Background Recognising deteriorating patients in acute hospitals, timely review and critical care admission are the focus of current NHS policy implementation (‘Martha’s Rule’). In alignment with the Scottish Government’s Data Strategy for Health and Social Care (2023), which seeks to improve health and wellbeing using data, we aimed to establish an innovative, data-driven programme to measure and improve care quality for patients referred to critical care.

Methods We established data infrastructure for identifying critical care referrals across a region in Scotland (NHS Lothian, population ~900,000), reporting baseline results and implementing a quality improvement (QI) programme of work to improve care and benchmark performance against national standards set out by UK professional societies (Faculty of Intensive Care Medicine, Intensive Care Society). Working with local NHS informatics teams we used free text recognition to identify critical care referrals documented in electronic health records (EHR). Since most information was only available as unstructured data, we used free-text analysis supplemented by manual casenote review in a subset and linkage with data from the local ICU audit database to develop a data pipeline for benchmarking against national standards. We automated data wrangling and reporting using R programming language, and reported a set of baseline results as a basis for a QI programme. The project was approved by the local QI committee.

Results From 05/07/2022-20/08/2023 there were 1242 referrals documented (mean 3/day). Through linkage between EHR and audit database, we were able to identify that only 42% of admissions had a documented referral. Free-text analysis substantially reduced time to review casenotes. Benchmarking against national standards, among the subset of 223 with manually collected data, in 42% of cases there was documented consultant involvement in decision-making; 93% of patients were seen within one hour of referral; 69% were admitted within 4 hours of decision-to-admit; 77% for whom admission was of no overall benefit had a documented Treatment Escalation Plan recommendation.

Conclusion We have established a data-driven system to efficiently identify and report key aspects of critical care referrals and drive improvements in care quality. Limitations include not identifying referrals missed by free-text recognition, and reliance on manual data collection. Additional data linkage is planned to minimise the burden of manual data collection, and a programme of QI work is underway employing this newly established data platform to drive improvement. This preliminary work shows the potential of using data-enabled systems to efficiently measure the impact of policy implementation.