Predictive performance and temporal dynamics of national early warning score 2 (NEWS2) in detecting clinical deterioration in general ward: A large-scale validation study in Singapore.

Background: The National Early Warning Score 2 (NEWS2) is widely used in electronic health records to identify patients at risk of clinical deterioration. However, evidence of its performance in Asian populations and across varying prediction time windows remains limited. Methods: In this retrospective validation study, we analysed 58,809 patient encounters from a 1700-bed tertiary hospital in Singapore. We evaluated NEWS2's discrimination (area under the receiver operating characteristic curve [AUROC]) and calibration for predicting clinical deterioration - defined as death, unanticipated intensive care unit/high dependency (ICU/HD) admission, or code blue activation - within 24 h and across multiple time windows (4 to 72 h), using 100 bootstrap iterations. Results: Event rates were 0.75 % for clinical deterioration, 0.51 % for death, 0.24 % for unanticipated ICU/HD admission and 0.05 % for code blue activation within 24 h. NEWS2 demonstrated strong overall discriminative performance for clinical deterioration (AUROC 0.898, 95 % CI: 0.881-0.915). Discrimination was excellent for mortality (AUROC 0.963, 95 % CI: 0.952-0.974), moderate for code blue activation (AUROC 0.795, 95 % CI: 0.714-0.876) and modest for unanticipated ICU/HD admission (AUROC 0.761, 95 % CI: 0.715-0.808). Calibration was excellent for mortality, but poor for unanticipated ICU/HD admission and code blue activation. Temporal analysis revealed consistent performance across different time windows for mortality prediction, with varying patterns for other outcomes. Conclusions: NEWS2 reliably predicts clinical deterioration and mortality in an Asian tertiary care setting. However, its predictive performance for unanticipated ICU/HD admission and code blue activation remains limited, suggesting the need for further model adjustment to enhance its clinical utility.