Development and Validation of a Prediction Model for Kidney Failure in Long-Term Survivors of Childhood Cancer

Abstract

PURPOSE

Kidney failure is a rare but serious late effect following treatment for childhood cancer. We developed a model using demographic and treatment characteristics to predict individual risk of kidney failure among 5-year survivors of childhood cancer.

METHODS

Five-year survivors from the Childhood Cancer Survivor Study (CCSS) without history of kidney failure (n = 25,483) were assessed for subsequent kidney failure (ie, dialysis, kidney transplantation, or kidney-related death) by age 40 years. Outcomes were identified by self-report and linkage with the Organ Procurement and Transplantation Network and the National Death Index. A sibling cohort (n = 5,045) served as a comparator. Piecewise exponential models accounting for race/ethnicity, age at diagnosis, nephrectomy, chemotherapy, radiotherapy, congenital genitourinary anomalies, and early-onset hypertension estimated the relationships between potential predictors and kidney failure, using area under the curve (AUC) and concordance (C) statistic to evaluate predictive power. Regression coefficient estimates were converted to integer risk scores. The St Jude Lifetime Cohort Study and the National Wilms Tumor Study served as validation cohorts.

RESULTS

Among CCSS survivors, 204 developed late kidney failure. Prediction models achieved an AUC of 0.65-0.67 and a C-statistic of 0.68-0.69 for kidney failure by age 40 years. Validation cohort AUC and C-statistics were 0.88/0.88 for the St Jude Lifetime Cohort Study (n = 8) and 0.67/0.64 for the National Wilms Tumor Study (n = 91). Risk scores were collapsed to form statistically distinct low- (n = 17,762), moderate- (n = 3,784), and high-risk (n = 716) groups, corresponding to cumulative incidences in CCSS of kidney failure by age 40 years of 0.6% (95% CI, 0.4 to 0.7), 2.1% (95% CI, 1.5 to 2.9), and 7.5% (95% CI, 4.3 to 11.6), respectively, compared with 0.2% (95% CI, 0.1 to 0.5) among siblings.

CONCLUSION

Prediction models accurately identify childhood cancer survivors at low, moderate, and high risk for late kidney failure and may inform screening and interventional strategies.

INTRODUCTION

As therapy for childhood cancers has improved over the decades, the population of long-term survivors is increasing. Survivors remain at significant risk for many treatment-related late effects.^1-3 Development of chronic kidney disease (CKD) or kidney failure is a significant concern for many cancer survivors treated with nephrotoxic chemotherapy, irradiation, or nephrectomy.^1,4-7 Kidney failure, the current preferred term for the entity previously known as end-stage renal disease or end-stage kidney disease, is defined as severe kidney dysfunction or kidney replacement therapy such as dialysis or transplantation.⁸ In the general population, the estimated lifetime risk of kidney failure is 2.8%-4.0%.⁹ Kidney failure carries a significant risk of morbidity and mortality¹⁰ and has a major impact on overall quality of life.¹¹ Despite the importance of understanding long-term complications in childhood cancer survivors, few studies have explored late kidney failure.

CONTEXT

• Key Objective

• Can we identify which childhood cancer survivors are at higher risk for developing kidney failure using individual demographic and treatment information?

• Knowledge Generated

• We developed externally validated prediction models to identify childhood cancer survivors at low, moderate, and high risk for late kidney failure (defined as dialysis, kidney transplantation, or kidney-related death occurring at least 5 years from cancer diagnosis). These models, accompanied by a new online risk calculator, may help guide screening by identifying individuals at higher risk of kidney complications who may benefit from closer monitoring and earlier intervention as indicated.

• Relevance (S. Bhatia)

• These risk prediction models, accompanied by a new online risk calculator, may help guide screening by identifying individuals at higher risk of kidney complications who may benefit from closer monitoring and earlier intervention as indicated.*

• *Relevance section written by JCO Associate Editor Smita Bhatia, MD, MPH, FASCO.

The Childhood Cancer Survivor Study (CCSS) represents one of the largest cohorts of long-term survivors of childhood cancer,^12,13 allowing for evaluation of rare but serious late effects, including kidney failure.¹⁴ Previous CCSS studies have estimated the cumulative incidence of late kidney failure (ie, occurring at least 5 years after cancer diagnosis) between 0.5% and 0.9% at 15 years after diagnosis,^2,15 with a more recent study estimating a cumulative incidence of 1.7% at 35 years after diagnosis.¹⁴ Specific risk factors independently associated with late kidney failure identified in a previous CCSS study include radiotherapy doses of ≥ 15 Gy to the kidney, exposure to ifosfamide or high-dose anthracycline chemotherapy, and nephrectomy.¹⁴ Among unilateral Wilms tumor survivors, the reported cumulative incidence of late kidney failure is 1.3% at 20 years after cancer diagnosis,^4,16 with higher rates of kidney failure observed in survivors with WT1-related genetic syndromes, congenital genitourinary anomalies, and younger age at the time of cancer diagnosis.¹⁶

In this study, we aimed to use the CCSS cohort to develop a model to predict late kidney failure among 5-year survivors of childhood cancers using demographic and readily available cancer treatment information. Previously, models using the CCSS cohort have been developed to predict cardiomyopathy and heart failure,¹⁷ ischemic heart disease and stroke,¹⁸ ovarian failure,¹⁹ and breast cancer²⁰ among childhood cancer survivors. The development of a prediction model for late kidney failure would help identify patients at higher risk for poor outcomes and potentially improve targeted screening and preventive measures in this at-risk population.^1,21

METHODS

Primary Study Population

Previous studies have reported on the patient enrollment and methodology of the CCSS cohort.¹² The cohort consists of 5-year survivors of the most common forms of childhood cancer (leukemia, CNS cancers, Hodgkin lymphoma, non-Hodgkin lymphoma, Wilms tumor, neuroblastoma, soft tissue sarcoma, and bone tumor) diagnosed at age <21 years between 1970 and 1999 and treated at one of the 31 participating North American institutions. To minimize survival bias, proxies completed the questionnaires on behalf of survivors who were deceased at the time of study entry but otherwise eligible for the 5-year survivor cohort. Survivors who developed kidney failure before study eligibility at 5 years from initial cancer diagnosis (n = 126) or for whom radiation doses to the remaining kidney could not be validated (n = 47) were excluded, leaving 25,483 CCSS survivors available for this analysis predicting late kidney failure. A sibling cohort (n = 5,045; Appendix Table A1, online only) served as a comparison population to reflect the risk of developing kidney failure in the general population. The protocol was approved by the human subjects committee at each institution. Participants or their legal guardians provided written informed consent. Treatment factors, including chemotherapy exposures and cumulative doses, surgeries, and radiotherapy were abstracted from medical records. Radiotherapy records were centrally reviewed and mean doses to the right and left kidneys estimated by reconstructing individual radiotherapy fields on a computational phantom scaled to patient's age at the time of treatment.²² If the mean doses to the right and left kidneys differed, the lower of the two were used in analyses.

Outcome Definitions

CCSS participants completed a baseline questionnaire including demographic characteristics and health conditions, and were subsequently followed prospectively with periodic questionnaires.²³ Proxy responses from family members were used for 5-year survivors who were deceased, age <18 years at the time of questionnaire completion, or otherwise unable to complete the questionnaires. Responses were classified and graded using the CCSS' adaptation of the Common Terminology Criteria for Adverse Events (version 4.03).^2,15 For this analysis, the primary outcome was life-threatening (grade 4; requiring dialysis or kidney transplantation) or fatal (grade 5; patient death due to kidney disease) kidney disease. The outcome of kidney transplant was corroborated by linkage with the Organ Procurement and Transplantation Network (OPTN) through December 31, 2013, and conducted for the baseline CCSS cohort living in the United States diagnosed with cancer between 1970 and 1986.²⁴ Of the 52 total kidney transplants self-reported by CCSS participants, 50 were subsequently validated by OPTN,²⁴ suggesting that concordance (C) between self-report and OPTN records is high. The cohort was also linked to the National Death Index to ascertain death due to CKD or kidney failure, with all deaths confirmed through 2017. Of the eligible survivor cohort (n = 25,483), 4,111 survivors died before the end of 2017; among the 204 kidney failure cases, 111 survivors had died. Outcomes were limited to kidney failure occurring at least 5 years since cancer diagnosis and before age 40 years, given the limited number of cohort members and kidney failure events beyond that age. Additional comorbidities, including the presence of congenital genitourinary conditions, hypertension requiring medications, and diabetes requiring medications, were also available from study questionnaires. Physiologic data, such as serum creatinine or cystatin C, urine albumin/protein, or blood pressure, have not been routinely collected as part of the CCSS cohort.

Statistical Analysis

Similar to our previous risk-prediction modeling in CCSS,^17,18 a set of predictors were specified a priori to be potentially included in our prediction models, which consisted of attained age (5-year intervals), sex, race/ethnicity (White/non-Hispanic, Black/non-Hispanic, Hispanic, and other), presence of congenital genitourinary anomalies, age at cancer diagnosis (<10 v ≥10 years), history of nephrectomy within 5 years of diagnosis, exposure to and cumulative dosing of potentially nephrotoxic chemotherapy (ie, ifosfamide, platinum agents, and anthracyclines), radiotherapy exposure to the abdomen and mean kidney dose from radiotherapy, and development of subsequent malignant neoplasm or comorbid conditions including hypertension and diabetes within 5 years of cancer diagnosis.

Two prediction models were developed: a simple model with all predictors categorized as yes/no only, and a dose-specific model with cumulative chemotherapy dosing and mean kidney dose from radiotherapy. Piecewise exponential models with each 5-year attained-age period to have a constant baseline hazard were used with backward selection for model selection^25-27 and estimated the relationships between potential predictors and late kidney failure, allowing for the estimation of absolute rates over time, starting the at-risk period at 5 years from cancer diagnosis and ending at the onset of late kidney failure, death, last questionnaire completion, or age 40 years, whichever occurred first. Consistent with our prior work, regression coefficient estimates were subsequently converted to integer risk scores (relative rate compared with siblings of <1.3, 1.3-1.9, 2.0-2.9, 3.0-4.9 and ≥5.0 corresponding to integer risk scores of 0, 1, 2, 3, and 4, respectively).^17,18 The methodology of the time-dependent area under the curve (AUC) and the C-statistic²⁸ were used to estimate the predictive power for late kidney failure by age 40 years (representing weighted average of respective statistic from study start through age 40 years). For both AUC and C-statistic, values of 0.5 suggest the model does no better than chance, while values approaching 1.0 indicate perfect prediction.²⁹ To address potential response bias, we conducted additional analyses using inverse probability weighting (on the basis of the set of variables available for all eligible survivors; Appendix Tables A2 and A3, online only).

The sum of the integer risk scores (cumulative risk score) was used to develop low-, moderate-, and high-risk groups on the basis of cumulative incidence by age 40 years. The three risk groups (low-, moderate-, and high-risk) were a priori chosen to represent what we felt to be clinically significant risk categories.

External Validation Cohorts

Data from the St Jude Lifetime Cohort Study (SJLIFE, n = 4,708) and the National Wilms Tumor Study (NWTS, n = 6,760) were used to validate the CCSS-based prediction models. The SJLIFE cohort defined outcomes via in-person clinical evaluation supplemented by medical records and death certificates.^30-32 The NWTS, similar to CCSS, accepted patient or family self-report with additional corroboration from medical records and the US Renal Data System.⁴ The NWTS data used for validation were based on a case-cohort design using a random sample of the overall cohort (310 of 6,760) as a subcohort that was used in prior analyses,¹⁷ plus all kidney failure cases (n = 91, with five of these cases also part of the subcohort). For each validation cohort, patients were limited to 5-year survivors without prior kidney failure, and outcomes were restricted to kidney failure (dialysis, kidney transplant, or death due to kidney disease) occurring >5 years after diagnosis. As it was possible for some participants to be part of more than one cohort, we ensured that no participant was counted more than once. Specifically, CCSS participants who overlapped with SJLIFE and/or NWTS were excluded from the validation cohorts, leaving 2,490 and 396 participants for validation, respectively. Time-dependent AUC and C-statistics were estimated for each validation cohort using the CCSS risk score algorithm.

RESULTS

Within the CCSS cohort, 204 individuals of 25,483 survivors (Table 1) experienced late kidney failure, representing a cumulative incidence of 1.0% (95% CI, 0.8 to 1.1) by age 40 years. Median follow-up was 22.2 years (interquartile range, 16.4-29.7 years). In comparison, siblings with median follow-up of 27.0 years (interquartile range, 19.8-34.7 years) had a cumulative incidence of 0.2% (95% CI, 0.1 to 0.5) by age 40 years.

TABLE 1.

Demographic and Clinical Characteristics of 5-Year Survivors of Childhood Cancer From the CCSS (training data set) and the St Jude Lifetime and NWTS (validation data sets)

For the simple model, the following binary predictors of late kidney failure were included in the final model: Black/non-Hispanic race/ethnicity (v all others, combined); age <10 years at cancer diagnosis; history of nephrectomy within 5 years of cancer diagnosis; exposure to ifosfamide, platinum, and anthracycline chemotherapy; history of abdominal radiation; congenital genitourinary anomalies; and onset of hypertension within 5 years of cancer diagnosis (Table 2). For the dose-specific model, cumulative ifosfamide dose and mean kidney dose from radiotherapy were also predictive and included. In both models, early-onset hypertension was a highly influential predictor with a risk score value of 4, exceeding the scores of all other predictors, including high cumulative ifosfamide dosing (≥60 g/m²) and higher mean kidney dose from radiotherapy (≥12 Gy), each with risk score values of 3 (Table 3). Cumulative platinum and anthracycline dose categories did not improve prediction compared with binary exposure history (yes/no) and therefore, dosing for these agents was not included. Early-onset diabetes was also not found to be an influential risk factor and was not included in the final models. The AUC/C-statistics for the CCSS-derived cumulative risk scores by age 40 years were 0.65 (95% CI, 0.62 to 0.69)/0.68 (95% CI, 0.65 to 0.72) for the simple model and 0.67 (95% CI, 0.63 to 0.70)/0.69 (95% CI, 0.64 to 0.72) for the dose-specific model. When we applied the CCSS-based cumulative risk scores to our external validation cohorts, the models performed better in the SJLIFE cohort (eight cases; AUC/C-statistics 0.83 [95% CI, 0.73 to 0.96]/0.86 [95% CI, 0.75 to 0.96] for the simple model and 0.88 [95% CI, 0.72 to 0.98]/0.88 [95% CI, 0.71 to 0.98] in the dose-specific model) and similar to CCSS in the NWTS cohort (91 cases; AUC/C-statistics 0.62 [95% CI, 0.54 to 0.68]/0.63 [95% CI, 0.56 to 0.70] for the simple model and 0.67 [95% CI, 0.61 to 0.77]/0.64 [95% CI, 0.59 to 0.73] in the dose-specific model).

TABLE 2.

Associated Rate Ratios for Predictors of Late Kidney Failure Included in Final Prediction Models

TABLE 3.

Integer Risk Scores Associated With Late Kidney Failure and Corresponding Prediction Model Performance

Cumulative risk scores were subsequently categorized into three groups (low-, medium-, and high-risk) with the intent to create distinct, clinically meaningful groups, on the basis of the corresponding cumulative incidence of kidney failure by age 40 years and its comparison with siblings (Table 4). The low-risk group (score <3) represented 77%-80% of the at-risk population, while the moderate-risk group (score, 3-5) and high-risk group (score ≥ 6) represented 17%-21% and 2%-3% of the population, respectively. The low-risk group had a cumulative incidence of 0.6% (95% CI, 0.4 to 0.7) in both models. The moderate-risk group had a cumulative incidence of 2.3% (95% CI, 1.6 to 3.2) and 2.1 (95% CI, 1.5 to 2.9) for the simple and dose-specific models, respectively. The high-risk group had a cumulative incidence of > 5% by age 40 years (9.4% [95% CI, 4.4 to 16.7] and 7.5% [95% CI, 4.3 to 11.6] in the simple and dose-specific models, respectively). The low-risk survivor group had a significantly greater risk of developing kidney failure compared with siblings (relative risk, 3.2 [95% CI, 1.6 to 6.1] and 3.0 [95% CI, 1.6 to 5.9] for the simple and dose-specific models, respectively; both P < .001). Additionally, the cumulative incidence of kidney failure continued to increase with time, particularly for the moderate- and high-risk groups (Fig 1). When we compared the categorization of the risk groups between the simple and dose-specific models, we found that the two models agreed well with a C rate of 92.0% (Appendix Table A4, online only).

TABLE 4.

Cumulative Incidence and Rate Ratios of Late Kidney Failure by Risk Group for the Simple and Dose-Specific Models Among Participants From the Childhood Cancer Survivor Study

FIG 1.

Cumulative incidence of kidney failure over time by risk group for the CCSS, SJLIFE, and NWTS cohorts (SJLIFE cumulative incidence curves start from age 30 years [>5 years from latest age at diagnosis], while CCSS and NWTS curves start from age 26 years [> 5 years from age 21 years]): (A) CCSS simple model, (B) SJLIFE simple model, (C) NWTS simple model, (D) CCSS dose-specific model, (E) SJLIFE dose-specific model, and (F) NWTS dose-specific model. CCSS, Childhood Cancer Survivor Study; NWTS, National Wilms Tumor Study; SJLIFE, St Jude Lifetime Cohort Study.

DISCUSSION

Survivors of childhood cancer are at risk for long-term kidney dysfunction, among other chronic medical conditions, following treatment.^1-3 Most studies on late kidney failure focus on patients with the same diagnosis (ie, Wilms tumor)^4,16 or provide recommendations on the basis of specific risk factors,²¹ but have not provided individual risk prediction for survivors. Using demographic, treatment, and outcomes data from one of the largest cohorts of childhood cancer survivors, we developed two models to predict individual risk of kidney failure through age 40 years. Although the additional knowledge of mean radiotherapy and chemotherapy dosing resulted in slightly improved prediction, both the dose-specific and simple model with binary exposure variables were able to characterize three distinct risk groups. Additional analyses using inverse probability weighting for the entire eligible cohort did not demonstrate significant differences in our prediction modeling, indicating minimal effect from response bias. Although outcomes in the CCSS cohort were primarily self-reported with potential for misclassification, we demonstrated that the prediction models could be reasonably applied to large external validation cohorts with different methods of ascertaining outcomes, which supports the robustness of our models. Of note, the models performed the best in the SJLIFE cohort, which defined outcomes on the basis of in-person clinical assessment and medical records, and therefore least subject to misclassification.

Compared with previously published prediction models developed for late outcomes in survivors of childhood cancer, our kidney failure models demonstrated comparable discriminatory power. Other prediction models using CCSS reported AUCs/C-statistics ranging from 0.71 to 0.77 for heart failure,¹⁷ 0.63-0.70 for ischemic heart disease and stroke,¹⁸ and 0.78-0.82 for ovarian failure.¹⁹ Our goal was to provide a user-friendly tool to aid patients and clinicians in assessing individual risk of future kidney failure following completion of cancer treatment, especially those at greatest risk for kidney morbidity. Our prediction models, available online as a risk calculator,³³ can readily identify a small group (2%-3% of the total population) at high risk of kidney failure (cumulative incidence of >5% by age 40 years). The cumulative incidence in our high-risk group contrasts greatly with our sibling comparison group, as well as the general US population,⁹ both approximately 0.2% by age 40 years.

Previous studies have explored the demographic and treatment-related factors associated with kidney dysfunction in pediatric cancer survivors.^14,32,34-36 In our prediction model, Black/non-Hispanic race/ethnicity and younger age at cancer diagnosis (<10 years) were associated with late kidney failure. Although the incidence of CKD is 2-3× higher in African American children compared with Caucasian children in the general population,³⁷ further studies are needed in this under-represented group. Younger age at diagnosis has also been associated with late-onset kidney dysfunction.^38-40 Although we did not account for genetic syndromes that are known to be associated with higher risk of kidney disease,^4,16 we included congenital genitourinary conditions as a risk factor that may serve as a proxy for many of these syndromic disorders.

In addition to inherent patient risk factors, various treatment-related exposures also contribute to long-term nephrotoxicity. Nephrectomy is a well-described cause of kidney dysfunction, related to volume loss⁴¹ or hyperfiltration injury to the remaining kidney.^42,43 Long-term NWTS studies report a 20-year cumulative incidence of kidney failure of 0.7% for nonsyndromic Wilms tumor patients following nephrectomy.^16,44 Radiation nephropathy can manifest as hypertension, proteinuria, or kidney insufficiency,^1,7 and risk for CKD is generally higher in patients who received higher doses of radiation.^1,21 Chemotherapy agents, particularly cisplatin, carboplatin, and ifosfamide, have been implicated in long-term kidney dysfunction.^1,7,45-49 Cumulative dose of ifosfamide ≥60 g/m² is associated with higher risk for CKD.²¹ Finally, anthracycline chemotherapy is classically associated with cardiotoxicity^17,18 but was identified as a novel risk factor for late-onset kidney failure in a recent CCSS analysis.¹⁴

Notably, early-onset hypertension (within 5 years of cancer diagnosis) was a more influential predictor for late kidney failure than any other factor in our prediction models. Although hypertension was defined variably across the three study cohorts, this represented an uncommon risk factor overall, <3% of cancer survivors in each cohort. Although we only included early-onset hypertension, given our goal of predicting future kidney failure at the 5-year survival time point, the prevalence of hypertension overall was 14.9% in the CCSS cohort at median age 33.7 years,⁵⁰ and current hypertension was associated with stage 3-5 CKD in a prospective study of pediatric cancer survivors.³² As a potentially modifiable risk factor, improved awareness of hypertension screening and timely treatment may help mitigate late kidney dysfunction in survivors, especially for patients with multiple risk factors.

There are several considerations in the interpretation of our results. Our models were purposefully on the basis of exposures present at the 5-year cancer survival time point, given our goal of developing prediction models anchored at this milestone. Consequently, we cannot account for later-onset hypertension or other risk factors occurring after the prediction time point. We also excluded survivors who developed kidney failure within 5 years of diagnosis; thus, the factors leading to early-onset kidney failure (eg, tumor lysis syndrome and methotrexate toxicity) may differ from the predictors in our models. Newer therapeutic classes of agents, including growth factor inhibitors, checkpoint inhibitors, and chimeric antigen receptor-T cells have been associated with kidney injury,^51-53 and future models will be needed to assess their long-term impact. Our models did not include laboratory-based measures (ie, creatinine or glomerular filtration rate). It is possible that models incorporating those data at the 5-year cancer survival time point, as well as other physiologic data such as blood pressure, could further improve predictive performance. Finally, despite small numbers of kidney failure events in our validation cohorts, the AUC/C-statistic values supported reasonable model performance and discriminatory accuracy.

In summary, using readily available demographic and treatment information, we were able to develop externally validated models that categorized childhood cancer survivors into low-, moderate-, and high-risk groups for developing kidney failure. These prediction models may help guide screening for late kidney dysfunction by identifying higher-risk individuals who may benefit from closer follow-up or earlier intervention.

APPENDIX

TABLE A1.

Demographic Characteristics of Sibling Cohort From the Childhood Cancer Survivor Study

TABLE A2.

Available Demographic and Clinical Characteristics of Included Participants Versus Nonparticipants Eligible for the Childhood Cancer Survivor Study

TABLE A3.

Integer Risk Scores Associated With Late Kidney Failure and Corresponding Prediction Model Performance for All Eligible Survivors Using Inverse-Probability Weighting^a,b

TABLE A4.

Cross-Tabulation of CCSS Participants in Low‐, Medium‐, and High‐Risk Groups for the Simple Versus Dose-Specific Model

AUTHOR CONTRIBUTIONS

Conception and design: Natalie L. Wu, Bryan V. Dieffenbach, Sangeeta Hingorani, John L. Jefferies, Kevin C. Oeffinger, Leslie L. Robison, Melissa M. Hudson, Wendy M. Leisenring, Gregory T. Armstrong, Eric J. Chow

Financial support: Leslie L. Robison, Melissa M. Hudson, Gregory T. Armstrong

Administrative support: Leslie L. Robison, Melissa M. Hudson, Gregory T. Armstrong

Provision of study materials or patients: Gregory T. Armstrong

Collection and assembly of data: Yan Chen, Bryan V. Dieffenbach, Rebecca M. Howell, Leslie L. Robison, Melissa M. Hudson, Wendy M. Leisenring, Gregory T. Armstrong

Data analysis and interpretation: Natalie L. Wu, Yan Chen, Bryan V. Dieffenbach, Matthew J. Ehrhardt, Sangeeta Hingorani, John L. Jefferies, Daniel A. Mulrooney, Kevin C. Oeffinger, Brent R. Weil, Yan Yuan, Yutaka Yasui, Melissa M. Hudson, Gregory T. Armstrong, Eric J. Chow

Manuscript writing: All authors

Final approval of manuscript: All authors

Accountable for all aspects of the work: All authors

AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST

Development and Validation of a Prediction Model for Kidney Failure in Long-Term Survivors of Childhood Cancer

The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated unless otherwise noted. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or ascopubs.org/jco/authors/author-center.

Open Payments is a public database containing information reported by companies about payments made to US-licensed physicians (Open Payments).