Integer Cystatin C Values: Impact on Discordance Group Assignment and Accuracy of GFR-Estimating Equations
Year: 2023
Type: article
Abstract: Fu et al.1 recently reported that eGFR based on both creatinine and cystatin C (eGFRcr-cys) outperforms eGFR derived from creatinine (eGFRcr) or cystatin C (eGFRcys) when eGFRcr and eGFRcys are discordant. Importantly, they presented cystatin C values as integers, a less informative format than the typically used real numbers,2,3 especially when moving from zero to one decimal. Use of integers may affect the accuracy of eGFRcys and eGFRcr-cys and thereby the assignment of participants to discordance groups, potentially invalidating conclusions. To explore this, we performed similar analyses to Fu et al.1 using 698 participants (49.8% male) with available data on iothalamate clearance. Mean (SD) age and iothalamate clearance were 57.5 (11.1) years and 81.4 (28.0) ml/min per 1.73 m2, respectively. Median (25th; 75th percentile) serum creatinine and cystatin C were 94.0 (78.3 to 117.0) μmol/L and 1.07 (0.88 to 1.29) mg/L, respectively. It is unclear whether they stratified bootstrapping by discordance groups—a critical step because its omission can impair performance of GFR-estimating equations by changing proportions of participants within discordance groups. We specifically focused our stratified bootstrap analysis on the eGFRcys>eGFRcr discordance group because this group represents participants with above-average muscle mass4 and contained the fewest samples (8%) in the original study.1 Use of cystatin C as integers or values with one or two decimals resulted in respective assignment of 168, 109, and 113 participants to the eGFRcys>eGFRcr discordance group, consistent with the substantial effect of omitting the first decimal of cystatin C values. When using integers or values with one or two decimals, respective median (95% confidence interval) biases for eGFRcys were 4.19 (1.05 to 6.94), −0.45 (−3.48 to 6.13), and −0.49 (−3.60 to 3.56) ml/min per 1.73 m2. For eGFRcr-cys, these biases were −4.06 (−5.51 to −1.29), −6.77 (−10.79 to −2.47), and −6.21 (−10.96 to −2.66) ml/min per 1.73 m2. For eGFRcr, these biases were −16.23 (−17.49 to −14.19), −20.75 (−24.01 to −15.31), and −21.05 (−24.41 to −15.34) ml/min per 1.73 m2. It seems that rounding cystatin C values to integers may substantially affect both the size of discordance groups and the resulting biases, not only of eGFRcys and eGFRcr-cys, with largest differences when moving from zero to one decimal but also for eGFRcr, albeit in the opposite direction. This suggests potential skewing of the conclusions of Fu et al.1 favoring eGFRcr over eGFRcys. If Fu et al. have cystatin C values with at least one decimal, we encourage them to report on the performance of GFR-estimating equations using these values. Alternatively, for comparability, they could use integers for both cystatin C and creatinine values in mg/dl.
Language: English
Subfield: Nephrology
Field: Medicine
Domain: Health Sciences
Open Access status: green
OpenAlex ID: W4388127084