Interactive decision support for patient discharge from ICU

Chief Investigator

Dr Chris Bourdeaux,  University Hospitals Bristol and Weston NHS Foundation Trust

Co-investigator

Dr Chris McWilliams, University of Bristol

Summary

Demand for intensive care unit (ICU) beds is growing from both emergency and elective patients. The timely identification of patients that are fit to leave ICU is therefore a pressing priority for clinicians. Discharging patients too early from ICU risks deterioration on the ward and early readmission to ICU. Any delay to a discharge from ICU after the patient is medically fit can result in delays to admission of emergency patients or cancellation of planned major surgery. Delayed patients also suffer increased hospital length of stay and delayed rehabilitation and NHS England consider low levels of delayed discharges as a key quality indicator.

Decisions on readiness for ICU discharge are currently made by clinicians on an ad hoc basis and are at risk of being overly subjective due to many factors such as bed pressures and level of clinical experience. Thus, these decisions may result in unwarranted variation in ICU discharge and poor outcomes for patients. Whilst it is difficult to determine an objective ready for discharge date, and therefore to measure this variation, a local audit demonstrated marked variation in individual consultant decisions on discharge with some consultants discharging nearly twice as many patients per day on average as others with no change in readmission rate or illness severity.

We have completed preliminary work demonstrating that a machine learning algorithm can detect patients that are fit to leave and performs better than previously published nurse-led discharge checklists which are extremely conservative. This technology has the potential to streamline the discharge process, reduce variation and improve important clinical and operational outcomes. We will submit an NIHR i4i PDA grant in order to further develop this technology into practice as a decision support tool and bring it to market.