Integrating Sarcopenia Into ICU-Acquired Weakness Risk Stratification: A Machine Learning–Based Prediction Model for Critical Care

Why It Matters

Integrating sarcopenia into a high‑accuracy prediction model enables clinicians to pinpoint patients at greatest risk of ICU‑AW, allowing targeted preventive measures and more efficient use of intensive‑care resources.

Key Takeaways

• Sarcopenia identified as third‑most important predictor of ICU‑AW
• XGBoost model achieved AUC ≈ 0.84 in both training and validation
• Six variables (age, APACHE II, sarcopenia, sepsis, MV, lactate) drive predictions
• Model outperformed logistic regression (AUC 0.84 vs 0.75)
• Early risk stratification can target nutrition and mobilization resources

Pulse Analysis

Intensive‑care units grapple with ICU‑acquired weakness, a condition that afflicts up to half of critically ill patients and prolongs ventilation, length of stay, and long‑term disability. Traditional risk scores capture severity but often miss underlying muscle reserve, limiting their ability to guide preventive interventions. Recent research highlights the growing importance of body composition metrics, particularly sarcopenia, as a biologically plausible driver of muscle dysfunction in the ICU environment.

By leveraging a large, prospectively collected cohort, investigators applied both LASSO regression and the Boruta algorithm to isolate six robust predictors, then compared ten machine‑learning approaches. XGBoost emerged as the superior model, delivering an area under the curve near 0.84 across training and external validation sets. SHAP analysis confirmed that sarcopenia ranked just behind age and APACHE II, underscoring its independent contribution beyond conventional severity indices. The model’s calibration and decision‑curve analyses suggest it can reliably differentiate high‑risk patients across a wide probability spectrum, outperforming standard logistic regression by a meaningful margin.

For clinicians, the practical implication is clear: early identification of sarcopenic patients enables focused nutrition optimization, aggressive early mobilization, and vigilant infection control—interventions known to mitigate muscle loss. While the reliance on CT‑derived muscle indices may limit bedside applicability, the study paves the way for integrating portable ultrasound or bioelectrical impedance assessments into similar predictive pipelines. Future work should validate the algorithm across diverse ICU settings and explore real‑time decision support tools that translate these insights into actionable care plans, ultimately reducing the burden of ICU‑AW on patients and health systems.