Assessing the impact of specialist home visiting upon maltreatment in England: a feasibility study of data linkage from a public health trial to routine health and social care data

Background: Follow-up for public health trials may benefit from greater use of routine data. Our trial of a home-visiting intervention for first-time teenage mothers assessed outcomes to the child's second birthday. To examine its medium-term impact, particularly upon maltreatment outcomes, we designed a study using routine records.

Methods: We aimed to establish the feasibility of our study design, which combines trial data with routine health, social care and education data using a dissent-based linkage model. Trial participant identifiers were linked to routine health, social care and education data if women did not dissent. Data were forwarded to a safe haven and further linked to de-identified trial outcome data. The feasibility study aimed first to establish the acceptability of data linkage through a discussion group of young mothers and by levels of dissent received by the research team. Second, we assessed levels of accurate linkage to both health (via NHS Digital) and education and social care (both via National Pupil Database, NPD). Third, we assessed the availability of data and levels of missingness for key outcomes received for a sample of target study years.

Results: Of 1545 mother-child dyads contacted, eight women opted out. The engagement exercise with stakeholders found support for the principle of data linkage, including in the context of maltreatment. Some contributors preferred opt-in consent. Most (99.9%) health records were matched on either three or all four identifiers. Fifty participants were not matched to any health data. Primary outcome data from NPD are derived from any one of three fields, all of which were satisfactorily returned and provided an indication of cases for analysis. Missing data for secondary outcomes varied from 0% (Child looked after status) to 70% (Anatomical Area A&E diagnosis) however when combined with other variables the levels of missingness for outcome decrease.

Conclusions: Through study set-up and in this pilot, we provide evidence that the main study is feasible, satisfies governance requirements and is likely to generate data of sufficient quality to address our main research questions. Observed levels of missingness or low event rates are likely to affect some secondary analysis (e.g. state transition modelling) although overall were satisfactory.