KConnect is developing sophisticated search and text analysis techniques to improve medical decision-making and care. The project involves researchers, commercial search experts and practitioners in health organisations from across Europe providing expertise in clinical health informatics.
The project aims to create a medical text Data-Value Chain – a series of steps to support decision-making – using commercial cloud-based services for multilingual Semantic Annotation, Semantic Search and Machine Translation of Electronic Health Records and medical publications.
Services are adapted to search through, and make sense of, free text in Electronic Health Records – a particularly challenging goal, due to misspellings, new words or phrases being adopted (neologisms), organisation-specific acronyms, and heavy use of negation (for example “not active” rather than “sedentary”) and use of vague or cautious descriptions.
The KConnect semantic index at South London and Maudsley NHS Foundation Trust is based on a case register derived from the Trust’s full patient record, via the Clinical Record Interactive Search (CRIS) database.
The system will initially be used to develop predictive models of adverse drug events, drawing on published literature relating to medications as well as analyses carried out within the records database. Records data are analysed up to the point of the adverse drug event, generating a timeline which will then be compared with control timelines of persons with similar characteristics who have not experienced an adverse reaction.
Predictive models from records text can then be compared with those drawn from the literature (e.g. on drug interactions), and the different models that arise are assessed. The result is a better understanding of who will be affected by adverse drug reactions, allowing patients and their clinicians to make more informed choices about their medication.
- Honghan Wu, Zina M. Ibrahim, Ehtesham Iqbal and Richard JB Dobson. Predicting Adverse Events from Multiple and Dynamic Medication Episodes. Accepted by AI-2016 Thirty-sixth SGAI International Conference on Artificial Intelligence. CAMBRIDGE, ENGLAND 13-15 DECEMBER 2016.