A blog by Co-chair Joseph Lam summarising parts of the discussion in a Q&A session held on 13 August 2025 on behalf of the ITALO interest group.
In a recent Q&A I organised for a community of researchers working with the Mental Health Services Data Set (MHSDS) across multiple Trusted Research Environments (TREs) and linked data resources —including ECHILD, UK LLC, and other MHSDS-to-cohort linkages, Tom Bardsley (NHS England) provided valuable guidance on interpreting and using this complex dataset, drawing on his decade of experience.
A key reminder from the outset: the MHSDS is not HES. Unlike the Hospital Episode Statistics, which focus on discrete episodes of admitted care, the MHSDS is referral-based, covering a far broader scope—community, inpatient, crisis, learning disability, and some autism/ADHD activity. Every record links back to a referral, and a single referral can appear in multiple monthly submissions until it is closed.
Discharge dates are generally reliable, but not infallible—some are estimated in advance, others missing. In absence of a discharge date, analysts have used the last month in which a referral appears as a proxy for inactivity. Seasonal effects can sometimes be observed in some services: providers may carry out end-of-year “spring cleaning” in March, closing long-inactive referrals before the new financial year. This can produce abrupt changes in recorded caseloads when monitored on a month-to-month basis.
While all providers are expected to submit rejected referrals, coverage can be inconsistent. Individuals may hold multiple simultaneous or sequential referrals for the same condition, reflecting variations in local service pathways.
Diagnosis records can be duplicated across months for ongoing referrals, appear in multiple coding formats (ICD-10, SNOMED), or legitimately reflect more than one primary diagnosis.
Over time, the number of submitting organisations has expanded from around 60 large NHS trusts to approximately 400, with independent and voluntary sector providers particularly contributing to growth in children and young people’s mental health data.
Finally, Tom advised against relying on the Referral To Treatment table for waiting time analysis; instead, derive these measures by linking referral and care contact data, being mindful of service-specific definitions of “treatment start.”
Call to action
The insights shared underscore the value of peer learning across TREs such as ECHILD, UK LLC, and others. We encourage analysts, researchers, and practitioners working with linked health data to engage with the ITALO network and the wider community of MHSDS users. By sharing methods, clarifying data nuances, and pooling expertise, we can collectively improve the quality, transparency, and impact of research using these rich but complex data.
This blog reflects my own interpretation of the discussion and does not represent the official position of NHS England, the MHSDS team, or any TRE. It should not be regarded as definitive guidance, and no responsibility is taken for its use. MHSDS content, coverage, and structure are subject to change over time.
Joseph Lam
Co-chair of ITALO
Research Fellow, UCL Great Ormond Street Institute of Child Health
Data Science Officer, NHS England