Delta MEWS combined with SpO2 for identifying MEWS-based high-alert status at unplanned transfer to the emergency resuscitation room: a multicenter retrospective study

Objective To develop and validate a practical model to identify MEWS-based high-alert physiologic status at the time of unplanned transfer to the emergency resuscitation room (UTER) among emergency patients under observation. Methods This multicenter retrospective study used complete-case data from three hospitals in China. The training cohort comprised 599 UTER patients from the First Hospital of Jilin University (January 1 – December 31, 2023). Temporal internal validation included 145 UTER patients from the same hospital (January 1 – March 31, 2024). Geographic external validation included 231 UTER patients from Hainan Provincial People’s Hospital and the Second Hospital of Shanxi Medical University (December 1, 2022 – March 1, 2023). M1 was the MEWS within 1 h after entry to the emergency observation room; M2 was the MEWS at UTER; Delta MEWS = M2 – M1. The modeled endpoint was MEWS-based high-alert status at UTER, defined as M2 ≥ 6. Logistic regression, nomogram construction, ROC analysis, calibration analysis, and decision-curve analysis were used. The positive class for precision-recall analysis was prespecified as MEWS-based high-alert status at UTER (M2 ≥ 6). Results In the training cohort, the optimal training-set cutoff for Delta MEWS was 3.5, corresponding to a sensitivity of 0.837 and a specificity of 0.924. In multivariable analysis, SpO 2 at UTER (OR 0.97, 95% CI 0.94–1.00; p = 0.027) and Delta MEWS ≥ 3.5 (OR 53.26, 95% CI 29.51–96.11; p < 0.001) were independently associated with high-alert status. The model achieved an AUROC of 0.890 (95% CI 0.853–0.927) in the training cohort, 0.957 (95% CI 0.917–0.997) in temporal internal validation, and 0.934 (95% CI 0.871–0.997) in geographic external validation. The corresponding AUCPR values were 0.729 (95% CI 0.673–0.781), 0.921 (95% CI 0.872–0.991) and 0.618 (95% CI 0.542–0.723), compared with cohort-specific positive-class fractions of 23.5, 33.1, and 6.5%, respectively. Conclusion Delta MEWS combined with SpO 2 provided bedside stratification of MEWS-based high-alert status at UTER. The nomogram may assist clinicians in recognizing severe physiologic worsening during transfer. Reporting precision-recall performance together with class balance clarifies model discrimination across cohorts with different prevalences.