Continuous vital sign monitoring on surgical wards: The COSMOS pilot

Study objectives: Alerts for vital sign abnormalities seek to identify meaningful patient instability while limiting alarm fatigue. Optimal vital sign alarm settings for postoperative patients remain unknown, as is whether alerts lead to effective clinical responses reducing vital sign disturbances. We conducted a 2-phase pilot study to identify thresholds and delays and test the hypothesis that alerts from continuous monitoring reduce the duration of vital sign abnormalities.

Design: Two-phase pilot.

Patients: 250 adults having major non-cardiac surgery.

Setting: Surgical wards.

Intervention: All patients had routine vital sign monitoring by nurses at 4-h intervals. We initially continuously recorded clinician-blinded saturation, heart rate, and respiratory rate in 100 patients. In the second phase, we randomized 150 patients to blinded versus unblinded continuous vital sign monitoring. In unblinded patients, nurses were verbally alerted to abnormal vital signs.

Measurements: In the first phase, we modeled expected alarm counts using 6082 h of continuous oxygen saturation, heart rate, and respiratory rate data. Thresholds and delays targeting ∼3 alarms per patient per day were selected for phase two. Primary analysis assessed the effect of unblinded monitoring across a 5-component primary composite of cumulative durations of vital sign abnormalities. Secondary outcomes included fraction of alerts deemed meaningful by nurses and number of clinical interventions.

Results: In phase one, we identified alarm settings that yielded an average of 2.3 alerts per patient per day. In phase two, the average relative effect ratio of geometric duration means for vital signs exceeding thresholds was 0.75 [95 % CI: 0.51, 1.1], P = 0.17. Sixty alarms (82 %) were deemed useful in unblinded patients, leading to 60 % more interventions in unblinded patients.

Conclusions: We were able to select continuous saturation, heart rate, and respiratory rate thresholds that generated about 2 alerts per patient per day, nearly all of which were considered useful by nurses. Unblinded monitoring and nursing alerts led to interventions (mostly increasing oxygen delivery) that non-significantly reduced vital sign abnormalities by 25 %.

Clinicaltrials: gov registration: NCT05280574.