Multivariate analysis of clinical, demographic, and laboratory data for classification of disorders of calcium homeostasis

Parathyroid hormone (PTH) nomograms combine total calcium and intact PTH (iPTH) measurements to classify disorders of calcium homeostasis. Our objective was to determine if using a combination of laboratory, demographic, and clinical parameters improves the accuracy of classification of these disorders. Chart data were collected for 236 patients with physician-ordered iPTH and total calcium tests. Classification was done using 3 approaches: (1) PTH nomogram plotting total calcium and iPTH results against known cases; (2) review of all available chart data ("gold standard"); and (3) multivariate model (classification and regression tree [CART] or logistic regression) using 24 variables. The CART model was developed using the gold standard patient classification and validated using leave-one-out cross-validation. The CART model was significantly (P = .002) more accurate (80.6%) than the PTH nomogram (59.7%) and logistic regression (66.2%) at classifying calcium homeostasis disorders. The CART model used 6 of 24 variables (iPTH, calcium, creatinine, renal transplantation, percentage of females, and urea nitrogen) and had a misclassification error rate of 0.194 (27/139). Classification of disorders of calcium homeostasis based on the PTH nomogram can be improved by using the CART model developed in this study.