Emergency Department (ED) Analytics
St. Michael’s Hospital sees 75,000 annual emergency department (ED) encounters, with an expected increase of 2 percent every year. Increases in ED volumes can put a strain on staff resources and can lead to increases in patient wait times.
The LKS-CHART team has developed analytical models, including volume forecasting, an early warning system, and flow optimization, to help clinical and administrative teams better determine short term and long term resource and staffing needs in the ED.
Volume forecasting:
The volume forecasting model was developed to determine how ED volumes will change - hours, days and months into the future. This information can help hospital administration make staffing decisions to better accommodate patient volumes, as well as assist with hourly resource planning in light of unexpected ED volume surges.
The goal of the project is to create a user-friendly volume forecast dashboard, based on the analytical model, so ED staff can display and breakdown patient arrival predictions by levels of care required (from low to high).
Early warning system:
This recurrent neural network model takes historical volumes, calendar variables and weather data, and forecasts up to 12 hours in the future when the waiting areas and beds in the ED will reach 90% capacity.
Flow optimization:
The ED at St. Michael’s Hospital is undergoing renovation for the next several years. As part of the renovation plan, temporary waiting areas were designed to replace the areas that are currently under construction.
To understand the amount of volume that each of the new waiting areas would see, we asked Emergency Department physicians to review a retrospective sample of Emergency Room Intake data and then “assign” each patient in the dataset to one of the new ED waiting areas, based on the patient’s triage note. We then created a classifier on this data to learn the physician’s assignment rules (i.e which patients go where, based on diagnosis/procedures/etc.) and then applied the model on several months of data to see if the new spaces could meet anticipated demand.
