Leveraging Machine Learning and Calcium Metabolism Indicators for Predicting Calcium Homeostasis Disorders

Abstract

Many physiological systems depend on calcium homeostasis, which can be disrupted to cause conditions like
hypercalcemia, hypocalcemia, and hyperparathyroidism. Effective management and treatment of many illnesses depend on an early
and precise diagnosis. In this work, we provide a machine learning model that uses key indicators of calcium metabolism, including
Ionized Calcium (Ca++), Total Calcium, and metabolic markers such as Albumin and Parathyroid Hormone (PTH), to predict
calcium-related illnesses. Carefully selected to consider the intricate interactions among important elements, the dataset consists of
clinical notes from individuals with calcium imbalance. We trained and assessed numerous machine learning techniques—including
Gradient Boosting, Random Forests, and Logistic Regression—to identify the top classification model. Our results show that when
ionized and total calcium levels are changed for albumin concentration, the prediction accuracy for identifying hypocalcemia and
hypercalcemia much increases. With 92.3% accuracy, 91.2% precision, and 91.8% recall, its clinical diagnostic performance—the
Random Forest model exceeded all others. Combining biological indicators with machine learning in this work emphasizes the
potential for tailored calcium management, thereby enabling early diagnosis and risk assessment of calcium-related diseases.