Personalised screening tool for early detection of sarcopenia in stroke patients: a machine learning-based comparative study

Aging Clinical and Experimental Research, Journal Year: 2025, Volume and Issue: 37(1)

Published: Feb. 20, 2025

Sarcopenia is a common complication in patients with stroke, adversely affecting recovery and increasing mortality risk. However, no standardised tool exists for its screening this population. This study aims to identify factors influencing sarcopenia develop risk prediction model evaluate predictive performance. Data from 794 stroke were analysed assess demographic clinical characteristics. Variable selection was performed using least absolute shrinkage operator (LASSO) regression, followed by multivariate regression analysis. Logistic (LR), random forest (RF) XGBoost algorithms used construct models, the optimal subjected external validation. Internal validation conducted via bootstrap resampling, involved an additional cohort of 159 stroke. Model performance assessed area under curve (AUC), calibration curves decision analysis (DCA). Seven variables identified through LASSO The LR achieved highest AUC (0.805), outperforming RF (0.796) (0.780) models. Additionally, exhibited superior accuracy, precision, recall, specificity F1-score. External confirmed model's robustness, 0.816. Calibration DCA demonstrated their accuracy applicability. A model, presented as nomogram online calculator, developed Early may facilitate timely interventions improve patient outcomes.

Language: Английский

Causal Relationship Between Blood Metabolites and Osteoporosis: A Two-Sample Mendelian Randomization and Genetic Correlation Analysis

Bioengineering, Journal Year: 2025, Volume and Issue: 12(5), P. 435 - 435

Published: April 22, 2025

Background: Osteoporosis (OP) is a systemic bone disease often undiagnosed until fractures occur. Metabolites may influence OP, offering potential biomarkers or therapeutic targets. This study investigates the causal relationship between circulating metabolites and OP-related phenotypes using Mendelian Randomization (MR). Methods: GWAS data on 233 metabolic traits from 136,016 participants were analyzed through two-sample MR. Linkage disequilibrium score regression (LDCS) was used to estimate genetic correlations phenotypes, leveraging European ancestry linkage scores account for polygenicity stratification. MR employed inverse-variance weighted (IVW) method, with sensitivity analyses via MR-Egger, MR-PRESSO, median methods address pleiotropy confounders. Results: LDCS identified significant mineral density (BMD) total body BMD (toBMD) showing strongest associations. Thirty-five metabolite traits, including apolipoprotein A-I, exhibited linkages. Among 79 influencing BMD, serum acetate levels significantly associated femoral neck (OR: 1.28, 95% CI: 1.02–1.62), lumbar spine 1.73, 1.32–2.27), 1.21, 1.04–1.42). Creatinine consistently linked reduced 0.88, 0.79–0.99). Triglycerides in IDL VLDL particles also contributed variation. Conclusions: Significant relationships observed specific highlighting key as of health. These findings enhance understanding OP pathogenesis suggest future preventive strategies.

Language: Английский

Personalized Screening Tool for Early Detection of Sarcopenia in Stroke Patients: A Machine Learning-Based Comparative Study

Research Square (Research Square), Journal Year: 2024, Volume and Issue: unknown

Published: Nov. 20, 2024

Background Sarcopenia often occurs in stroke patients and contributes to worse recovery a higher risk of death. There is no standardized tool for screening sarcopenia patients. The objective this study explore the factors influencing patients, develop prediction model, evaluate its predictive accuracy. Methods Demographic clinical characteristics 794 were collected. LASSO regression analysis was used variable selection, selected variables analyzed using multivariate regression. Logistic Regression (LR), Random Forest (RF), XGBoost construct models, with optimal model external validation. Bootstrap resampling internal validation training cohort, another 159 collected performance models evaluated AUC, calibration curve, Decision Curve Analysis (DCA). Results Based on logistic analysis, seven selected. AUC value LR 0.805, surpassing that RF (0.796) (0.780). also outperformed terms accuracy, precision, recall, specificity, F1-score. In validation, achieved an 0.816, curve along DCA demonstrated has nice accuracy applicability. Conclusions study, we developed presented it as nomogram detect such early may benefit these

Language: Английский