How Can Hospital Teams Detect Deterioration Without False-Alarm Overload?

Why does our early warning system generate so many false alerts?

It generates so many alerts because sensitivity creates work, and each lower threshold casts a wider net.

The issue is not that an alert proves nobody is ill. An alert is a request for evaluation. In an external validation study across six Belgian hospitals, a NEWS score of 5 or more had a positive predictive value of 6.8% for unexpected death, CPR-treated cardiac arrest, or unplanned ICU admission. Its negative predictive value was 99.5%. The score can be good at ruling out danger while still creating many evaluations for patients who do not reach the defined outcome.

The workload changes sharply with the cut-off. The Journal of Clinical Nursing study found that 20.1% of patients had NEWS of 4 or higher, 11.3% had NEWS of 5 or higher, and 3.2% had NEWS of 7 or higher. A one-point change is not a small policy choice.

Alert noise also comes from timing and workflow. Nurses in a BMJ Quality & Safety study described alerts that arrived after deterioration was already recognised, perceived false positives, interruptions, unclear next actions, and missed nursing skill. An alert without a named action is not decision support. It is another task.

Measure the whole chain. Count alerts, then count clinical evaluations, response times, documented plans, and adverse deterioration that occurred without a preceding alert. The systematic review in the Journal of Hospital Medicine recommends positive predictive value or the workup-to-detection ratio alongside alert frequency and outcome definitions. A high AUC does not settle the staffing question.

How should a hospital choose an alert threshold for NEWS2 or another early warning score?

Choose the threshold through a local policy that balances the patients you serve, the response staff available, and the harm of both missed and excessive alerts.

NICE directs hospitals to specify measured parameters, observation frequency, explicit cut-off points, and a graded response. It also says thresholds should be set locally and reviewed regularly to optimise sensitivity and specificity. The threshold is only half the rule. The other half is what happens after it fires.

The Royal College of Physicians states that NEWS2 normally uses a single score of 3 and aggregate scores of 5 and 7. It also permits patient-specific adaptation after clinical assessment. That exception needs a rule: document the variance, the reason, the next monitoring frequency, and the point at which staff must escalate again.

Do not set one number and call the work finished. A retrospective study of 332,682 adult admissions reported that recorded registrar escalations trebled after NEWS2 was introduced in June 2019. The study authors linked that workload rise to higher sensitivity, lower specificity, and the escalation protocol. Threshold design and staffing design are the same problem viewed from different sides.

Write the policy as an operating agreement. The Royal College of Physicians says local protocols should define urgency, responder competence and seniority, further monitoring, and care setting. This mirrors the discipline of deciding whether routine work should stay with the primary team or move to specialist review: the handoff must have an owner, a time, and a record.

How can repeated alerts for the same patient be handled safely?

Repeated alerts should be handled as one active episode when a clinician is already responding, a documented plan is active, or the patient is improving under planned observation.

Suppressing a repeat alert without a record hides risk. Treating every repeat score as a new emergency floods the ward. The safer middle path is a structured reason for non-escalation, visible to the bedside nurse and the next responder.

The Royal Cornwall Hospitals NHS Trust policy permits staff to record that a repeat NEWS2 trigger does not need a new escalation because the care team is already present, an escalation plan exists, or the patient is improving. It also requires an individualised documented variance when normal physiology would otherwise produce inappropriate triggers or unnecessary increases in observation frequency.

Each episode needs four fields: who owns it, when escalation occurred, what response arrived, and when the score must be reviewed again. Royal College guidance calls for documentation of escalation time, response time and grade, assessment and treatment plan, and the individualised threshold for further response. That record turns alert handling into a closed loop.

Do not let an episode rule mute bedside concern. NICE says clinical concern can trigger response even when the score is low. Royal Cornwall states that concern about clinical condition overrides NEWS2. The score is a signal. The bedside assessment is the decision.