Objectives: Creation of linked mental health, social and education records for research tosupport evidence based practice for regional mental health services.
Setting: The Clinical Record Interactive Search (CRIS) system was used to extract personal identifiers who accessed psychiatric services between September 2007 and August 2013.
Participants: A clinical cohort of 35,509 children and young people (aged 4-17)
Design: Multiple government and ethical committees approved the link of clinical mental health service data to Department for Education (DfE) data on education and social care services. Under robust governance protocols, fuzzy and deterministic approaches were used by the DfE to match personal identifiers (names, date of birth, and postcode) from NPD and CRIS data sources.
Outcome measures: Risk factors for non-matching to NPD were identified, and the potential impact of non-match biases on ICD-10 mental disorder and persistent school absence (<80% attendance) were examined. Probability weighting and adjustment methods were explored as methods to mitigate the impact of non-match biases.
Results: Governance challenges included developing a research protocol for data linkage which met the legislative requirements for both NHS and DfE. From CRIS 29,278(82.5%) were matched to NPD school attendance records. Presenting to services in late adolescence (aO.R 0.67, 95% C.I 0.59-0.75) or outside of school census timeframes (aO.R 0.15, 0.14-0.17) reduced likelihood of matching. After adjustments for linkage error, ICD-10 mental disorder remained significantly associated with persistent school absence (aO.R 1.13, 1.07-1.22)
Conclusions: The work described sets a precedent for education data being used for medical benefit in England. Linkage between health and education records offers a powerful tool for evaluating the impact of mental health on school function, but biases due to linkage error may produce misleading results. Collaborative research with data providers is needed to develop linkage methods that minimize potential biases in analyses of linked data.
- data linkage
- health informatics
- school and education