Related Experiment Video
Updated: Jun 26, 2026
Implementation of a Real-Time Psychosis Risk Detection and Alerting System Based on Electronic Health Records using CogStack
Published on: May 15, 2020
A novel face scale and National Early Warning Score 2 prediction model to assess deteriorating patients in hospital
Alanna Wall1, Jack A Fisher1, Emma Kate Watkinson1
1Department of Intensive Care, North Middlesex Hospital, Sterling Way N18 1QX, London, United Kingdom.
Background:
Early warning scores like the National Early Warning Score 2 (NEWS2) identify patient deterioration but do not incorporate clinician-intuitive assessment. Previous research suggests facial expressions can be critical indicators of deterioration risk.
Objectives:
The aim of this study was to evaluate whether a prototypical facial expression scoring scale (quick visual early warning score [qVIEWS]) used alongside NEWS2 improves identification of intensive care unit (ICU) admission in hospital ward patients.
Methods:
A single-centre observational cohort study of 333 adult ward patients was conducted in a London Acute Hospital, assessed by trained outreach members at the initial (Index) review using both pictographic qVIEWS and NEWS2. The primary outcome was admission to the ICU. Random forest modelling compared NEWS2+qVIEWS versus NEWS2-only prediction and Cox proportional hazards modelling assessed time to ICU admission. The study was approved by the United Kingdom Health Research Authority (Integrated Research Application System 289191, Research Ethics Committee 21/LO/0197).
Results:
Among 333 patients (ICU admission in 17.4%), higher qVIEWS were associated with a higher Index NEWS2 (p < 0.001), sustained physiological instability to 48 h, increased hospital mortality (risk ratio: 2.62, 95% confidence interval [CI]: 1.39-4.95), and greater ICU admission risk (risk ratio 2.03, 95% CI: 126-3.27). Inter-rater reliability for qVIEWS was 0.98, and qVIEWS moderately correlated with NEWS2 (r = 0.24, p < 0.001). Random forest discrimination for ICU admission was similar between NEWS2+qVIEWS and NEWS2-only models (area under the curve: 0.73 vs 0.74; p = 0.37) with comparable calibration and decision curve net benefit. Permutation importance demonstrated meaningful contribution of qVIEWS, exceeding several NEWS2 components. In exploratory multivariate Cox analysis, qVIEWS was independently associated with earlier ICU admission (high risk: 1.65 per unit increase, 95% CI: 1.02-2.66; p = 0.041) alongside respiratory rate.
Conclusion:
qVIEWS did not improve predictive performance beyond NEWS2 and delivered comparable classification metrics across prespecified thresholds. Exploratory analyses suggested an association between qVIEWS and earlier ICU admission although findings should be interpreted cautiously. While results do not support immediate clinical implementation, they provide proof of concept for facial expression assessment as a novel dimension in acute illness evaluation. Further research, particularly into serial measurements and automatic facial recognition, is warranted.