Imperial College London

Academic Institute

Anthony Gordon

Unknown

No

NIBDAPC_2021_0008

Sepsis (severe infections with a high risk of death) represents a global healthcare challenge, a leading cause of mortality and the most expensive condition treated in hospitals. Additionally, sepsis is a central contributor to most deaths related to COVID-19 infections. It was recognized as a top priority by the James Lind Alliance, a consortium bringing together patients and clinicians to prioritise the most pressing unanswered questions and inform the NIHR. A cornerstone of the treatment of sepsis is the administration of intravenous fluids (sterile salty water given directly in the veins) and vasopressors (drugs that constrict the blood vessels to normalise the blood pressure). However, there is huge controversy around the individual dosing of these drugs in a given patient. A tool to personalise these medications could improve patient outcomes. Our contribution to the field was the development of a new method to suggest the correct dose of medications to doctors, which was created using artificial intelligence algorithms applied to large medical databases in the USA. This tool has the potential to drastically improve sepsis management, save lives and precious ICU resources. Now, we would like to test this AI system retrospectively using UK data from ICHT, without influencing patient care or actually using the AI in the NHS. One way to do this is to check whether patients who received (in the past) the dose recommended by the AI had better outcomes. We also intend to re-calibrate the model using UK data, which involves re-training the existing model with new UK data, and check whether this improves model performance in this patient population. To conclude, accessing ICHT data to validate the model represents a crucial step towards clinical validation of our AI tool, which we are conducting in parallel via an NIHR/NHS-X AI in Health and Care Award. We aim to publish the output of this work in the scientific and lay press, to maximise its impact.

First and foremost, it is essential to improve the management of sepsis in order to decrease sepsis-related morbidity mortality and to reduce healthcare expenditures. Reducing sepsis mortality by even a few percent represents hundreds or thousands of lives saved annually in the UK alone, and several tens of millions of pounds in direct and indirect costs. The COVID-19 pandemic, with staff shortages and increased number of critically ill patients, has only rendered the need for such systems more acute. The importance of the problem for the NHS and the NIHR was clearly highlighted in several reports and strategy documents. The James Lind Alliance (which informs the NIHR by bringing together patients, carers and clinicians to identify and prioritise top unanswered questions that they agree are the most important) lists our exact same question in the top priorities for Emergency Medicine: “In patients with sepsis does a liberal fluid resuscitation strategy versus early vasopressor use result in increased morbidity and mortality?”. Sepsis is also listed in the James Lind Alliance top 10 priorities in intensive care. The 2014 National Information Board strategy report “Personalised health and care 2020” sets out proposals intending to “‘bring forward life-saving treatments and support innovation and growth”, and to support the development of new medicines and treatments “particularly in light of breakthroughs in (…) tackling infectious diseases”. Tailoring clinical management to an individual patient is the very principle of my approach. NHS Digital is now in charge of delivering this strategy in its mission to “transform health and care through technology”. The 2018 NHS-commissioned report “Thinking on its own: AI in the NHS” stated that “AI could support the delivery of the NHS’s Five Year Forward View, which aims to narrow (…) gaps in health provision“ (Harwich & Laycock, 2018). More specifically, it highlighted that AI could support “the reduction of the care and the quality gap (…) as [AI] can give all health professionals (…) access to cutting edge diagnostics and treatments tailored to individual need” (Harwich & Laycock, 2018). The report also stressed that public trust was “vital for the successful development of AI”, and that solutions had to be found to “overcome concerns of both the public and healthcare professionals”, which is why we put such a strong emphasis on involving these entities and other stakeholders such as HDR UK (Harwich & Laycock, 2018).

Public Health Research

17/12/2021

De-Personalised

Not applicable

Not applicable

One-off

TRE