NEWS2 shows poor predictive accuracy for deterioration on virtual wards

Scott et al. (PLoS One 21(4):e0347678; PMID 42048369) conducted a retrospective observational cohort study of all remotely monitored patients aged 16 years and over admitted to virtual wards in Bristol, North Somerset and South Gloucestershire between October 2023 and February 2025. The study included 2,533 admissions across five care pathways: respiratory (41%), outpatient parenteral antimicrobial therapy (19%), frailty (18%), general (18%), and heart failure (7%), with pathways not mutually exclusive.

The principal finding was that NEWS2 had poor predictive accuracy for hospital readmission. Both first NEWS2 and maximum NEWS2 produced an area under the curve (AUC) of 0.55 (95% CI 0.51–0.60 and 0.50–0.59 respectively) — barely above chance. During the study period, 177 (7%) virtual ward admissions resulted in hospital readmission and 9 (<1%) died. Median virtual ward length of stay was 10 days (IQR 6–14).

The acuity profile of the cohort is informative. First NEWS2 was 0–2 in 67% of admissions, 3–4 in 23%, 5–6 in 8%, and 7 or above in 2%. Maximum NEWS2 during admission showed a broader distribution: 32% remained at 0–2, 36% reached 3–4, 23% reached 5–6, and 9% reached 7 or above. The clinical acuity of most patients remained stable, with 30% not deteriorating at all and 41% deteriorating by only 1–2 points. The authors note that the NEWS2 distribution on virtual ward admission was similar to that observed in acute hospitals in the 2022 Society for Acute Medicine Benchmarking Audit — the same score distribution, but without the predictive discrimination that makes it clinically useful in the acute setting.

This finding raises a specific governance question for any health system expanding virtual ward capacity. NEWS2 was designed for and validated in acute hospital settings. Its application to virtual ward populations — who are typically lower-acuity, remotely monitored, and managed across heterogeneous care pathways — does not produce the predictive discrimination needed for reliable clinical escalation decisions. An AUC of 0.55 means the tool does not meaningfully distinguish patients who will require escalation from those who will not. The authors advise that the low hospital readmission rate (7%) means this finding should be interpreted with caution, but the direction of the result is clear: NEWS2 alone is insufficient.

The implication is not that virtual ward patients do not need monitoring, but that the monitoring tools must be appropriate to the setting. As Hospital-at-Home and virtual ward models continue to expand — supported by growing randomised evidence for their safety and effectiveness — the risk stratification instruments that underpin their clinical governance must be validated in community rather than hospital populations. Services relying on NEWS2 as their primary virtual ward safety metric should assess whether supplementary or alternative approaches to deterioration detection are needed for this distinct clinical context.