Machine learning-based demand forecasting in cancer palliative care home hospitalization

Area of Science:

• Health Informatics
• Artificial Intelligence in Healthcare
• Palliative Care Research

Background:

• Palliative care home hospitalization services face challenges with fluctuating patient demand and limited resources.
• High costs associated with service backlogs and shortages impact operational efficiency.
• Effective Management Information Systems (MIS) are crucial for managing these complex demands.

Purpose of the Study:

• To develop an effective MIS for palliative care home hospitalization, utilizing predictive models for demand forecasting.
• To forecast patient demand at both individual and population levels to ensure smooth service operations.
• To address the challenges of fluctuating demand, resource limitations, and high operational costs.

Main Methods:

• Proposed two Long Short-Term Memory (LSTM) based deep learning models for demand forecasting.
• Developed an individual-level model to predict specific service needs for patients.
• Created a population-level model to forecast weekly service demand for a given patient group.

Main Results:

• Experiments conducted on over 4000 cancer patients in home hospitalization.
• The proposed LSTM models demonstrated superior performance compared to conventional time-series forecasting methods.
• Achieved high accuracy in forecasting patient demand at both individual and population scales.

Conclusions:

• Intelligent demand forecasting is vital for palliative care home hospitalization systems.
• Accurate forecasting helps manage progressive demand growth and sudden drops in patient needs.
• The study highlights the potential to improve resource utilization and enhance the quality of care concurrently.