Forecasting and Optimization of Patient Flow Through Predictive Machine Learning

Executive Summary

Patient overflow in Emergency Departments (ED) has become a significant challenge for healthcare institutions in the United States. It has led to overcrowding, long wait times, and decreased quality of care. The consequences of patient overcrowding include increased morbidity and mortality rates, decreased patient and staff satisfaction, and higher healthcare costs.

A trusted healthcare institute in the US operating across multiple states in the country, sought a solution to predict the patient volumes in each department for better staff planning and to reduce the length of stay for patients. The team at Factspan helped the healthcare institution to develop a customized and scalable predictive data modeling solution that uses patient volume history and resource availability to manage patient flow across the departments. By focusing on anticipating patient demand and optimizing resource allocation, the model enabled the hospital to reduce wait times and improve patient experience.

About the Client

The client, a faith-based not-for-profit organization, is one of the most trusted healthcare institutions in the United States. With nearly 2 million patient visits per year and with over 4000 physicians, they have earned a reputation for providing exceptional medical care and are recognized as a leader in the healthcare industry. The organization is known for its commitment to patient-centered care, with a focus on personalized treatment plans, advanced medical technologies, and a team-based approach to care.

Business Challenge

Sudden spike in patient volume in critical departments often resulted in a very chaotic scenario. When patient admissions exceeded the capacity of the hospital, it resulted in inadequate staffing, longer wait times, delayed diagnosis and treatment, increased risk of infections, and compromised patient outcomes. The entire scenario put undue stress on the overburdened hospital staff, accelerating mental and physical burnout. With a focus on continuous improvement and innovation, our client was looking for a solution to reduce patient overcrowding in the Emergency Department (ED). The healthcare institute needed answers to some pertinent questions:

• What’s the daily admissions estimate for hospitals?
• When does the emergency department experience high patient flow?
• How many procedures can the surgical staff perform in a day?
• And more…

Our Solution

To develop a ML model for patient inflow management, the experts at Factspan gathered relevant patient information regarding admissions, discharges, volumes etc. The team then engineered features that will be used to train the model, selecting the most relevant variables for the task, and developing an algorithm that balances performance and efficiency considerations. The actual patient inflow data was used as feedback to fine-tune the model accuracy. Finally, the team deployed the model into the healthcare institution’s existing software infrastructure, ensuring that it is secure and reliable in a production environment.

The machine learning model had been designed to forecast demand, prevent delays, reduce readmission rates which led to improved outcomes and reduction of costs for the hospital. The insights provided by the ML model have also helped the hospital to manage patient flow, reduce ED overcrowding, and enhance the quality of care provided to the patients.

The model was built on the following algorithms: Performance Clustered eXtreme Gradient Boosting on Elastic Net Predictions, eXtreme Gradient Boosted Trees Regressor with Early Stopping, Performance Clustered eXtreme Gradient Boosted Trees Regressor, Seasonal AUTOARIMA with calendar of special events, eXtreme Gradient Boosted Trees Regressor with Early Stopping, Temporal Hierarchical Model with Elastic Net and XGBoost

Business Impact

• Patient inflow prediction with an accuracy over 89%
• Reduction in patient overcrowding and long wait times by 50%
• 30-40% improved resource allocation and utilization through data-driven decision
• Reduction in preventable readmissions and adverse events by 8-20%