152 Automating the assessment of first seizure care pathways and clinical outcomes using electronic patient records

Purpose Electronic Patient Records (EPR) are ideal sources of information to provide evidence for the optimisation of health management. However, beyond direct patient care, their use for secondary purposes such as research or service improvement has been limited. Aims We aimed to explore the feasibility of using novel patient data analytics tools (CogStack & MedCAT) at King’s College Hospital trust (KCH) to identify suspected first seizure patients and compare their man- agement against NICE guidelines. Method We utilised CogStack to search 1.4 billion EPR documents at KCH to identify suspected first seizure patients. We then retrieved their subsequent records, and used a combination of manual extraction and a natural language processing tool (MedCAT), to extract information about their symptomatic presenta- tion, final diagnosis, timing of investigations, and specialist appointments. Results 226 patients attended the emergency department with suspected first seizures. MedCAT could be feasibly trained with expert knowledge to increase the accuracy of automatically extracting relevant textual information from EPR. Our analysis of patient records identified steps in the clinical pathway that frequently fell short of the NICE guidelines. Conclusion EPR are feasible to be mined at scale for the rapid analysis of service demand and monitoring patient health trajectories. The insights of this study have been used to improve first seizure management in KCH, demonstrating the value of NLP application in healthcare. anthony.shek@kcl.ac.uk 107