Machine Learning-Based Prediction of Elevated PTH Levels Among the US General Population

Area of Science:

• Endocrinology and Metabolism
• Biostatistics and Machine Learning in Health
• Public Health and Epidemiology

Background:

• Elevated parathyroid hormone (PTH) levels are linked to increased mortality risk.
• Current evidence is limited regarding when to measure PTH in the general population.
• Predictive modeling can identify individuals at risk for elevated PTH.

Purpose of the Study:

• To develop a machine learning-based prediction model for elevated PTH levels.
• To utilize demographic, lifestyle, and biochemical data for prediction in US adults.
• To improve early detection of elevated PTH in clinical practice.

Main Methods:

• Population-based study using National Health and Nutrition Examination Survey (NHANES) data (2003-2006).
• Included US adults aged 20+ with serum intact PTH measurements.
• Trained and evaluated six machine learning models (logistic regression, random forest, GBM, SuperLearner) using separate NHANES cohorts.

Main Results:

• 9.2% of 8208 adults had PTH > 74 pg/mL.
• Random forest, GBM, and SuperLearner models achieved the highest predictive accuracy (AUC ≈ 0.79).
• Estimated glomerular filtration rate (eGFR) was the most significant predictor in top-performing models.

Conclusions:

• A machine learning model was successfully developed to predict elevated PTH in a US adult population.
• The model demonstrates potential for accurate and early detection of high PTH in clinical settings.
• Further research is needed to evaluate the clinical utility of this prediction tool for improving health outcomes.