Machine learning models for predicting readmission after stroke: A systematic review and meta-analysis

Clinical Snapshot

Key Findings

• P Value: Not reported in abstract

• Effect Size: Overall pooled AUROC: 0.74; 30-day readmission pooled AUROC: 0.73; 90-day readmission pooled AUROC: 0.75

• Primary Outcome: Predictive performance of ML models for post-stroke hospital readmission, measured by AUROC

• Nnt Or Sensitivity: AUROC of 0.74 indicates moderate discrimination (conventionally: 0.70–0.79 = acceptable, 0.80–0.89 = excellent). Sensitivity, specificity, PPV, NPV, calibration metrics, and net benefit (decision curve analysis) are not reported. No NNT equivalent calculable from available data.

• Confidence Interval: Overall: 95% CI 0.69–0.78; 30-day: 95% CI 0.67–0.79; 90-day: 95% CI 0.69–0.80

Clinical Application

Clinical implementation of ML readmission prediction models requires electronic health record integration, data governance frameworks, and prospective validation — none of which are addressed in this review. The wide range of input variables (6 to 10,047) across included models makes direct clinical translation impractical without further model standardisation and external validation. A pooled AUROC of 0.74 is insufficient alone to justify deployment without calibration data and impact analyses. Stroke readmission is a monitored quality indicator in Australian hospitals under the Australian Commission on Safety and Quality in Health Care (ACSQHC) and is relevant to the National Stroke Foundation's clinical guidelines. The RACGP and Stroke Foundation Australia emphasise post-discharge care coordination to reduce readmission. Australian hospitals using My Health Record and state-based EHR systems (e.g., NSW Health's eMR) could theoretically support ML model deployment, but TGA regulatory pathways for Software as a Medical Device (SaMD) would apply to any clinically deployed prediction tool. PBS implications are indirect — better readmission prediction could reduce avoidable hospitalisation costs but no cost-effectiveness data are presented. AIHW data indicate stroke remains a leading cause of disability and hospitalisation in Australia, reinforcing the clinical relevance of this research question. Adult patients hospitalised with stroke across diverse healthcare settings. Mean patient age range of 64.5–79.7 years across studies aligns with typical stroke demographics in high-income countries. Applicability to specific stroke subtypes (ischaemic vs. haemorrhagic vs. TIA) is unclear from the available data.

Abstract

BACKGROUND: Hospital readmission following stroke poses a significant challenge for healthcare systems. Machine learning (ML) offers the potential to improve prediction models for readmission risk, surpassing traditional statistical methods. However, the performance of ML models in such context has not been systematically evaluated. We aim to evaluate the performance of ML models in predicting post-stroke hospital readmission, as well as identifying the most important readmission predictors. METHODS: A comprehensive systematic literature review and meta-analysis was conducted.Studies included were those utilizing ML models for stroke readmission prediction. The primary outcome was the predictive performance of ML models for all time stroke readmission as reported by the Area Under the Receiver Operating Characteristic Curve (AUROC). Pooled AUROC values were calculated using a random-effects model. RESULTS: A total of eleven studies involving 380,254 patients were analyzed. Mean age of the patients ranged between 64.5 and 79.7 years. In total, 49 ML models were reported with Logistic regression (LR) and Random Forest being the most used. The number of input predictive variables used for model training ranged from 6 to 10,047. The overall pooled AUROC for predicting hospital readmission was 0.74 (95% CI: 0.69 to 0.78). For 30-day readmission, the pooled AUROC was 0.73 (95% CI: 0.67 to 0.79) while for 90-day readmission, it was 0.75 (95% CI: 0.69 to 0.80). The most frequently reported variables as being top predictors of readmission were Length of stay (LOS) (3 studies), Age (2 studies), National Institutes of Health Stroke Score (NIHSS) (2 studies), HbA1c (2 studies), Homocysteine blood level (2 studies). CONCLUSION: Machine learning models demonstrate moderate predictive performance in stroke readmission risk prediction. Future research should focus on validating and refining existing models and adopting unified methodological approaches to aid in drawing more accurate conclusions.

References

1. 1.Alajlouni, Y., Zayed, Y. S., Nofal, Y., Musleh, A., & Ghannam, M. (2026). Machine learning models for predicting readmission after stroke: A systematic review and meta-analysis. International Journal of Medical Informatics. https://doi.org/10.1016/j.ijmedinf.2026.106484