Harmonised Data: Linking Mental & Physical Health

We wanted to know:

• What EHR records are available in each of the four nations of the UK?
• Can we use data collected as part of existing longitudinal and cohort studies to learn about the physical health of people with severe mental illness (SMI)?
• How do different parts of the UK define SMI, and how do they record it?
• Do EHRs in the UK have important health information for people with SMI, like BMI, smoking status, blood pressure, cholesterol levels, blood sugar, and HbA1c? If so, how well recorded are these?

Key Findings

Numerous datasets have relevant health information across the UK. These range from local repositories to national databases like the SAIL databank covering Wales. However, no single resource provides complete coverage across all four UK nations. Despite advances in EHR accessibility, such as the introduction of trusted research environments, challenges persist. For example differences in the ways SMI is defined. Key information, such as BMI, smoking status and blood pressure are well recorded in EHRs for people with SMI, but not always regularly. While some cohort or longitudinal studies may provide information on people with SMI, only those with a large number of patients or focus on mental health will be suitable for research.

Conclusions

While UK EHRs offer a rich source for mental health research, inconsistencies in SMI definitions and data structures hinder research into of physical health aspects of people’s lives. Harmonising data resources, improving access to data, and standardising definitions are essential to propel research forward.

Ready to dive deeper? Download or view the complete report for comprehensive details on methodologies, results, and recommendations.

Key resources

Cohort resources

• The UK Longitudinal Linkage Collaboration (UK LLC) provides a record linkage service to UK Longitudinal Population Studies and a secure mechanism for researcher access.

EHR resources

• The HDR UK phenotype library provides phenotyping algorithms and code lists for EHR research

We wanted to know:

• How feasible is it to replicate studies on severe mental illness and physical health across EHR databases covering England, Scotland and Wales?
• What are the key differences and challenges in data availability and structure across these databases?
• How complete is the recording of key cardiometabolic risk factors, such as BMI, smoking status, blood pressure, cholesterol, blood glucose, and HbA1c across the three databases?

Key Findings

The CPRD, SAIL and DataLoch databases provide a rich resource for research to improve the physical health of people with SMI. While CPRD covers practices in England, Scotland and Wales, it only a subset of GP practices in these regions. For Wales and Scotland, a lack of data on the location of practices makes it hard to determine how representative these practices are. In contrast, SAIL and DataLoch contain data for almost all practices in the Wales and Lothian in Scotland, respectively.

We found that across the three databases BMI, blood pressure, and smoking status were well-recorded (>70%), but cholesterol and glucose were recorded less often. While the majority of people with SMI had CVD risk factors recorded at least once, regular recording was less common. Researchers should be aware that completeness of CVD risk factors also likely vary by patient characteristics.

In this report, we used code lists generated in CPRD. It is therefore perhaps unsurprising that recording of CVD risk factors was generally more complete in CPRD. While our code lists likely identified the majority of observations in SAIL and DataLoch, it is clear that for optimal case finding code lists should be developed specifically for each data source. Other sources of differences in CVD risk factor recording likely reflect differences in database structures, geographic coverage, clinical practices, and incentivisation schemes. , Differences in coding and data structures make direct comparisons across databases challenging and further work is required to improve capabilities of federated data analysis across a range of data sources, both at regional and national levels.

Conclusions

These datasets offer valuable insights into severe mental illness (SMI) and cardiovascular disease (CVD) risk factors, though differences in data structure, coding, and recording practices may affect comparability. Key risk factors like BMI, blood pressure, and smoking status were generally well recorded, while cholesterol and glucose data were less consistent. Understanding these variations can help researchers adapt their approaches for more reliable analysis. While challenges remain, these data sources hold significant potential for improving knowledge on the physical health of people with SMI, with further efforts needed to enhance data harmonisation across the UK.

Ready to dive deeper? Download or view the complete report for comprehensive details on methodologies, results, and recommendations.

Key resources

Cohort resources

• The UK Longitudinal Linkage Collaboration (UK LLC) provides a record linkage service to UK Longitudinal Population Studies and a secure mechanism for researcher access.

EHR resources

• The HDR UK phenotype library provides phenotyping algorithms and code lists for EHR research