Key result
Machine learning outperforms scoring systems in predicting unplanned ICU readmission with ~71% sensitivity.
Why the study?
Intensive care unit readmission is critical for discharge timing and transitional care, but a systematic review was needed to assess the performance and applicability of various predictive models.
Do machine learning models improve the prediction of unplanned ICU readmission compared to scoring systems?
Meta-Analysis
Do machine learning models improve the prediction of unplanned ICU readmission compared to scoring systems?
Absolute Event Rate: 0.711% vs 0.607%
Machine learning models outperform traditional scoring systems in predicting unplanned ICU readmission, though current models have high risk of bias and lack external validation.
May improve ICU readmission prediction; leaves open external validation amid high bias risk.
BACKGROUND: Intensive care unit (ICU) readmission is a critical factor in determining discharge timing and transitional care and is predicted by various models using different approaches. A systematic review is needed to assess the performance and applicability of these models. AIM: To identify prognostic models for unplanned ICU readmission and compare the performance of machine learning models with scoring systems. STUDY DESIGN: This is a systematic review and meta-analysis. We searched 11 databases up to August 21, 2024 for cohort studies on ICU readmission prediction models. The Prediction Model Risk of Bias Assessment Tool assessed model applicability and risk of bias, and meta-analysis was performed using the Hierarchical Summary Receiver Operating Characteristic Curve model in Stata 16.0. RESULTS: Of 2150 articles, 67 were included, describing 335 models and 67 scoring systems. Common predictors included mechanical ventilation, age, blood pressure, gender and heart rate. The meta-analysis of 199 models showed pooled sensitivities of 0.607 for scoring systems and 0.711 for machine learning models, with specificities of 0.699 and 0.899, respectively. Deep learning models had higher sensitivity (0.745) but lower specificity (0.709). All studies had a high risk of bias. CONCLUSIONS: Machine learning outperformed scoring systems but ignored clinical notes. Including unstructured text could improve predictions. Models need external validation to ensure reliability across institutions. RELEVANCE TO CLINICAL PRACTICE: Models for ICU readmission prediction will aid critical care nurses in identifying high-risk patients and prioritizing post-ICU care needs. This can support nurse-led interventions, improve patient safety and optimize resource allocation for transitional care.
No takes yet. Share an insight, caveat, or question.
Hu et al. (2025) conducted a meta-analysis in Unplanned intensive care unit (ICU) readmission. Machine learning models vs. Scoring systems was evaluated on Pooled sensitivity for predicting unplanned ICU readmission. Machine learning models outperformed scoring systems for predicting unplanned ICU readmission, demonstrating higher pooled sensitivity (0.711 vs 0.607) and specificity (0.899 vs 0.699).
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: