intermediate-high (5), and high-high (6) is shown here.

Supplemental Table 6. Summary of the characteristics of the study cohort stratified into six risk combination groups, which were modeled in both directions by the three risk classes of the MIC and the PROPKD score (low-low/intermediate-low/low-high/intermediate-intermediate/high-intermediate/high-high). For the purposes of this analysis, any patients treated with tolvaptan for a period exceeding 1 year were excluded.

Supplemental Table 7. In this additional excel sheet, all PKD1 and PKD2 germline variants in study participants are listed, together with their ACMG classification and known ClinVar or HGMD variants. Genome Reference Consortium Human Build 38 (Hg38). GnomAD versus 4.0.0; PKD1 NM_001009944.3; PKD2 NM_000297.4.