BACKGROUND Circulating calcium mainly carries out its physiologic function in its ionized form (iCa). Clinically, iCa is usually estimated by multiplying the total calcium (TCa) level by 0.5 in the general population, but this method is not accurate when applied to patients on long-term hemodialysis (CHD). Accordingly, this study aimed to develop a predictive function for iCa in patients on CHD by incorporating TCa and other additional variables. MATERIAL AND METHODS This was a retrospective cross-sectional study consisting of 2 cross-sectional datasets: a derivation set including 469 CHD patients in June 2019, and a validation set including 446 CHD patients in September 2019. The derivation set's data were analyzed using the stepwise model selection of machine learning with 10-fold cross-validation to develop a predictive function for iCa. This predictive function was then applied to the validation set's data, and the predictive function's estimated iCa was compared with the actual laboratory iCa by using the paired-samples t test and intraclass correlation coefficient. RESULTS After analyzing the routine laboratory data parameters of patients in the derivation set, the following 5 variables were included in the predictive function of iCa: blood urea nitrogen, creatinine, phosphate, TCa, and albumin. This predictive function was applied to the validation set to yield an estimated iCa level that was not significantly different from the laboratory-measured iCa level of the validation dataset (P=0.676) with an excellent ICC of 0.905. CONCLUSIONS We developed a new predictive function that accurately measures the iCa in patients on CHD by using routine laboratory data.