Author name: Benjamin Holgate
‘Insights from HealTAC 2025’s Mental Health and the New NLP DATAMIND Workshop’
Introduction
How recent shifts in Natural Language Processing (NLP) have and will in future affect mental health research and care was the central theme of a ground-breaking workshop held in Glasgow last month.
The ‘Mental Health and the New NLP DATAMIND Workshop’ was held on 16 June as a prelude to the annual HealTAC conference, this year located at the University of Glasgow in Scotland. The workshop was the first of its kind for HealTAC, a forum that brings together academics, clinicians, industry, and patients to discuss current trends and techniques in analysing healthcare texts.
The workshop, organised by Dr Angus Roberts, Senior Lecturer in Health Informatics at King’s College London, was attended by about 80 people, who reflected the mix of professional and patient backgrounds at the conference. The all-day workshop involved 13 talks by 15 speakers (including two dual-speakers) who came from academia, industry, and patient groups, with each talk lasting about 15 minutes. Audience participation through Q&A at the end of each of the four sessions throughout the day was both lively and engaged.
Key Themes
The aim of the workshop was to gain insights and promote discussion around how new NLP techniques are influencing research into mental health and the treatment of patients. Perhaps inevitably, the dominant topic was the application of Large Language Models (LLMs), which have emerged over the past few years as a powerful and flexible NLP and analytical tool. LLMs featured in nine talks, or 69% of total papers. As Dr Marcos Del Pozo Banos, Senior Lecturer in Health Data Science at Swansea University, pointed out in his opening talk that the invention of the pre-trained Transformer architecture a decade ago, and the release of BERT, ChatGPT-1 and other early LLMs soon after, have since led to a proliferation of increasingly powerful and larger LLMs that have revolutionised NLP and AI.
Different speakers spoke to a wide variety of LLM applications in mental health research and treatment. For example, Dr Del Pozo Banos told how therapeutic chatbots are being used to deliver scalable psychological support for patients. This application was reinforced by Dr Maximilian Droog-Hayes, Principal AI Scientist at ieso Health, who detailed how his company has developed a therapeutic chatbot for mental health patients that responds in a personalised and empathetic manner.
Other speakers told of how LLMs can help with extracting information from medical texts for temporal reasoning, in order to track changes in a patient’s condition over time (Dr Tao Wang, Research Fellow in Heath Text Analytics and Data Science, King’s College London), as well as how LLMs are assisting in gaps in a patient’s depression timeline by identifying under-coded generic consultation encounters relevant to the depression phenotype in primary care text data (Matthew Iveson, Senior Research Fellow, and Matúš Falis, Research Fellow and Associate NLP Analyst, at University of Edinburgh).
Notwithstanding the near-ubiquity of LLMs in NLP medical research, other topics were highlighted throughout the workshop. For instance, Arlene Casey, Vivensa Foundation Senior Research Fellow at the University of Edinburgh, raised the issue of privacy risks for researchers working with texts that contain patients’ mental health narratives, especially with references to things such as suicide, depression, or trauma. Gloria Roque, AI Product Owner, and Jack Richmond, AI Scientist, at Akrivia Health discussed the need for gold-standard annotations and evaluation methodologies when measuring NLP models given the high standards required for safe and accountable AI in healthcare. Dr Darren Cook, Research Fellow at City St George’s / VISION Consortium, talked about how various NLP tools are aiding research into domestic violence, especially in relation to women, based on data recorded by police.
Another area covered in multiple talks was the different kinds of datasets that are being used to facilitate NLP-oriented mental health research. Two speakers separately illustrated the evolution and importance of the CRIS (Clinical Research Interactive Search) dataset from South London and Maudsley NHS Trust, which contains about 30 million documents and has helped generate more than 300 research papers (Jytoi Sanyal, NLP Lead – Operational at SLaM NHSFT, and Prof. Rob Stewart, Professor of Psychiatric Epidemiology and Clinical Informatics at King’s College London). Other kinds of datasets covered included insurance claims (Dr Rashmi Patel, Assistant Professor in Real-World Data Analytics at the University of Cambridge), police records, and GP primary care reports.
A highlight of the workshop, however, was an insightful talk on the patient’s perspective given by Dr Jane Taylor, a Patient Advocate in the DATAMIND Super-Research Advisory Group. She canvassed several issues arising from the use of NLP with electronic health records, such as: filters used by both patient and health professionals before, during, and after clinical encounters; what might be missed or misinterpreted in transcriptions of clinic visits; and bias in regard to status, gender, ethnicity, and sexuality.
Summary
A key takeout from the workshop was that NLP has expanded far beyond just being a tool to extract information or analyse data from medical texts. Rather, NLP has evolved into a multi-faceted field that also assists in improving patient treatment and therapeutics, often in a personalised manner. The workshop consensus was that rapid innovations in AI, especially LLMs and generative AI, will accelerate this shift.
The workshop’s speakers and their abstracts can be found at https://healtac2025.github.io/workshop/.
