Answers are grounded in this paper and cite their sources.
Purpose: to develop a predictive model for femoral bone mineral density (BMD) recovery within one year after parathyroidectomy (PTE) in patients with symptomatic primary hyperparathyroidism (PHPT) using machine learning algorithms and three-dimensional modeling parameters of the proximal femur (3D-DXA). Materials and methods. A retrospective single-center study (n=116; 96,6% women; age 57,5 [45,75; 64] years) was conducted at Endocrinology Research Center. The binary outcome was a BMD increase of ≥3% in the total hip (BMD TH) and femoral neck (BMD FN) one year after PTE. Predictive models were built using categorical gradient boosting (CatBoostClassifier). Univariate subgroup comparison analysis was performed for preliminary predictor selection. Model performance was assessed by diagnostic sensitivity (DS), diagnostic specificity (DSp), positive predictive value (PPV), and negative predictive value (NPV). Results. A multi-output model with a single set of predictors and common hyperparameters showed unstable performance: for BMD FN, DS decreased from 78% during cross-validation to 60% on the test set. Individual models with separate predictor sets and 3D-DXA parameters demonstrated substantially better results. The model for BMD FN achieved a PPV of 74±6% and NPV of 74±4% during cross-validation; the model for BMD TH achieved a PPV of 82±6% and NPV of 81±12%. Conclusion. For the first time, predictive models for femoral BMD recovery after PTE in PHPT have been developed based on machine learning algorithms. These models serve as a clinical decision support tool for personalizing postoperative management of PHPT patients: stratifying the risk of insufficient BMD recovery, planning the timing of follow-up densitometry and justifying the prescription of anti-resorptive therapy.
No takes yet. Share an insight, caveat, or question.
Pershina-Miliutina et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: