New equations to estimate 24-h urinary creatinine excretion

Abstract

We aimed to develop and validate prediction equations to estimate the 24-h creatinine excretion in Brazilian adults. We executed cross-sectional randomly distributed measurements in 330 adults from Vitória, a south-western Brazilian city. Anthropometry, blood pressure, and blood analytes were measured on the same day. The 24-h urine collection was initiated at the clinic and ended on the next day at home. Completeness of the 24-h collection period was confirmed by personal records in a diary. We considered valid 24-h urine collections for a period between 23 and 25 h, no report of urine loss, and volume > 500 ml. Four equations: All together with weight, height and age (WHA) and weight, height, age and lean mass (WHAL), and split by sex were developed using Least absolute shrinkage and selection operator (LASSO) regression with tenfold cross-validation iterated 50 times. Root Mean Square Error (RMSE), Mean Absolute Percentage Error (MAPE), Pearson correlation (r), Lin's Concordance Correlation Coefficient (CCC), Coefficient of Determination (R2), Median Absolute Error (MedAE), mean bias, and 95% limits of agreement (95% LOA) were used to assess their performance. From 269 participants (54% women, mean age 44.1 ± 14 years) the 24-h urine volume was 1766 ± 744 mL containing 1331 ± 399 mg creatinine (higher in men than in women; 1608 ± 337.1 vs. 1088 ± 268 mg; p < 0.01). The predicted equations by sex presented stronger correlation (r = 0.77-0.80), the lower values of RMSE (263.8-248.4) and the higher CCC (0.74-0.78). The equations including weight, age, height and lean mass were those that presented the best performance metrics. However, the equations split by sex including age, height and weight sub estimated 24-h creatinine excretion by only 0.10 mg/day, in average, with lower and upper 95% LOA - 517.2 and 517.0 mg/day, respectively. Since the developed equations showed accuracy and outperformed previous ones, we recommend their use according to variables availability.

Keywords: Equations development; Lasso; Urinary creatinine excretion; Urine collection; Validation study.

Fig. 1

Regularization of a Lasso regression model showing the effects of Lambda on model performance and coefficient stability, evaluated through tenfold CV iterated 50 times. (a) Lambda versus NRMSE curve. The solid blue line represents the mean NRMSE across CV iterations. (b) Plot of coefficient values for each predictor variable as a function of Lambda. Each colored line represents the mean coefficient value for a specific variable. For both, the dark shaded area corresponds to ± 1SD, and the light shaded area represents the minimum and maximum values observed across the iterations.

Fig. 2

Relationship between (a) the Kawasaki’s, (b) Tanaka’s, (c) All-WHA, (d) All-WHAL, split by Sex (e) All-WHA and (f) All-WHAL equations and 24-h creatinine excretion in Brazilian adults. Black dots represent male and the squares represent female.

Fig. 3

Bland–Altman plots of the difference between (a) the Kawasaki’s, (b) Tanaka’s, (c) All-WHA, (d) All-WHAL, split by Sex (e) All-WHA and (f) All-WHAL equations and 24-h creatinine excretion in Brazilian adults.