DPhil Studentship: Language-grounded Causal Inference from Electronic Mental Health Records
[Scholarship covers course fees up to the value of home fees, a tax-free stipend of at least £20,780 per annum, plus additional support for research expenses, conference attendance, and consumables. Overseas students must fund the remainder of their fees unless additional support is granted at the funder’s discretion.]
The University of Oxford's Department of Psychiatry, in association with St Cross College, invites applications for a fully funded DPhil (PhD) studentship focused on language-grounded causal inference from electronic mental health records. This 3-year doctoral programme, commencing October 2026, aims to address the challenge of estimating causal effects of psychiatric treatments from observational data, particularly where confounding variables are richly documented in clinical free-text but rarely captured in structured records.
The project comprises three interconnected studies: (1) distilling knowledge from advanced large language models (LLMs) into small, secure local language models (SLMs) for extracting clinically relevant variables and phenotypes from mental health records; (2) comparing SLM-extracted variables with expert judgement to identify confounders and evaluating causal discovery algorithms across datasets; and (3) tasking SLMs with designing emulated trials, benchmarking against gold-standard emulations and expert assessments. The overarching goal is to determine whether privacy-preserving language models can reliably support large-scale causal inference in psychiatry.
Supervision will be provided by Dr Andrey Kormilitzin (University of Oxford) and Professor Erin Evelyn Gabriel (University of Copenhagen), with support from the SMARTBiomed consortium. The studentship offers multidisciplinary training in language models, natural language processing, statistical machine learning, and causal inference, as well as access to large-scale psychiatric datasets and structured researcher development through SMARTBiomed.
The scholarship covers course fees up to the value of home fees, a tax-free stipend of at least £20,780 per annum, and additional support for research expenses, conference attendance, and consumables. Overseas students must fund any fee shortfall unless further support is granted at the funder’s discretion.
Applicants should hold or expect to obtain at least an upper second-class honours degree in a relevant field (medical science, statistics, or machine learning), with strong Python programming skills. Prior experience in causal inference and clinical NLP is advantageous. Applications must be submitted online, including all supporting materials and a £20 application fee, by 12:00 midday (UK time), Friday 10 April 2026. Informal enquiries are welcome via email to Dr Andrey Kormilitzin.
