Nurse-led escalation of early warning scores and digital/AI deterioration alerts in adult inpatients: a systematic review and meta-analysis of ward cardiac arrest, unplanned ICU admission, and in-hospital mortality

1. Review title Nurse-led escalation of early warning scores and digital/AI deterioration alerts in adult inpatients: a systematic review and meta-analysis of ward cardiac arrest, unplanned ICU admission, and in-hospital mortality 2. Original language title English). 3. Anticipated or actual start date of the review [2026-01-20] 4. Anticipated completion date [2026-03-05] 5. Stage of review at time of registration · Searches: [Completed] · Screening: [Completed] · Data extraction: [Ongoing] · Risk of bias assessment: [Ongoing] · Data analysis: [Ongoing]6. Named contact Name: [Ugwu Okechukwu Paul-Chima]Email: [ugwuopc@kiu.ac.ug]Institution: [Kampala International University] 7. Review team members and affiliations 1,2Fadia Ahmed Abdelkader Reshia, 3Ugwu Okechukwu Paul-Chima and 4,5Ashiru Muhammad 1Assistant Professor of Medical Surgical Nursing Department, College of Nursing, Jouf University, Saudi Arabia. 2Assistant Professor of Critical Care and Emergency Nursing, Faculty of Nursing, Mansoura University, Mansoura, Egypt. 3Department of Research and Publication Kampala International University Uganda 4Department of Nursing Sciences Bayero University Kano, Nigeria 5Faculty of Nursing Sciences Kampala International University Uganda. 8. Review question What are the effects of nurse-led escalation of early warning scores (manual or digital) and/or digital/AI deterioration alerts on adult inpatient ward outcomes (in-hospital mortality, ward cardiac arrest, unplanned ICU admission/transfer) and on escalation process outcomes (e.g., timeliness, escalation frequency, protocol adherence) where reported? 9. Condition or domain being studied Adult inpatient clinical deterioration on hospital wards, including outcomes such as ward cardiac arrest, unplanned ICU admission/transfer, sepsis progression, length of stay, and mortality. 10. Participants/population Adults (≥18 years) admitted to acute-care hospital wards (medical/surgical and mixed wards). Mixed-population studies are eligible if adult inpatient outcomes are extractable. 11. Intervention(s), exposure(s) Nurse-led escalation strategies, defined as models in which nurses have primary operational responsibility for initiating/delivering/enforcing escalation triggered by:(i) early warning scores (manual or digital) and/or(ii) electronic surveillance or AI/ML-generated deterioration alerts, including nurse-activated pathways embedded within rapid response systems. Intervention groupings (for subgrouping/narrative structure): 1. score-based escalation and protocolisation 2. digitised surveillance and alerting pathways 3. AI/ML-enabled deterioration prediction and alerting 4. response-team activation and escalation timeliness pathways 12. Comparator(s)/control Usual care; pre-intervention baseline; waitlist/attention control; or alternative escalation/monitoring practice. 13. Types of study to be included · Tier 1 (effectiveness): Randomised trials (including cluster and stepped-wedge) and comparative quasi-experimental designs with extractable quantitative data (e.g., controlled before–after, interrupted time-series with appropriate modelling, comparative cohort studies). · Tier 2 (contextual/prediction/association): Prediction-performance studies (AUROC/AUPRC/NRI) and association studies (e.g., escalation delay vs outcomes), synthesised narratively. 14. Context Acute-care hospital ward settings (non-ICU), including systems integrated with rapid response teams. 15. Primary outcome(s) Primary outcomes (hierarchically prespecified): 1. In-hospital mortality 2. Ward cardiac arrest (or in-hospital arrest outside ICU where reported) 3. Unplanned ICU admission/transfer 16. Secondary outcome(s) Time-to-review; rapid response team activation; delay to ICU transfer; hospital length of stay; deterioration/sepsis process outcomes (e.g., time-to-antibiotics) when paired with a clinical endpoint; escalation timeliness/frequency and protocol adherence outcomes where reported. 17. Search strategy and information sources Databases: MEDLINE/PubMed, Embase, CINAHL, Web of Science/Scopus from inception to final search date; no language restrictions; hand-search reference lists of included studies and relevant reviews. Full search strategies to be provided in an appendix. Draft search concepts (example):(nurse-led OR nursing OR nurse-initiated) AND (escalation OR rapid response team OR medical emergency team) AND (early warning score OR NEWS OR MEWS OR electronic observation) AND (clinical deterioration OR cardiac arrest OR ICU transfer OR mortality) AND (machine learning OR artificial intelligence OR predictive alert OR sepsis alert) 18. Data extraction (selection and coding) Two reviewers independently screen titles/abstracts and full texts; disagreements resolved by discussion/third reviewer. Extract: setting, population, intervention characteristics, comparator, implementation supports, follow-up duration, outcome definitions, ascertainment methods, and effect-estimate data. Cluster trials: extract cluster parameters (cluster size, ICC) and whether analyses adjusted for clustering. Missing variables coded as NR and not imputed. 19. Risk of bias (quality) assessment RoB 2 (cluster extension where applicable) for randomised evaluations; ROBINS-I for non-randomised comparative studies. Risk-of-bias judgements inform sensitivity analyses and certainty assessment. 20. Strategy for data synthesis Random-effects meta-analysis when pooling is clinically/methodologically appropriate and ≥2 independent study estimates exist. When k