Comprehensive Summary
In their article, Nakano et al. (2025) developed and validated a machine learning (ML) model to predict Intensive Care Unit Acquired Weakness (ICU-AW) in critically ill patients. This retrospective cohort study analyzed clinical strength assessments between 2008-2010, using the Medical Research Council (MRC) scale to diagnose ICU-AW. The primary outcome of this study was the model’s predictive accuracy for ICU-AW using demographic (e.g., BMI, Sex) and clinical variables, including bilirubin, CRP, and PaO2 levels. Model predictions were benchmarked against MCRC-based ICU-AW diagnoses, with performance evaluated using its AUROC, specificity, and sensitivity. In the cohort of 600 patients, the random forest model incorporating all descriptors achieved an AUROC of 0.76, specificity of 62%, and a sensitivity of 79%. Based on the model’s performance, Nakano et al. concluded that the model is well-calibrated for predicting ICU-AW by day 9 in the ICU and could guide early intervention and resource allocation in the ICU.
Outcomes and Implications
ICU-AW is associated with prolonged ICU stays, increased healthcare costs, and worse overall patient outcomes. Currently there is no known effective treatment for ICU-AW, highlighting the importance of early risk identification. Although the MRC scale accurately diagnoses ICU-AW, it requires patient cooperation, which is often limited in critically ill patients. Predicting ICU-AW by day 9 enables clinicians to allocate preventative strategies and rehabilitation resources more efficiently, potentially mitigating morbidity in high-risk patients.