This blog was written for MQ by Dr Amy Ronaldson.
In May 2025, the MQ DATAMIND Data Science biannual meeting took place at the London HQ of Deutsche Bank. As an MQ Research Fellow, I was excited to attend and gain insights from leading researchers, clinicians, policy makers, and those with lived experience of mental health challenges.
About me
I’m Amy, an MQ Research Fellow using large amounts of routinely collected health data to understand infection outcomes in people with severe mental illness. Mental health data science is central to my work, making this meeting an invaluable space to exchange knowledge, share experiences, and learn from others in the field. I never miss it!
The event featured presentations, panel discussions, and Q&A sessions spanning career stages and disciplines. Several key themes emerged at this meeting which I believe reflect the direction of travel within mental health data science:
-
The role of Artificial Intelligence (AI) in mental health data science
With AI rapidly advancing, I was eager to hear how it’s being applied within mental health data science. Dr. Elizabeth Ford outlined how AI is currently being used, from administrative applications to forecasting patient needs and predictive modelling to innovations such as AI-driven therapy and medical scribes. While promising, significant concerns remain. Mental health data is often highly sensitive, recorded in unstructured formats, and can contain unexpected identifiers. This makes data protection and informed consent critical.
Public attitudes seem generally supportive of opt-out models if data is securely de-identified, but the nuances of mental health records—such as prior misdiagnoses and changes in diagnostic criteria—pose challenges for AI interpretation. Bias in records, particularly regarding LGBTQ+ individuals, homeless populations, and gender disparities, risks reinforcing existing inequalities.
Ford argued that while AI can assist clinicians, final decisions should remain with human experts to avoid exacerbating biases or unintended consequences.
-
Early Career Researcher Insights
For me, the early career researcher (ECR) flash presentations are always the highlight of MQ meetings. Showcasing the next generation of talent in mental health data science offers a valuable glimpse into the emerging trends and future direction of the field.
One key theme that emerged from the ECR presentations was the recurring challenge in mental health data science of polypharmacy and nuanced prescribing patterns. Flash talks touched on many aspects of this challenge from assessing drug interactions (e.g. metformin and antipsychotic-induced weight gain), to the application of large language models to measure patterns in antidepressant treatment. Huge efforts are being made to understand the best way to leverage prescribing data within mental health data science.
-
Machine learning versus traditional epidemiology
Professor Honghan Wu examined how deep learning models perform on mental health prediction tasks, combining structured electronic health records with unstructured text. Unstructured text within health records is a massive, somewhat untapped, resource within mental health data science. A panel discussion towards the end of the afternoon about the use of clinical text in research sparked much debate, particularly around machine learning vs traditional epidemiology. One key question emerged: Can AI outperform conventional methods when dealing with complex datasets? The consensus seemed to be that machine learning has advantages when it comes to handling large amounts of data but still needs careful oversight to ensure important nuance is not missed.
-
Data Sharing & the Future of Mental Health Science
Professor Andrew McIntosh gave a compelling talk on the future of collaborative mental health research in the UK. He presented challenges in data harmonization, noting that increasing dataset sizes through collaboration has unlocked insights into genetic underpinnings of psychiatric disorders. The discussion emphasized the importance of replicability, adequate sample sizes, and the invaluable efforts by organizations like MQ and DATAMIND to improve data governance.
