Accurately predicting when a patient will be ready to go home is at the centre of every effective discharge plan ensuring patients have a smooth transition from hospital to home.
Getting this timing right allows the patient’s family or care partners to prepare for their arrival home and helps staff with coordinating timely community and home supports.
But knowing exactly when a patient can go home, historically, has been informed by clinician estimations rather than real-time data.
“Traditionally, clinicians estimate patient discharge times during rounds or morning huddles,” explains Teresa O’Callaghan, executive director, Abbotsford Regional Hospital and Cancer Centre and Mission Memorial Hospital. “While clinicians are experienced, human estimations can be inaccurate. This unpredictability makes planning difficult.”
Now, thanks to an innovative AI solution developed by Fraser Health’s Centre for Advanced Analytics, Data Science and Innovation (CAADSI) team, staff have a more precise idea of when a patient will be ready for discharge.
“Using the AI predictive discharge model, our staff and medical staff are able to see on any given day who could be ready to go home within 24 hours,” explains Teresa. “This helps our staff plan and makes the discharge process smoother and more efficient.”
This custom AI model combines classical demand forecasting and generative AI to analyze over 72 variables, including clinical data, patient history, vital signs and unstructured case notes, forecasting discharge readiness within 24 hours.
Prior to integrating the AI model into the electronic health record and deploying at Fraser Health’s 12 hospital sites, it was trained on 100,000 patient records from two years of real data and tested over two months using a third year’s data. When comparing the tool’s predictions with the actual experience in discharges, the AI's discharge predictions achieved 86 per cent accuracy. This is four times more accurate than the traditional process which relied on human estimates.
“We took an agile approach, experimenting alongside clinical teams with how best to integrate the tool into clinical workflows,” says Sheazin Premji, executive director, CAADSI.
According to Sheazin, the inputs to the AI model were meticulously chosen through extensive research involving physicians, nurses, scientific literature and an assessment of Fraser Health’s data availability and data quality. Inputs include patient information such as age, gender, medical history and recent lab results, among others.
“This partnership allowed the AI model to evolve iteratively based on real-world clinical insights, reinforcing that AI thrives on continuous refinement and adaption, rather than being a one-and-done effort.”
Staff and medical staff can now access these predictions through a live dashboard, which is embedded into their current daily workflow.
This dashboard and the discharge prediction tool have proven highly effective. Prior to implementing the AI, hospital staff might achieve 250 to 300 discharges a day, across the region. Now, with the help of this tool, it is not unheard of for our staff to complete 600 discharges in a single day.
Discharging patients as soon as safely possible is important. Patients recover best at home in familiar surroundings where they can rest. Plus, their risk of hospital-acquired infections is reduced the earlier they leave the hospital. Discharging patients also frees up hospital beds and resources for other patients who need care.
The AI solution delivers daily, patient-level discharge probability predictions, enabling proactive discharge planning, early identification of barriers and efficient resource allocation. This innovative approach enhances care and improves patient outcomes.
AI can uncover subtle data relationships that may not be immediately apparent to clinicians, capturing complex patterns across numerous variables and a large volume of data. Ultimately, the patient’s care team has the final say on when it is safe for a patient to be discharged. The AI solution’s use of extensive data helps to inform our medical staff and enhance the accuracy of predictions, without making the final decision.
This success exemplifies our digital health strategy in action, showcasing how we are transforming health care through technology. We are deeply grateful to all the teams involved in this journey toward a more efficient system that ultimately enhances patient care.
