People with severe mental illness (SMI) including schizophrenia and bipolar disorder, have life expectancies up to 15 years shorter than the general population. Little progress has been made over the past twenty years in closing the gap in this population.
Most of the important information on physical health conditions in psychiatric hospitals is stored in free text, which is very difficult for computers to ‘understand’. Natural language processing (NLP) toolkits – such as SemEHR – utilise artificial intelligence technologies to ‘read’ such data so that computers can make the best use of this ‘hidden’ information. For example, SemEHR has been used to study multimorbidities (>2 physical conditions) of SMI patients in South London and Maudsley hospital and showed promising performance.
In this exemplar project, we will use different NLP approaches, including SemEHR, on different types of free text documents including GP Referral letters and discharge summaries at different hospitals across the UK. The aim is to estimate the frequency of physical health conditions and, more importantly, identify important and detailed information about them. Such information will be combined with other types of data (such as national data registries) to get a comprehensive picture about patients so that clinicians can accurately infer the effects of multimorbidities on outcomes and mortality of SMI patients.
This study will generate a valuable national research resource for patients with SMI detailing physical multimorbidity and outcomes e.g. premature mortality. Such a resource will support development of strategies for better treatment and management of people with SMI.