Forecasting patient demand at urgent care clinics using explainable machine learning

Urgent care clinics and emergency departments around the world periodically suffer from extended wait times beyond patient expectations due to surges in flows. The delays arising inadequate staffing levels during these periods have been linked with adverse clinical outcomes. Previous research into forecasting flows has mostly used statistical techniques. These studies also predominately focussed on short-term forecasts, which limited practicality for resourcing of medical personnel. This study joins an emerging body work seeks explore potential machine learning algorithms generate accurate forecasts presentations. Our uses datasets covering 10 years two large urgent develop long-term flow up one quarter ahead using a range state-of-the-art algorithms. A distinctive feature this is use eXplainable Artificial Intelligence (XAI) tools like Shapely LIME that enable in-depth analysis behaviour models, would otherwise be uninterpretable. enabled us ability models adapt volatility demand COVID-19 pandemic lockdowns identify most impactful variables, resulting valuable insights their performance. results showed novel combination advanced univariate Prophet as well gradient boosting, ensemble, delivered consistent solutions average. approach generated improvements 16%–30% over existing in-house methods estimating daily 90 days ahead.