Ward deterioration technology: no reliable winner among alerts, models and continuous monitoring

Design

Systematic review and Bayesian network meta-analysis of randomised trials

Population

28 trials; 6 in the mortality network (13,716 observations), non-ICU inpatients

Primary outcome

In-hospital or 30-day mortality; unplanned ICU transfer

Effect

Mortality OR: rule-based 0.91 (0.35–2.30), predictive 1.22 (0.61–2.31), continuous 0.70 (0.36–1.29)

A Bayesian network meta-analysis in the Journal of Medical Internet Research (24 September) compared three approaches to spotting deterioration on general wards — rule-based electronic alerts, predictive-model alerts and continuous physiological monitoring — against standard care, using randomised trials.

None clearly changed mortality: odds ratios were 0.91 for rule-based alerts, 1.22 for predictive models and 0.70 for continuous monitoring, all with credible intervals crossing 1. Results for unplanned ICU transfer were similar. Expanded rule-based systems did not clearly reduce cardiac arrests (OR 0.94). Confidence in every comparison was very low.

The authors' point is practical: sophistication of the algorithm is not what determines benefit. What happens after the alert — who responds, how fast, and with what authority — probably matters more than how the alert is generated. For emergency and critical care teams who receive these calls, that is a case for investing in the response pathway rather than the software.

• Do not expect a predictive algorithm alone to reduce ward deaths.
• Continuous monitoring showed a favourable but imprecise trend.
• Invest in a clear, rapid response pathway for any alert system you use.
• Audit alert fatigue and response times alongside outcomes.

Why it matters

Hospitals are buying predictive tools on promise, and the randomised evidence does not yet support a hierarchy.

Don't overread it

Very low confidence in all comparisons; absence of a clear effect is not evidence that these systems do nothing.

The statistics, in plain English

A credible interval is the Bayesian version of a confidence interval. For continuous monitoring, 0.36 to 1.29 means the data are consistent with anything from a large benefit to modest harm. Very low confidence means these estimates could change substantially with more trials.

Read the rest in the app

You have read your two free briefings this month. The app carries all 27 specialties, every morning, free — and this finding is waiting in it.

Get Daily Doctor — free

Scan to keep reading on your phone. No account needed to start.