Individualized Dynamic Patient Monitoring Under Alarm Fatigue

Hospitals are rife with alarms, many of which are false. This leads to alarm fatigue, in which clinicians become desensitized and may inadvertently ignore real threats. We develop a partially observable Markov decision process model for recommending dynamic, patient-specific alarms in which we incorporate a cry-wolf feedback loop of repeated false alarms. Our model takes into account patient heterogeneity in safety limits for vital signs and learns a patient’s safety limits by performing Bayesian updates during a patient’s hospital stay. We develop structural results of the optimal policy and perform a numerical case study based on clinical data from an intensive care unit. We find that compared with current approaches of setting patients’ alarms, our dynamic patient-centered model significantly reduces the risk of patient harm.