Data Analytics for Patient Scheduling

COVID and NHS appointment backlogs

The management of patients with long term rheumatic conditions is complex and requires significant healthcare resources. Following the COVID pandemic creating a backlog of appointments, rheumatology services are now having difficulties meeting the demand. Optimising the use of clinic resources whilst maintaining best patient outcomes is required to balance capacity and demand.

Source: Institute of Fiscal Studies “Could NHS waiting lists really reach 13 million?”

The Royal National Hospital for Rheumatic Diseases based at the Royal United Hospitals, Bath, has collected a significant amount of anonymised data pertaining to 84,906 outpatient appointments across the unit over the four-year period 2015-2018. Pilot data has indicated logistical inefficiencies and psychological/medical diagnostic behavioural traits, which reveal opportunities to algorithmically “learn” optimal scheduling policies and suggest a restructuring of the procedural delivery.

Reinforcement learning approach

The entirety of the patient treatment process can be considered as a complex high-dimensional, multi-parameter temporal process which includes factors such as e.g. the referral from a GP, consultation logistics, staffing regimes and the complexity of rheumatological treatment. A primary objective of the health service is to optimise the safe discharge of patients. To this end, this project uses machine learning methods to understand the relationship between appointment outcomes and operational and clinical characteristics. The principle idea is to use reinforcement learning algorithms to develop a dynamic real-time, self- adjusting appointment system.