Introduction: Intensive care unit-acquired Weakness (ICUAW) significantly affects patient prognosis and quality of life. This study aimed to develop a machine learning (ML) model that integrates bedside muscle ultrasound measurements and clinical features for early ICUAW prediction and to identify key predictors that could guide targeted interventions. Methods: This prospective multicenter cohort study enrolled 1203 adult ICU patients. Within 24 hours of ICU admission, we measured the thickness of the rectus femoris (RF) and vastus intermedius (VI) muscles, as well as the RF cross-sectional area (CSA), using bedside ultrasound under both pressurized and unpressurized conditions. Relevant clinical features were recorded concurrently. Nine ML algorithms were trained to develop predictive models incorporating both muscle ultrasound and clinical data. We selected the best-performing model and evaluated its performance on an independent external validation set. Results: Between September 2023 and May 2024, 858 patients from 16 tertiary hospitals comprised the training cohort, and an additional 345 patients (June 2024- April 2025) comprised the external validation cohort. The random forest model demonstrated superior performance, achieving an area under the curve (AUC) of 0.914 (95% CI: 0.891-0.936) and 0.907 (95% CI: 0.882-0.931) for the unpressurized and pressurized models, respectively. In the external validation, the unpressurized model achieved a sensitivity of 0.983 (95% CI: 0.952-0.997) and specificity of 0.726 (95% CI: 0.651-0.792). The pressurized model achieved a sensitivity of 0.978 (95% CI: 0.944-0.994) and a specificity of 0.506 (95% CI: 0.427-0.585). Key predictors identified in the unpressurized model included the Sequential Organ Failure Assessment (SOFA) score, VI or RF muscle thickness, albumin, and fraction of inspired oxygen (FiO2). In the pressurized model, the prominent predictors were SOFA score, VI thickness, Interleukin-6 (IL-6), and creatinine. Conclusions: The random forest-based machine learning model effectively predicted ICUAW using bedside ultrasound and clinical features, demonstrating high accuracy and suggesting potential clinical utility for early risk stratification and intervention.
Zou et al. (Sun,) studied this question.