The implementation of a real time early warning system using machine learning in an Australian hospital to improve patient outcomes

BACKGROUND: Early Warning Scores (EWS) monitor inpatient deterioration predominantly using vital signs. We evaluated inpatient outcomes after implementing an Artificial Intelligence (AI) based intervention in our local EWS. METHODS: A prior study calculated a Deterioration Index (DI) with logistic regression utilising demographics, vital signs, and laboratory results at multiple time points to predict any major adverse event (MAE-all cause mortality, ICU admission, or medical emergency team activation). The current study is a single hospital, pre-post study in Australia comparing the DI plus the existing EWS (Between the Flags-BTF) to only BTF. Data were collected on all eligible inpatients (≥16 years, admitted ≥24 hours, in general non-palliative wards). Controls were inpatients in the same hospital between January and December 2019. The DI was integrated into the electronic medical record and alerts were sent to senior ward nurse phones (July 2020-April 2021). RESULTS: We enrolled 28,639 patients (median age 73 years, IQR: 60-83) with 52.3% female. The intervention and control groups did not show any statistically significant differences apart from reduced admissions via the emergency department in the intervention group (40.4% vs 41.6%, P = 0.03). Risk for an MAE was lower in intervention than control (RR: 0.81; 95%CI: 0.74-0.89). Length of hospital stay was significantly reduced in the intervention group (3.74 days, IQR 1.84-7.26) compared to the control group (3.86 days, IQR 1.86-7.86, P = 0.002) CONCLUSIONS: Implementing the DI in one hospital in Australia was associated with some improved patient outcomes. Future RCTs are needed for further validation.