High dimensional forecasting of patient flow in acute healthcare
Professor Richard Dobson
Professor of Medical Bioinformatics
Biostatistics and Health Informatics
Dr James Teo (King’s College Hospital), Dr Dan Bean (King’s College London)
Patient flow through a hospital is a key determinant of efficient hospital functioning and quality of care. Previous studies of patient flow have mainly focused on linear flow processes, queuing and single patient pathways which fail to capture the complex topology of a real-world hospital comprising hundreds of patient journeys overlapping in time and space. Recently, we published the first analysis of patient flow in two sites of Kings College Hospital NHS Foundation Trust using network analysis to represent and understand these complex patterns. Whilst we were able to identify key pathways, subsequent advances in machine learning and big data analysis may enable the development of more advanced models which are more accurate and enable forecasting of flow. Accurate forecasting and alerting methods would allow preventative steps to be taken such that patient flow is sustained during high-strain periods.
Electronic Health Records (EHRs) contain vast amounts of data on patient care. This includes timestamped transfers between various wards, demographic data, test results and free text (such as medical history and diagnosis). The EHR in Kings College Hospital is supplemented by state-of-the-art analytics tools including CogStack (information search and retrieval) and MedCAT (medical natural language processing). This project will combine flow data (admission and discharge times, route through the hospital, wait for treatment) with patient-level data to build a predictive model to forecast patient flow. This model will be trained using the vast historical data available at Kings College Hospital.
Once the model is trained, the next phase of the project is to deliver these analytics insights to hospital staff. This will involve developing a dashboard for end users that plugs into the existing CogStack ecosystem.
The main dataset is pseudonymised flow data from King’s College Hospital. The student will require a KHP Passport for access. This data can be used for audit without specific ethical approval. For more detailed analysis requiring further patient data, ethical approval will be sought through the KERRI committee. All data will be deidentified and will not leave KCL/KCH hardware.
Patient flow, networks, predictive modelling, visualisation