Abstract
ABSTRACT
Intensivists face a difficult decision deciding when a patient should be discharged from the intensive care unit (ICU) to a unit of lesser acuity. Accurate algorithms don’t exist to aid in that decision. Therefore, we developed and validated a stratification tool using physiologic data obtained from a patient’s electronic medical record for assignment to risk categories of mortality or hospital discharge to hospice.
Population: 33,039 admissions at 41 adult ICUs using an electronic medical record from 1/1/2012-6/30/2017.
Outcome: Mortality on a unit of lesser acuity or discharge to a hospice after leaving the ICU (“mortality/hospice”).
Methods: Vital signs from three hours to one hour before actual discharge were obtained. We used the 12 most proximate values for heart rate, respiratory rate, and mean arterial pressure, respectively. A letter was assigned to the median of every three measurements based on the underlying distribution of the vital sign. Four consecutive letters were concatenated to form a pattern, which were candidates for triggers (i.e. risk alerts). A patient could have three triggers, one for whether or not each vital sign contained a word that increased risk, along with a fourth trigger if a patient received mechanical ventilation. Using a genetic algorithm that weighted the outcome of mortality/hospice, we acquired a set of patterns that maximally increased risk. Those patterns were then validated as triggers for increased risk.
Results: The overall mortality/hospice rate was 4.7%. Fifteen patterns were identified that increased risk. Patients without triggers had a mortality/hospice rate of 3.2%, while patients with one, two, and three to four triggers had rates of 7.5%, 13.3%, and 27.5%, respectively,
Conclusion: It’s possible to use vital signs proximate to when a discharge decision is made to identify patients with an increased risk of either mortality or discharge to hospice after leaving the ICU.