A methodological comparison study of the use of Natural Language Processing within Mental Health Services in the National Health Services in the United Kingdom (Preprint)

Abstract

UNSTRUCTURED Background: Mental illness has a high disease burden within the UK, attributing to 22.8% in comparison to Cancer (15.9%) and Cardiovascular disease (16.2%). Costs of mental illness in England have been evaluated at £105.2 billion each year. This burden could be reduced by effective use of Electronic Health Records that could provide vital information around diagnosis, prevalence and incidence of mental illnesses to better understand the nuances of clinical and patient reported outcomes. To better evaluate some of the technical methods that could be better used, we explored Natural Language Processing. Objective: Our primary objective was to evaluate the use of Natural Language Processing methods and it’s association with unstructured EHR text data from U.K.-CRIS. Methods: We used a descriptive methodology to demonstrate the use of NLP and validated the method using Southern Health NHS Trust electronic health data. Conclusions: We can conclude that the method used is suitable for the mental health service. However, to generalize our findings, a wider validation study across mental health organisations in the UK would be required.