Challenges in improving the equation for estimating 24-h urinary sodium excretion from casual urine in hypertensive patients taking antihypertensive drugs: addressing overestimation, especially at low sodium excretion levels.

The estimation formula by Tanaka et al. for predicting the 24-h urinary sodium (Na) excretion (24Na) from a single causal urine sample is widely used. However, it overestimates values in the low 24Na range. We aimed to develop a formula to improve the accuracy, particularly for samples with 24Na < 2 g/day. Stored data from 187 hypertensive patients (mean age, 66.1 years; 56.7% female) who underwent both 24-h home urine collection and a fasting morning causal urine test the following day were analyzed. We used a machine learning approach to extract conditional branches based on the threshold relationships among the variables. The proposed estimation formula was constructed by adding a correction term to the Na/Creatinine(Cr) ratio in Tanaka's formula and the modified formula was applied to each conditional branch. The correction terms included body mass index (BMI), age, and concentration of causal urine Na and were applied in different forms according to each branch. Compared with the Tanaka method, our method improved the agreement rate by ~25% and reduced the disagreement rate by 25% in samples with 24Na < 2 g/day. The correlation coefficient was higher (Ours: 0.51, Tanaka: 0.29), the range of error with 24Na was narrower (Ours: 4.89, Tanaka: 5.69), and the percentage of absolute errors for <1 g improved by 9.8%. Although developed from a specific dataset, our formula is useful for low-24Na samples prone to misestimation by the conventional formula and may improve the accuracy of dietary salt intake assessments from causal urine.