A nomogram integrating machine learning with clinical predictors for osteosarcopenia risk prediction in type 2 diabetes mellitus.
Type 2 diabetes mellitus (T2DM) predisposes patients to osteosarcopenia, a debilitating condition characterized by concurrent bone loss and muscle wasting. This study aimed to develop and internally validate a nomogram for predicting osteosarcopenia risk in T2DM patients aged ≥ 40 years.
The test cohort included 5,412 hospitalized T2DM patients (January 2010-July 2024), and the temporal validation cohort included 1,671 patients (August 2024-December 2025) from the First Affiliated Hospital of Fujian Medical University. Logistic regression and machine learning algorithms (Boruta, random forest, LASSO) were combined for feature selection. The nomogram was constructed via multivariable logistic regression. We carried out receiver operating characteristic (ROC) curve analysis, calibration, decision curve analysis (DCA), and bootstrap validation for assessing the nomogram. Restricted cubic splines were employed for exploring potential nonlinear associations.
Eight independent predictors, which encompassed gender, age, BMI, WHtR, fracture history, diabetic foot ulcer (DFU), smoking status, and diabetic kidney disease (DKD), were identified. These predictors were incorporated into the nomogram. The nomogram achieved AUCs of 0.864 and 0.904 in the test cohort and validation cohort, respectively. Accordingly, favorable calibration and positive net benefit on DCA was demonstrated. Higher BMI served as a protective factor (OR = 0.56, 95% CI: 0.53-0.59). Besides, higher WHtR acted as a risk factor (OR = 1.47, 95% CI: 1.28-1.69). Restricted cubic spline analysis revealed a significant negative nonlinear relationship between BMI and osteosarcopenia risk, and a significant positive nonlinear relationship between WHtR and osteosarcopenia risk.
This nomogram, based on eight readily available clinical variables, exhibits excellent discriminative performance and clinical utility for predicting osteosarcopenia risk in T2DM patients aged ≥ 40 years. Further multicenter external validation is warranted.
The test cohort included 5,412 hospitalized T2DM patients (January 2010-July 2024), and the temporal validation cohort included 1,671 patients (August 2024-December 2025) from the First Affiliated Hospital of Fujian Medical University. Logistic regression and machine learning algorithms (Boruta, random forest, LASSO) were combined for feature selection. The nomogram was constructed via multivariable logistic regression. We carried out receiver operating characteristic (ROC) curve analysis, calibration, decision curve analysis (DCA), and bootstrap validation for assessing the nomogram. Restricted cubic splines were employed for exploring potential nonlinear associations.
Eight independent predictors, which encompassed gender, age, BMI, WHtR, fracture history, diabetic foot ulcer (DFU), smoking status, and diabetic kidney disease (DKD), were identified. These predictors were incorporated into the nomogram. The nomogram achieved AUCs of 0.864 and 0.904 in the test cohort and validation cohort, respectively. Accordingly, favorable calibration and positive net benefit on DCA was demonstrated. Higher BMI served as a protective factor (OR = 0.56, 95% CI: 0.53-0.59). Besides, higher WHtR acted as a risk factor (OR = 1.47, 95% CI: 1.28-1.69). Restricted cubic spline analysis revealed a significant negative nonlinear relationship between BMI and osteosarcopenia risk, and a significant positive nonlinear relationship between WHtR and osteosarcopenia risk.
This nomogram, based on eight readily available clinical variables, exhibits excellent discriminative performance and clinical utility for predicting osteosarcopenia risk in T2DM patients aged ≥ 40 years. Further multicenter external validation is warranted.