Addressing variation in Mental Health Outcomes - From Inequalities data to ‘Caseness’
As part of our OpenDataSavesLives series, we hosted a two day workshop at Open Innovations in Leeds, where we set out to address the data around inequalities and how we could use it to understand variations in mental health outcomes by geography.
We started with two possible research questions:
Challenges with the Data
A search for data began, which characterised what is common within the NHS; if the exact data is not available to answer the question, we must look for the data that is closest to help give us answers. We then find that different datasets are not recorded in the same way, making it difficult to analyse and compare. Variations in simple things like age bands, to more complicated variables such as ethnicity right through to different diagnoses conventions such as Reed codes, ICD-10 codes and Snomed-CT codes. While there are attempts to make this simpler, The NHS Digital Terminology Server - NHS Digital for example, it remains difficult and an initial barrier to analysis.
Because of these difficulties, a new management science is growing to try and make chaotic conversations a bit more orderly, which currently looks a bit like this:
What data is available?
- Some data is available, but this data doesn’t actually answer the question.
Well that's all I have.
- Well maybe we could just grab it all and see if anything useful comes out?
If we published where we found it and some of the problems we encountered then it might make it easier for the next person.
- It might…
OR we could use Nesta’s model, which gives us a helpful model to check against.
In our example, the specific problem is relatively clear - the NHS would like to assure itself that services are being offered equitably in these three areas:
- Patient experience
The defined action is less clear in our example - if we do find out that people of a certain ethnicity wait longer for treatment, what could be done beyond alerting the system of this finding? Trusts or GPs might simply do case-finding; going through one by one those patients identified as being at higher risk or using the now more common population management methodologies, designing interventions to target particular communities.
The clear data product required is also relatively simple to describe. We would, for example, like to know for each patient their characteristics (age/sex/ethnicity/learning difficulties) and a range of access/patient experience/outcome data for them. However, because of information governance, this is very difficult to get access to. Instead, aggregated data is reported; rates of invasive treatment by ethnic group, levels of ‘caseness’ (severity/acuity) by CCG and so forth.
Acknowledging this and moving to accessible data, an attempt has been made to collect, describe and present three datasets for mental health; IAPT, CMHS and MHSD. Just looking at their levels of geography and age bands used, it is possible to illustrate how difficult it is to link them:
|Description||Level of Geography||Age|
Data containing information on activity, waiting times and outcomes from talking therapies
Clinic / Trust Level
Specifies how many child referrals but not age range
Data on children, young people and adult in contact with secondary services for mental health and wellbeing, learning disability, autism and other neurodevelopmental conditions
0 to 18, 18 and over
|CMHS||Patient experience survey using mental health survey between months Sept-Nov||Trust||18-35, 36-50, 51-65, 66+|
Alongside these datasets, the innovation process led to a search for other potentially useful datasets to work with. This led us to consider:
- The linked datasets that are being built locally (KID and Kernel for example in Kent (www.kmkernel.org);
- Data on ethnicity beyond the census ethnicity categories from www.originsinfo.eu;
- Data from regional care record providers, Graphnet for example, www.graphnet.com.
Answering the Question with Data
Over the two days, we were able to pull together the sources above into a visualisation.
Mental Health Data Explorer tool
This tool explores the relationship between demographics, such as population and ethnic origin, and outcomes across England, summarised by Clinical Commissioning Group. A heat map allows for regional comparison and the correlation plot beneath highlights any potential correlations between metrics from the two drop-down menus.
The visualisation was based on previous work by Open Innovations, exploring relationships between demographic and Covid-19 vaccination rates at the MSOA level. Some inconsistencies in reporting led to some data being lost and there was conversation around deciphering the ambiguity of labels and descriptions in government datasets. However, we mapped over age and geographic level as closely as possible and, given the challenges described above, pulled it all together in a short space of time. It demonstrates what can be done right now with open data and existing tools, and the insight that can be gained simply by doing the work, bringing people together to innovate over two days.
In the end, delivering the workshop has helped to paint a picture of the landscape for mental health open data and highlighted the challenges of navigating information governance processes around health statistics. We hope to explore this topic further throughout the year as the mission of #OpenDataSavesLives continues.
Please take a look at the tool, interact with it and see what relationships you can find by viewing our Github repository and visualisation tool. The next question we ask is - what does the data tell us, and what can we do about it?