Research Square · Mar 19, 2026
Sarcopenia as an Independent Predictor of Injurious Falls in Postmenopausal Women with Type 2 Diabetes Mellitus: A One Year Prospective Cohort Study with Machine Learning Risk Stratification and mHealth Educational Intervention from a Tertiary Care Center in New Delhi, India
By Raut J
Background Postmenopausal women with type 2 diabetes mellitus (T2DM) face substantially elevated risk of sarcopenia and injurious falls through converging mechanisms of insulin resistance, neuromuscular atrophy, and chronic hyperglycaemia. Despite this, concurrent assessment of sarcopenia and fall risk in T2DM clinical practice remains uncommon, and validated risk stratification tools tailored to this population are absent. Objective To determine whether sarcopenia independently predicts incident injurious falls over 12 months in postmenopausal women with T2DM, to construct and validate an interpretable machine learning fall risk model, and to evaluate the efficacy of a multilingual mHealth fall prevention educational intervention. Methodology: A one year prospective cohort study enrolled 312 postmenopausal women with T2DM aged 50 to 75 years at the Endocrinology and Diabetes Clinic of All India Institute of Medical Sciences (AIIMS), New Delhi, India. Sarcopenia was diagnosed using Asian Working Group for Sarcopenia 2019 (AWGS 2019) criteria combining dual energy X ray absorptiometry (DXA), dynamometry, and 6 metre gait speed. Fall events were ascertained through computer vision based clinic monitoring and monthly community telephone surveillance. An interpretable XGBoost ensemble model incorporating SHAP feature attribution was trained on 14 baseline clinical variables. A multilingual AI educational intervention was delivered in five languages at 3, 6, and 9 months. Results Sarcopenia was identified in 118 of 312 participants (37.8%). Over 12 months, 94 injurious falls occurred in 71 participants (23.7%). In multivariable Cox regression, sarcopenia was the strongest independent predictor of falls (adjusted hazard ratio 2.84; 95% CI 1.91 to 4.22; p < 0.001). The machine learning model achieved AUC 0.883 (95% CI 0.841 to 0.922), significantly outperforming clinical scoring (AUC 0.714; p < 0.001). The multilingual educational intervention reduced fall hazard by 34% versus standard care (aHR 0.66; 95% CI 0.47 to 0.92; p = 0.014). Conclusions Sarcopenia nearly triples injurious fall risk in postmenopausal women with T2DM at a New Delhi tertiary centre. Interpretable machine learning risk stratification and multilingual digital education significantly improve clinical fall prevention outcomes. Routine sarcopenia screening and ML assisted risk assessment should be integrated into T2DM care pathways for postmenopausal women.