Every year, thousands of brain injury patients in UK rehab units develop infections that their clinical teams fail to spot in time. Acute brain injury suppresses the immune system, and patients often cannot communicate symptoms clearly. Standard early-warning scores, designed for general hospital wards, miss the subtle vital-sign changes that signal infection in these patients. This delay allows infections to escalate into sepsis, causing worse outcomes and prolonged hospital stays. Sanome has developed a machine-learning tool called MEMORI that continuously analyses patient data—comorbidities, vital-sign trends, and other clinical information—to flag infection risk hours before conventional methods would. Already deployed at a specialist neuro-disability hospital, the tool outperforms the current standard of care. The team now plans to improve its performance by adding blood results, pathology reports, and clinical notes, giving clinicians an even earlier warning. If successful, MEMORI could reduce infection-related deaths and complications in brain injury patients, cut emergency admissions from rehab units, and lower treatment costs for the NHS. For patients and families, it means fewer crises and better recoveries.
View original technical description
More than 1.3M people in the UK live with Acute Brain Injury (ABI), costing the NHS c£15B per year. ABI patients are highly susceptible to healthcare-acquired infections (HAIs) during rehab due to compromised immune systems[1]. Over 15% of ABI patients develop an HAI, with mortality rates of 20-30%[2]. HAIs in ABI patients lead to worse outcomes, higher care costs (over £20,000 per incidence)[3], and patient distress[4]) . Studies show that 30-55% of HAI impacts could be avoided with earlier detection[5]. Due to communication difficulties and atypical vital signs, HAIs are hard to diagnose in ABI patients. Therefore, tools like the National Early Warning Score 2 (NEWS2) are less effective, leading to delayed interventions and worse outcomes. Sanome's MEMORI.v01 is a Machine Learning-model that estimates a patient's risk of developing an HAI. Co-designed with the Royal Hospital for Neuro-Disability(RHN), a leading ABI care facility, it has been deployed within the hospital's Electronic Patient Record system since May-24. The tool, incorporating patient meta-data, comorbidities, and recent trends in vital signs, continuously evaluates subtle changes in the data to alert clinical teams to potential infection risks, therefore giving them a "window of opportunity" to intervene earlier. Early clinical results are positive indicating that MEMORI outperforms NEWS2, the current clinical practice[6]. Feedback from clinical teams at both the RHN, as well as other ABI rehab centres (including Cambridge, Portsmouth, Devon and East Kent) suggest that clinical adoption and ultimately patient outcomes could be improved further, if clinicians were to be given a greater "window of opportunity". Literature indicates that incorporating additional data modalities, such as blood results, pathology reports, and clinical notes, could enhance model performance[7] and therefore provide a greater window of opportunity. This presents a significant opportunity: Patients benefit from earlier interventions and therefore better outcomes The NHS experiences reduced care costs through fewer A&E admissions and step-ups from neuro-rehab/ ABI facilities, as well as fewer complications such as sepsis Sanome gains increased revenue through faster adoption and greater realised benefits*. The UK benefits from innovation, tax revenue, and NHS cost savings *includes shared benefit agreements with hospitals. By enhancing the model's performance, we deliver greater value, accelerating revenue growth and adoption. We regularly conduct PPI workshops and patients and relatives were enthusiastic about using their data to continue to develop MEMORI. A key success factor for this project is the high-levels of clinician engagement at the RHN, as well as readily accessible data via our HRA and CAG-approved research database (REC:4/NE/0008).
Plain English summaries and category classifications on this site are generated by AI and may not perfectly reflect the original research.
Is something wrong? Let us know