Digitally adapted, hyper-local realtime bed forecasting to manage flow for NHS wards

For the proposed project, we request funding for 2-years covering one software engineer, and one post-doctoral research associate supported by the assembled project team including health care design and quality management support. We will take an existing functioning bed demand forecast, and make it fit for scaling across the NHS.

Work is divided into the following packages: Governance, PPI and approvals and team onboarding Application foundation: building an application suitable for different levels of digital maturity; respecting interoperability standards; and providing a round trip for user interaction to improve forecasting Application integration: connecting the application to live data feeds and rapid prototyping and iteration cycles Upgrading the AI components of the model including adaptions to predict staff demand, consider COVID-19 adaptions, and explore transfer learning to allow deployment where training data is sparse Application user design: working closely with ward managers and clinicians through-out to ensure the application delivers actionable insights that lead operational efficiencies Model evaluation: both quantitative and qualitative evidence of safety, reliability and utility Quality management: ensuring the model and the application move toward CE marking A package of PPI interventions The application will deliver hyper-local demand forecasts for beds and eventually staff, and other resources.

We forecast the counterfactual demand rather than the observed provision.

The forecast is adapted and delivered to the hospital ward team since this is the tactical unit that can use the information to improve operational efficiency. Staff can be flexed, surgical timing adjusted, and case mix adjusted to make the most of the resource available.

The assembled team brings together expertise in software engineering, mathematics, machine learning and health informatics. We are embedded in a busy teaching hospital. We will take every advantage to embed the learning we gather from our users in the application we build.