Ambulatory care (on-going)

Historically, hematology and oncology treatments were administered in an inpatient setting considering their complexities and the need for lengthy admissions. This put healthcare providers in unviable positions due to increasing expenditures and the ever-rising demand (increasing number of patients, long-lasting care, and longer follow-up time). For instance, Ontario had doubled its chemotherapy budget from 2003 to 2008, yet wait times had not decreased. Thus, Ambulatory Care (AC) (or outpatient care) has been the primary method of chemotherapy treatments over recent years. However, hematology and oncology AC wait times are still excessive after more than a decade, which is a direct consequence of earlier diagnoses, an ageing population and improved survival through better surgical techniques, advances in radiotherapy and more effective procedures. Delayed hematology and oncology treatments can have a substantial negative impact on health outcomes. In hematology and oncology AC, long wait times are the main source of patient (also staff) dissatisfaction, depression and anxiety, which are their most important psychopathological comorbidities. Apart from affecting patient and staff satisfaction, long wait times in hematology and oncology AC may adversely affect patient adherence to scheduled appointments.

In this project, we investigate one of the largest Canadian hospitals with geographically distributed AC campuses. The daily challenge facing the booking clerk is to allocate the available capacity between various patient classes and priorities in order to decrease the number of patients whose wait time exceeds a pre-specified threshold with a greater weight given to any late bookings of higher-priority demand. Thus, our project objective is to design a proactive decision support tool for scheduling patients in a distributed AC center. We consider a dynamic setting with uncertain patient arrival and use of the emergency department. This problem is formulated as an infinite-horizon Markov decision process model and solved as an approximate dynamic program. The proposed approximate dynamic programming algorithm is hybridized with a neural network, which accelerates the sub-problem in the column generation and ensures the feasibility of solutions. Simulation results demonstrate that my heuristic policy performs well compared to the approximate optimal policy and considerably outperforms well-known incumbent scheduling policies.