Improving outcomes of sepsis, using precision antimicrobial prescribing

Lead Research Organisation: University of Bristol
Department Name: Electrical and Electronic Engineering


This project intends to identify patients at risk of sepsis earlier, predicting their severity and clinical outcome; and provide informed and timely decisions on interventions such as antimicrobials and supporting care. It will answer the question of "what antibiotic, when, and in whom?". It will evaluate and potentially integrate scoring systems such as Sepsis-related Organ Failure Assessment score (SOFA). By linking data on patients presenting to acute care with sepsis with their primary care data, predictive values of different sepsis indicators / prognostic markers will be studied based on their comorbidities and pathway to care.

One in five deaths worldwide is caused by sepsis. Early identification and timely intervention are key to increased survival, but diagnosis can be difficult in the early stages, especially at extremes of age, and clinical manifestations can be non-specific, depending on the pathogen type, portal of entry, and progression of the condition. Potential benefits would reduce mortality and reduce length of hospital stay due to timely interventions, including the administration of working antibiotics as well as the effect of reducing population antimicrobial resistance. In the study region, 5,000 patients will benefit per year, with an estimated 200,000 cases and 50,000 sepsis deaths in the whole of the UK.

Clinical decision support system (CDSS) support for antimicrobial prescribing has previously been based on static models built with expert guidance from published evidence. Now for the first time, underpinned by Bristol's new Health Data Research UK South-West Partnership, antimicrobial use and resistance histories of each patient across primary and secondary care can be integrated with local susceptibility testing and genomic analysis of resistance mechanisms. Hence this will lead to more in-patient precision through a step-change in breadth and depth of data available to train novel models - namely, all individuals in the Bristol, North Somerset and South Gloucester (BNSSG) clinical commissioning area. The methodology is also novel, using cutting-edge Gaussian Process classification models.

For the first year, focus will be on patient data at presentation, integrated with population-level predictive modelling. There will be "sandpit" events to consider user-requirements for the clinical decision support tool.

In the second year, identifiable data and predictions back to individual providers will enable patient-specific AM use and AMR histories from primary and secondary care to be integrated into the clinical decision support model.

In the third year, there will be co-design of dashboards for the CDSS with clinicians, and the CDSS will be evaluated in a clinical setting.

The project will be based on an underpinning platform of Bayesian learning and prediction. This is now possible because of the HDR-UK partnership and BNSSG Systemwide dataset, which has integrated routine primary and secondary care electronic healthcare records with the Severn Pathology laboratory database (the diagnostic test provider for BNSSG). There will be statistical analysis and machine learning applied to assess relation between National Early Warning Score (NEWS) and SOFA score at baseline and on presentation during a sepsis episode, as well as dynamics of NEWS and SOFA score in relation to interventions.


10 25 50

Studentship Projects

Project Reference Relationship Related To Start End Student Name
EP/S023704/1 01/04/2019 30/09/2027
2270575 Studentship EP/S023704/1 23/09/2019 22/09/2023 Edward Barker