Clinical validation of the AI Clinician decision support system for sepsis treatment

Sepsis is life-threatening organ dysfunction due to severe infection and affects 250,000 patients annually in the UK, of whom 48,000 die. In addition, virtually all Covid-19 intensive care unit (ICU) deaths had sepsis.

As part of ICU treatment, the cornerstone of sepsis resuscitation is the administration of intravenous fluids and/or vasopressors to maintain blood flow to prevent organ failure.

However, uncertainty around correct dosing and timing of treatments, partially due to clinical syndrome heterogeneity, leads to poorer outcomes and increased ICU-resource use.

We tackle this problem using an AI that learned from ICU electronic healthcare records to provide personalised treatment recommendations to clinicians.

Our AI recommender system, the “AI Clinician”, extracted implicit knowledge from large ICU-databases and learned optimal sepsis treatment by analysing myriads of clinical decisions.

We demonstrated in proof-of-principle retrospective studies that our AI s treatment strategy was on average better than clinicians: In a 100,000-patient validation cohort, patient mortality was lowest where clinicians actual doses matched AI recommendations: mortality rates rose, in a dose-dependent manner, as clinicians actual decisions diverged from AI recommendations.

Thus, the 3 our technology s competitive advantages are that it can 1) extract relevant knowledge from our data sets, which are much larger than human clinicians could experience in a lifetime; 2) learn optimal strategies from suboptimal previous clinical decision examples; 3) optimise sequential decisions in real-time as patient conditions evolve.

Our AI technology requires now prospective validation in relevant, operational environments - we propose this in four ICUs across two NHS trusts.

Therefore, we will build a software platform capable of interfacing with many different NHS/ICU IT systems and running our AI algorithm in real-time, to provide dosing suggestions and their justifications to ICU physicians.

We will protect intellectual property and follow the processes for medical device software regulation to allow certification and commercialisation.