NHS England and NHS Improvement
The COVID-19 pandemic forces the NHS to make hard decisions every day to prioritise the safe delivery of care for patients.
In the first wave of the pandemic, the NHS was operating blind; it was difficult to predict the demand on health and care services from COVID-19 positive admissions or areas where the system would face strain.
National, regional and local teams, ranging from national incident response to frontline staff, needed to be forewarned of potential upticks in COVID-related hospital activity, so that they could plan effectively. Vitally, they needed to be confident in the forecasts and trust that they could be used to facilitate planning of services for patients.
As part of our work supporting the NHS’ data response strategy to the pandemic, Faculty helped to build the COVID-19 Early Warning System (EWS).
The EWS is a first-of-its-kind toolkit that forecasts vital metrics such as COVID-19 hospital admissions and required bed capacity up to three weeks in advance, based on a wide range of data (e.g. COVID-19 positive case numbers, 111 calls and mobility data). The techniques applied allow forecasts to use specific information from that trust, as well as incorporating a broader set of contextual information that impacts the forecast, such as what’s happening in COVID-19 admissions in other local hospital trusts. This allows the EWS to incorporate much more relevant information into the forecasts for individual trusts than was previously possible.
The EWS predictions can be viewed on a national, regional, system or trust level and are updated on a daily basis. For the first time ever, people working at all levels of the NHS have access to timely insights about the potential impact of the virus on patient demand for frontline services. This enables organisations to plan and put mitigations in place at both a national and local level.
A forecast of estimated daily COVID-19 admissions from the NHS Early Warning System dashboard.
The EWS is now part of daily workflows for hundreds of analysts and senior decision makers across the NHS. The live product is updated as users’ needs change in response to different phases of the pandemic, allowing them to access new views of the data when needed.
The EWS leverages Faculty’s pre-built AI safety technology tools, and model validation analysis has been included to support operational decision makers. Faculty’s latest explainability techniques help users to understand and interpret how each input, such as past hospital admissions or local testing data, influences the outcome of the forecast. For example, users can see how much recent historical admissions data for a particular trust is driving that trust’s forecast, versus how much local testing data is influencing it.
All modelling is validated statistically by comparing the forecasts with real data. The user’s dashboard shows the reliability of the forecasted confidence intervals and how the forecasts have evolved over time. Giving users access to this kind of model performance analysis is driving trust in the EWS tool and helping decision makers interpret forecasts more accurately.
Forecasts are generated using Bayesian hierarchical modelling, which allows the model to learn from trends within regions from data reported by hospitals on a daily basis. For example, if a few geographically close trusts were experiencing an uptick in COVID-19 admissions but another geographically close trust wasn’t, this information would be incorporated into the forecasts of all of the trusts.
There are currently around 1,000 users of the forecasts across the NHS who are accessing the EWS or using the outputs for their own analysis and to help inform the prioritisation of safe care delivery for patients on a daily basis. Users include both the central strategic incident response team, as well as regional analysts in local and regional trusts.
In conjunction with local intelligence and other data sources, the NHS has used the EWS to support decision making about how to respond to short- and medium-term pressures on hospitals across England. For example, it has helped national supply chain leads to ensure ventilators and oxygen supply is targeted where it is needed so no hospital has run out.
Professor Keith Willett, NHS England and NHS Improvement’s Senior Incident Director for COVID-19 said:
‘As a national incident team, we were able to review the likely issues that we were going to face in the next one to two weeks. This enabled proactive conversations with key system leaders to ensure that resilient mitigations were in place and the likely period of time that they would be required for. It also strengthened our position in national conversations as we engaged in constant discussions over the course of our national COVID-19 pandemic response about what the focus should be for the NHS – to provide the best quality of care for patients. Very helpfully, the team managing and curating the EWS would also be able to highlight trusts, STPs, and regions of interest and offer useful insights to explain possible changes in key system metrics over time.’
Locally NHS trusts are using the EWS to help them to plan how to use their available capacity for both COVID-19 patients and routine care and operations, with the benefit of advance knowledge of how the need to care for more or less patients with the virus might change in the coming one to three weeks. It is also being used to support the recovery of critical services, giving leaders confidence to move forward with increasing elective care capacity.
Ming Tang, Chief Data & Analytics Officer, NHS England and NHS Improvement said:
‘By using this leading technology developed with Faculty, we are helping to support frontline staff in their ongoing mission to save as many lives as they can by equipping them with the most accurate information. This tool is incredibly important for helping local teams plan to bring back on services for other patients safely, while at the same time flexing capacity locally to support COVID-19 care.’