New Model for Estimating Glomerular Filtration Rate in Patients With Cancer

Abstract

Purpose The glomerular filtration rate (GFR) is essential for carboplatin chemotherapy dosing; however, the best method to estimate GFR in patients with cancer is unknown. We identify the most accurate and least biased method. Methods We obtained data on age, sex, height, weight, serum creatinine concentrations, and results for GFR from chromium-51 (⁵¹Cr) EDTA excretion measurements (⁵¹Cr-EDTA GFR) from white patients ≥ 18 years of age with histologically confirmed cancer diagnoses at the Cambridge University Hospital NHS Trust, United Kingdom. We developed a new multivariable linear model for GFR using statistical regression analysis. ⁵¹Cr-EDTA GFR was compared with the estimated GFR (eGFR) from seven published models and our new model, using the statistics root-mean-squared-error (RMSE) and median residual and on an internal and external validation data set. We performed a comparison of carboplatin dosing accuracy on the basis of an absolute percentage error > 20%. Results Between August 2006 and January 2013, data from 2,471 patients were obtained. The new model improved the eGFR accuracy (RMSE, 15.00 mL/min; 95% CI, 14.12 to 16.00 mL/min) compared with all published models. Body surface area (BSA)-adjusted chronic kidney disease epidemiology (CKD-EPI) was the most accurate published model for eGFR (RMSE, 16.30 mL/min; 95% CI, 15.34 to 17.38 mL/min) for the internal validation set. Importantly, the new model reduced the fraction of patients with a carboplatin dose absolute percentage error > 20% to 14.17% in contrast to 18.62% for the BSA-adjusted CKD-EPI and 25.51% for the Cockcroft-Gault formula. The results were externally validated. Conclusion In a large data set from patients with cancer, BSA-adjusted CKD-EPI is the most accurate published model to predict GFR. The new model improves this estimation and may present a new standard of care.

Fig 1.

Schematic representation of study workflow. eGFR, estimated glomerular filtration rate.

Fig 2.

Bland-Altman plots of estimated GFR (eGFR) and measured GFR (mGFR) for the new model and each of the published models. The mean of mGFR and eGFR was plotted against the difference of the two for the internal validation data set. Positive differences indicate underestimation and negative differences indicate overestimation. The plots are ordered in ascending order of root-mean-squared error of eGFR from top left to bottom right. The solid line on each plot represents the mean of the difference and the dashed lines are drawn at the mean ± 1.96 times the standard deviation of the difference. Points are colored by sex (blue, female; gold, male). Adj, adjusted; CKD-EPI, chronic kidney disease epidemiology; MDRD, Modification of Diet in Renal Disease.

Fig 3.

Graphical illustrations of statistics used to compare the new model and published models. Box plots of the residuals (measured glomerular filtration rate [GFR] minus estimated GFR) for all published models and the new model using (A) the internal validation data set and (B) the external validation data set are shown. Notches delineate an approximate 95% CI for the median residual, calculated as ± 1.58*interquartile range/n^0.5. A positive or negative value for the median residual indicates underestimation or overestimation bias, respectively. (C) Graphical illustration of GFR root-mean-squared error (RMSE) in the internal and external validation data sets. Error bars describe the 95% CI on the basis of the χ² distribution for the calculated RMSE. (D) Graphical illustration of percentage of patients with a carboplatin dosing absolute percentage error (APE) > 20% in the internal and external validation data sets. For all plots, blue represents the internal validation set and gold represents the external validation data set. Adj, adjusted; CKD-EPI, chronic kidney disease epidemiology; MDRD, Modification of Diet in Renal Disease.

Fig 4.

Predictive CIs for glomerular filtration rate (GFR) of each patient in the internal validation data set. To obtain this figure, the new model fitted on the development data set was applied to all patients in the internal validation data set. The measured GFR (gold points) and the estimated GFR (blue points) for each patient are illustrated. Each horizontal line represents a 95% predictive CI for the patient, with patients ordered in accession by their estimated GFR. The vertical dashed line highlights the boundary at a GFR of 50 mL/min, below which cisplatin administration would be considered with caution by most clinicians. Of the 494 patients in the internal validation data set, 24 (4.9%) had measured values outside their prediction interval.