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

Abstract

Objective: To develop an effective Management Information System (MIS) that is empowered by predictive models that can forecast the demand of end-stage cancer home hospitalized patients in individual and population levels, and help palliative care service systems operate smoothly where the demand is highly fluctuating, resources are limited, expensive, and hardly adjustable in a short time, and the backlog and shortage costs are high.

Method: Inspired by real problems faced by a palliative care center providing various medical, nursing, psychological, and social services in a home-based setting, two Long Short-Term Memory (LSTM) based deep learning models are proposed for demand forecasting at both individual and population levels. The individual-level model can predict the type and time of the next service required for a specific patient with a given demographic and health profile, and the population-level model helps with the prediction of next week's demand for various services in a center supporting a specific patient population. Predicted demand informs on optimal resource and operations plan through a well designed MIS.

Results: Experiments were conducted on a dataset consisting of more than 4000 cancer patients with a Palliative Performance Scale (PPS) of 40 and below discharged from hospital to home under a national palliative care center's home hospitalization service in Iran from September 2012 to July 2019. The models outperformed conventional time-series forecasting methods where applicable. Results indicate that the proposed models were capable of forecasting patients' demand with astonishing performances both individually and on larger scales.

Conclusion: Intelligent demand forecasting can help palliative care home hospitalization systems to overcome the challenge of progressive demand growth when a considerable portion of patients are approaching death, followed by a sudden drop in demand when those patients pass away. It helps to improve resource utilization and quality of care concurrently.

Keywords: Cancer palliative care; Deep learning; Demand forecasting; End of life care; Home care; Home hospitalization; Machine learning; Management information system (MIS).