ECE2023 Eposter Presentations Calcium and Bone (99 abstracts)
1Endocrinology Research Centre, Department of Parathyroid Pathology, Moscow, Russia; 2Endocrinology Research Centre, Director, Moscow, Russia
Background: Primary hyperparathyroidism (PHPT) can present with not only classic complications like bone and kidney pathology but also with different metabolic disorders and cardiovascular diseases. The prevalence of diabetes mellitus type 2 (DM2) in patients with PHPT is higher than in general population but the mechanism of carbohydrate metabolism alteration in PHPT remains unclear.
Aim: The aim of this study is to estimate parameters associated with impaired glucose metabolism in patients with PHPT using a method of machine learningdecision tree.
Material and Methods: The study included clinical data of 307 patients with PHPT. None of the included patients received drugs that affect mineral metabolism. We assessed age, PHPT duration, the presence of prediabetes or DM2, serum calcium, phosphorous, iPTH, osteocalcin levels, BMI. Decision tree (DT) is a classic supervised machine-learning method used for solving classification problems. DT consist of decision nodes and end nodes. Decision nodes contain decision rules, end nodesvalue of target variable with Gini index characterized quality of classification. DT was built in Python using library Scikit-learn. The Bayes theorem was used to determine the conditional probability of impaired glucose metabolism in PHPT.
Results: 55 patients with PHPT had prediabetes/DM2 (17.9%). They were older (60 [55;70] vs 58 [50;66] years, P=0.011), had higher BMI (32.7 [28.1; 39.4] vs 27.2 [24.2; 30.4] kg/m2, P<0.001) and lower serum osteocalcin levels (33.1 [20.8; 51.8] vs 48.1 [34.0; 76.3] ng/ml, <0.001) compared to those without carbohydrate metabolism disorders. Thus, DT was built using these parameters (age, the presence of obesity, osteocalcin level). To form a decision rule, a terminal leaf node with the prevailing class 1 (presence of prediabetes/DM2) and the lowest Gini index were identified. The absence of obesity in combination with osteocalcin >36.8 ng/ml in patients with PHPT was less often associated with impaired glucose metabolism (OR=0.04, 95% CI [0.01-0.17]). Vice versa, obesity with osteocalcin level ≤36.8 ng/ml can be the potential marker of hyperglycemic state in PHPT. The probability of carbohydrate metabolism disorders in such patients is 19.8%.
Conclusion: Patients with PHPT and obesity with osteocalcin level less than 36.8 ng/ml should be considered as the risk group of prediabetes and DM2. In such patients extensive evaluation of carbohydrate metabolism is preferred.