A Two-Stage Prediction Strategy for Primary Hyperparathyroidism

Background: Primary hyperparathyroidism parathyroid hormone (PTH) hypersecretion is difficult to recognise when calcium is normal and difficult to distinguish from secondary hyperparathyroidism once PTH is found to be elevated. These are separate problems arising at separate points in care. We developed and evaluated two interpretable models corresponding to these decision points using an internal–external validation design. Methods: We analysed 2788 adults undergoing PTH testing across three campuses of one hospital (development n = 1843; internal–external validation n = 445 and n = 500); those with an estimated glomerular filtration rate below 30 mL/min/1.73 m2 were excluded. PHPT was adjudicated in 504 (18.1%) by two endocrinologists using biochemistry, imaging and histopathology. Model 1, for use before PTH is measured, comprised total calcium, ln(alkaline phosphatase) and age. Model 2, developed in the Main campus subgroup with PTH above the reference limit (>65 pg/mL, n = 429 of 623), comprised ln(PTH), total calcium and phosphate. Results: Model 1 achieved AUCs of 0.920 in the development cohort and 0.882–0.908 in the validation cohorts; because serum calcium contributed to the outcome definition, these estimates reflect agreement with the reference standard rather than independent diagnostic accuracy. Among participants with elevated PTH, Model 2 achieved an AUC of 0.904 in the development cohort and 0.892 in the pooled validation cohort compared with 0.803 and 0.791 for PTH alone, respectively. In the normocalcaemic subgroup, Model 2 achieved an AUC of 0.790 compared with 0.699 for PTH alone. Applied sequentially, the models referred 18.0–26.1% of participants while capturing 78.1–85.0% of PHPT cases and identified a second-stage subgroup with an observed PHPT prevalence of 93.1–95.1%. Conclusions: This two-stage strategy may support further testing and referral for PHPT using routinely available measurements. Prospective validation in broader clinical populations is required before clinical implementation.