CRIS blog: Eight years on

The Clinical Record Interactive Search – CRIS – system is a unique mental health care database developed by the NIHR Maudsley Biomedical Research Centre (BRC).  Anonymised information from electronic patient records provides researchers with a unique set of data to use for research. A recent paper published in the journal BMJ Open summarises the lessons learned and progress made with CRIS since it was set up in 2008.  Lead author Professor Robert Stewart tells us here about the CRIS research journey:

In 2006, when our partner South London and Maudsley NHS Trust made the transition from paper to electronic health records, it was immediately clear that these had huge potential to be used in research.  As a result, in 2008 my colleagues and I launched CRIS.  For the first time, CRIS enabled researchers to search anonymised patient records, allowing them to spot trends and patterns in mental health treatment and service use, which had previously been all but invisible.

There were considerable challenges to overcome in building CRIS – not least, in creating a system that was useful to researchers without inconveniencing the busy medical professionals who use electronic records on a daily basis.  We also needed to make sure that records maintained patient confidentiality, which involved setting up not just a technical solution, but also a system of governance to ensure that projects using CRIS received appropriate oversight.  We were fortunate to be advised a committed group of service users, whose work continues to inform our development of the system, and who provide leadership of our ethical approval and governance processes.

Once CRIS was up and running, it was clear that there was a major inherent limitation, in that it was only possible to analyse specific ‘structured’ fields of the health record (i.e. simple facts like age, gender, diagnosis, and dates of referral, admission, or discharge).  The most important information in health records tends instead to be entered as text, particularly in mental health. This is the electronic equivalent of a doctor or nurse writing down details using a notepad and pen as they talk to a patient and discuss their experiences and history.

Analysing text information electronically is difficult – different people use different terms to describe the same things, they use abbreviations (especially when taking records in a medical setting), and, of course, there is the complexity of English language and grammar.  Because of the challenges inherent in analysing text, researchers using earlier versions of CRIS had to code information manually from text fields in order to identify trends and other patterns in treatment and service use.  This was an extremely laborious process, and essentially meant that studies using text fields were limited to no more than a few hundred case notes – a great frustration, given the hundreds of thousands of patient records contained on CRIS.

We have been able to get round these problems by turning to a field of computer science called Natural Language Processing, which allows computers to extract meaning from human (natural) language.  After several years of painstaking development, and working closely with expert colleagues at the University of Sheffield, we are now able to run a large number of Natural Language Processing applications within CRIS, and the result is that we have a mental health database of unrivalled depth and breadth. Not only do we have access to over a quarter of a million anonymised medical records, but we are also able to extract and analyse detailed information from text fields which was previously inaccessible to researchers, including fine-grain details of the symptoms people report, the treatments they receive, and the outcomes they experience.

At the same time, we have been working with organisations like the Department of Health and the Office of National Statistics to link the information in CRIS with other databases, such as those recording physical health, and treatments received, in general hospitals and general practice.  Understanding and treating any illness involves looking at the wider picture, and linking with these information sources will help us investigate what puts a person at risk of mental illness, and how it affects their life chances.

Our work to date using CRIS already forms a substantial body of research with over 70 published papers to date.  For example, we have provided accurate estimates of homelessness among inpatients on mental health wards; uncovered the abuse experiences of people who have been trafficked; investigated medication use and pregnancy outcomes in women with schizophrenia and bipolar disorder; and highlighted the lower life expectancy experienced by people with mental disorders and the reasons for this (such as the fact that level of disability and environmental circumstances contribute more than symptoms).  All of this work informs further research into how we can improve mental health treatment and services for everybody.

Despite the great strides we’ve made since 2008, our work with CRIS has only just begun.  In addition to current and future studies making use of our advanced text mining capabilities and the ability to link with other data sources, we have extended CRIS to other hospitals and health trusts via the D-CRIS programme.  In particular, we are now working closely with colleagues at University College London and University of Cambridge on a range of collaborative projects using CRIS and have developed a number of information-processing applications which could be used in any mental health trust regardless of whether it has set up the CRIS platform.

Having created what we believe to be the world’s deepest and widest-ranging mental health records database, we see no reason to stop.  We look forward to expanding our horizons, uncovering new findings from CRIS and its descendants, and making sure that these are translated into improvements in the treatment and care of mental disorders.