CRIS blog: Using CRIS to evaluate mental health diagnoses in routine national statistics

Dr Katrina Davis is a Research Associate working with Professor Matthew Hotopf and with the UK Biobank Mental Health Outcomes Consortium. She will be starting a BRC PhD studentship in October this year, working with the CRIS database.

Many health research studies make use of information originally collected for administrative purposes as well as, or instead of, collecting this directly from patients. This offers advantages in terms of avoiding burdening participants with extra visits and interviews. It is also helpful in making sure that follow-up is as complete as possible, and is often more cost-effective and efficient. However, researchers do need to check carefully the accuracy of this sort of administrative information and how well it corresponds to real health-related outcomes. As described in a recently published paper in PLOS ONE, we investigated this for mental health diagnoses recorded on a national hospital database.

The UK Biobank project is currently following around 500,000 volunteers from around the UK. Because it is difficult to re-interview that many people, the project relies a lot on routine health records information, particularly the NHS England database of hospital activity called Hospital Episode Statistics (HES). This includes information on diagnoses (classified using the ICD-10 coding system) for each inpatient episode. There have been previous audits on the accuracy of the diagnoses in HES records compared to clinical notes, but these have not included diagnoses from mental healthcare. Our aim was to assess the accuracy of HES diagnoses for three major mental health diagnoses: i) schizophrenia and related disorders; ii) bipolar disorder; and iii) depression.

CRIS was an ideal platform for this research, as it includes a linkage between detailed mental healthcare records and HES. For example, it was possible to check individual symptoms recorded in the notes, and then read and evaluate these episodes to build up a broad longitudinal picture of mental health. In 250 people across the three mental healthcare groups who had been admitted to a mental health hospital at least once, a “research” diagnosis was assigned following a detailed record review, which was then compared against HES diagnoses they had received.

All patients whose case notes were examined were confirmed to have a mental health disorder. Those who received any HES diagnosis of schizophrenia, bipolar disorder and unipolar depression had this confirmed by the research diagnosis in 74%, 65% and 61% of cases respectively. Combining schizophrenia and related disorders with bipolar disorder to create a “severe mental illness” group in HES correctly identified the same group of disorders 89% of the time. Performances were better if only the most recent mental health diagnosis from the HES record was used.

We were therefore able to conclude that the accuracy of mental health diagnoses in people with at least one mental health admission was reasonable, particularly when diagnoses were grouped together rather than looked at individually. The accuracy was also in a similar range to that of physical disorder diagnoses on HES. Our findings suggest that HES records are suitable for identifying people with mental disorder severe enough to precipitate admission, and could possibly be used with caution to identify specific diagnoses. In the future we hope to compare HES to other outcome measures, such as GP records and questionnaire results, to build up a richer picture of mental health for large national projects like UK Biobank.

Read the paper online at: Using data linkage to electronic patient records to assess the validity of selected mental health diagnoses in English Hospital Episode Statistics (HES) https://doi.org/10.1371/journal.pone.0195002