Predictive Modeling for Readmission to Intensive Care: A Systematic Review

Area of Science:

• Critical Care Medicine
• Health Informatics
• Biostatistics

Background:

• Intensive care unit (ICU) readmissions pose significant challenges to patient outcomes and healthcare resource allocation.
• Developing accurate prediction models for ICU readmissions is crucial for effective patient management and resource planning.

Purpose of the Study:

• To evaluate the methodological rigor and predictive performance of existing ICU readmission prediction models.
• To identify characteristics of optimal prediction models for ICU readmissions.
• To explore the link between appropriate triage decisions and patient outcomes.

Main Methods:

• Systematic literature search across PubMed, Web of Science, Cochrane, and Embase databases for studies published between 2010 and 2021.
• Independent data extraction and bias assessment using the Critical Appraisal and Data Extraction for Systematic Reviews of Prediction Modelling Studies (CRD) checklist and the Prediction model Risk Of Bias ASsessment Tool (PROBAST).
• Critical evaluation of data sources, modeling techniques, outcome definitions, performance metrics, and risk of bias.

Main Results:

• Thirty-three studies on ICU readmission prediction models were included; six had a high risk of bias, and four had an unclear risk of bias.
• Common sources of bias included univariate analysis for predictor selection (50% of studies).
• Higher-performing models utilized homogenous patient populations, clearly defined outcomes, and routinely collected, time-analyzed predictors, outperforming models relying solely on existing clinical scores.

Conclusions:

• Models predicting ICU readmissions achieve superior performance through longitudinal time series modeling.
• The use of homogenous patient populations and tailored predictor variables significantly enhances the accuracy and utility of ICU readmission prediction models.