The DATAMIND & MQ ECR Workshop, Unlocking Research Insights and Harmonising Clinical & Survey Data, recently took place in London. In this blog, Hatem Mona, Data Scientist and Psychiatric Epidemiologist at Swansea University and DATAMIND, reflects on key takeaways from the day, including the challenges and opportunities of integrating varied data sources and how tools like Harmony can support this crucial work.
I recently attended a thought-provoking workshop focused on data harmonisation and its potential to transform mental health research. With a growing amount of electronic health data available across the UK, the session highlighted the pressing need to make these diverse sources more usable, accessible, and interoperable. The key focus was on harmonisation: the process of standardising data and metadata to ensure better quality, consistency, and integration across datasets. This is especially vital in mental health research, where diverse data – clinical, digital, and patient-reported – can unlock new insights when brought together
Exploring the Range of Data Sources
One of the major themes of the workshop was the wide variety of data sources available to mental health researchers. From electronic health records and primary care datasets like Clinical Practice Research Datalink (CPRD), to digital health tools, social media, and national cohort studies, the opportunities are vast. Each source offers unique insights: clinical data provides diagnoses and treatment history, digital tools capture behavioural patterns, and patient-reported outcomes reveal lived experiences and sensitive personal data.
Harmonising these diverse data streams can open new doors for research. For example, combining large-scale UK datasets enables comparisons across regions and ages and allows researchers to capitalise on natural experiments and study rare mental health conditions with improved statistical power. Stratified analysis by risk factors becomes more feasible with larger, harmonised samples, and cross-study comparisons can help validate findings across different populations, contexts, and healthcare systems.
Challenges in Data Integration
However, the path to integration is not without obstacles. A recurring challenge is the conceptual misalignment across datasets – particularly how key mental health constructs like severe mental illness (SMI) are defined and recorded. Unlike physical health measures like blood glucose, MH conditions are often described in less standardised ways, leading to inconsistencies in coding and diagnosis.
A presentation by Naomi Launders at the DATAMIND & MQ ECR Workshop outlining conceptual and structural barriers to data harmonisation in mental health research.
Structural barriers further complicate integration: data are stored using different coding systems (e.g., SNOMED, ICD-10), housed in various formats, and subject to different access protocols (like Trusted Research Environments). The CPRD dataset, for example, lacks full clarity on representativeness, making generalisability difficult to assess. Another key takeaway was the trade-off between longitudinal depth and population coverage (between linked cohort and longitudinal cohort studies respectively). While some datasets offer rich follow-up data over time, they may lack national representativeness, and vice versa. Harmonisation also doesn’t always support patient-level analysis, which can limit how findings are interpreted or applied. Researchers must also navigate practical hurdles – finding the right dataset, accessing it, and understanding local clinical and contextual nuances.
The Role of Tools Like Harmony
Tools like Harmony offer promising solutions to many of these challenges. Harmony was presented as a resource for retrospective data harmonisation – allowing researchers to align variables, recalibrate measurement scales, and standardise item-level data across multiple datasets. What stood out to me was how Harmony not only simplifies data structuring but also supports analysis. It allows for consistent variable definitions, helping bridge the gap between differing data structures. Natural Language Processing (NLP) was highlighted as a powerful companion to harmonisation efforts – transforming free-text clinical notes into structured symptom scales that can be used in research. These tools are vital for enabling cross-cohort analyses, particularly in studies requiring large sample sizes (e.g., rare disorders or stratified subgroup analyses). Harmonisation through platforms like Harmony allows for more accurate and meaningful comparisons, ultimately increasing the robustness of mental health research findings.
Conclusion
This workshop deepened my appreciation for the power of data harmonisation in mental health research. Despite the conceptual and structural hurdles, harmonisation can significantly enhance the quality and scope of our analyses, particularly when tools like Harmony and NLP are employed. Personally, I came away with renewed motivation to explore how these strategies could support my work using longitudinal cohort data to study self-harm and suicidality. Understanding how to standardise variables across waves and sources will be crucial for drawing meaningful conclusions and informing interventions.
Get Involved
If this sparked your curiosity, don’t miss the next workshop! It’s a great opportunity to dive deeper into these tools and strategies and continue learning about how we can transform mental health research through data. To stay updated on future events, follow @DatamindUK and @MQmentalhealth on X (formerly Twitter), or visit www.datamind.org.uk for more information and updates.
