Deep learning models for ICU readmission prediction: a systematic review and meta-analysis

Critical Care Koumantakis et al. Critical Care (2025) 29:442 https://doi.org/10.1186/s13054-025-05642-x Open Access RESEARCH Deep learning models for ICU readmission prediction: a systematic review and metaanalysis Emanuele Koumantakis1 , Konstantina Remoundou2,3 , Nicoletta Colombi4 , Carmen Fava5 , Ioanna Roussaki2,3 , Alessia Visconti1*† and Paola Berchialla1† Abstract Background Intensive Care Unit (ICU) readmissions are associated with increased morbidity, mortality, and healthcare costs. Therefore, determining an appropriate timing of ICU discharge is critical. In this context, deep learning (DL) approaches have attracted significant attention. Methods We conducted a systematic review of studies developing or validating DL models for ICU readmission prediction, published up to March 4th, 2025, and indexed in PubMed, Embase, Scopus, and Web of Science. We summarised them along multiple dimensions, including outcome and population definition, DL architecture, reproducibility, generalizability, and explainability, and provided a meta-analytic estimate of model performance. Results We included 24 studies encompassing 49 DL models, predominantly trained on US-based datasets, and rarely subjected to external validation. There was considerable variability across study settings, including the definition and timeframe of the ICU readmission outcome, as well as DL architecture used, alongside a substantial risk of bias. Technical reproducibility and model interpretation were rare. A meta-analysis of AUROC values from 11 studies yielded a mean of 0.78 (95% CI = 0.72–0.84), with very high heterogeneity (I2 = 99.9%). Models targeting diseasespecific ICU subpopulations achieved significantly higher performance (mean AUROC = 0.92, 95% CI = 0.89–0.95, p = 0.002), and substantially lower heterogeneity (I2 = 17.1%). Conclusions DL models showed promising performances in predicting ICU readmissions, but exhibited several shortcomings, including low reproducibility, over-reliance on a few US-based datasets, and limited explainability. Additionally, the high heterogeneity and risk of bias limited our ability to assess their pooled performance through meta-analysis. Taken together, our observations suggest that the quality of the evidence regarding the application of DL approaches to ICU readmission prediction is poor, thus hindering their clinical applicability. Keywords ICU readmission, Deep learning, Risk prediction, Systematic review. † Alessia Visconti and Paola Berchialla jointly supervised this work. *Correspondence: Alessia Visconti 1 Center for Biostatistics, Epidemiology, and Public Health, Department of Clinical and Biological Sciences, University of Turin, Turin, Italy 2 School of Electrical and Computer Engineering, National Technical University of Athens, Athens, Greece 3 Institute of Communication and Computer Systems, Athens, Greece 4 Federated Library of Medicine F. Rossi, University of Turin, Turin, Italy 5 Department of Clinical and Biological Sciences, University of Turin, Turin, Italy © The Author(s) 2025. Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creati vecommons.org/licenses/by-nc-nd/4.0/. Koumantakis et al. Critical Care (2025) 29:442 Background The decision to discharge a patient from the intensive care unit (ICU) is a complex process. Discharging a patient too early increases their risk of adverse outcomes, particularly ICU readmission and mortality, whereas a delayed discharge contributes to higher healthcare costs [1]. Given that ICU readmissions are themselves associated with a more than double mortality risk [2–4], extended hospital stays [2, 4], greater morbidity [2, 4], and the use of more resources [1, 3], identifying patients at high risk of ICU readmission is crucial [1, 3]. Numerous attempts have been made to develop models for risk of ICU readmission [5]. One of the earliest is the Stability and Workload Index for Transfer (SWIFT), a multivariable logistic regression model estimating readmission risk based on the original source of ICU admission, ICU length of stay, and patient stability [6]. Others are the Acute Physiology and Chronic Health Evaluation (APACHE) II and III scores [7, 8], and the Simplified Acute Physiology Score (SAPS) II [9]. However, the clinical utility of these models is limited by insufficient evidence regarding their performance, potential biases, and overall lack of effectiveness in real-world settings [5]. Additionally, they depend on strict statistical assumptions, such as specific error distributions, additive effects in linear models, and proportional hazards, which are often violated in real-life scenarios [10], and are unsuited for high-dimensional multimodal datasets [11]. Machine learning (ML) methods are better equipped to handle complex interactions and to uncover intricate relationships between numerous risk factors [12]. Among those, deep learning (DL) models have gained significant attention in healthcare, especially in the analysis of medical images [13–16], and in the prediction of prolonged hospital stay and readmissions [17]. However, there is a significant gap in the literature regarding the application and validation of DL models for ICU readmission prediction. Published systematic reviews primarily focus on traditional statistical models or broader ML techniques, and typically overlook the unique capabilities of DL approaches [5, 18, 19]. Here, we perform a systematic review and comprehensive meta-analysis on the current state-of-the-art DL-based methods. We assess their predictive performance, present their associated challenges, and discuss current and future prospects. Our work synthesizes the current status of the literature and critically appraises the quality of existing studies, thus aiding clinicians in making informed decision regarding the application of such approaches in critical-care settings. Page 2 of 13 Methods Data sources, search strategy, and study selection We performed a systematic review of studies describing DL models for the prediction of ICU r (...truncated)