Development and use of a privacy-protecting total population record linkage system to support observational, interventional, and policy relevant research

Background The 2006 Cooksey review of UK health research identified the NHS cradle-to-grave records of the 60 million UK residents as having the greatest potential to reinvigorate the UK research environment. Subsequently, national and regional e-health research units were set up across the UK, including in Wales, the Secure Anonymised Information Linkage (SAIL) system. We aim to draw attention to the uses, findings, and research potential of the SAIL system. Methods The SAIL system is a suite of privacy-protecting technologies, including multistage encryption in which unique anonymised numbers replace individual, household, and organisational identities (appendix). There is strong input from patient and public representatives and an independent panel that scrutinised proposals. Many different health, educational attainment, and housing datasets are included in SAIL, which now supports various research designs including pure population-based electronic cohorts, traditional cohorts, embedded individual and cluster randomised trials, and assessments of natural experiments. Examples from more than 50 studies supported so far include the Wales Electronic Cohort for Children (WECC), NIHR-funded evaluations of natural experiments (housing improvement and changes in alcohol outlet density), cluster randomised trials of prehospital care, and an assessment of the Welsh Government Flying Start childhood initiative. We will focus on some of the results from WECC. Findings WECC contains anonymised records of the 804 290 children living in Wales between 1990 and 2008 compiled from nine datasets (linkage rate >97% for three population registers: births, NHS registrations, and community child dataset). 590 042 children were born in Wales, and electronic education records were available for 122 817 born after 1994. This analysis will focus on the effect of pregnancy, birth, and childhood factors on health status and educational attainment to key stage 1 (age 7 years). For example, very low birthweight (hazard ratio [HR] 2·4, 95% CI 1·9–2·9), prematurity (3·07, 2·76–3·41, for <28 weeks; 1·36, 1·32–1·41, for 33–36 weeks), small for gestational age at term (1·09, 1·05–1·12), and birth by caesarean section (1·14, 1·09–1·20) all increased the likelihood of respiratory admissions. Prematurity was also associated with educational underachievement (odds ratio 1·5, 95% CI 1·0–2·5 for <28 weeks' gestation; 1·3, 1·2–1·4, for 33–36 weeks), as was social deprivation (2·6, 2·2–3·0). Children who moved house frequently or at particular periods were also at increased risk of poor health and educational underachievement. 10% of children moved within the first year and a third by age 5 years. For most outcomes (with the exception of vaccination) there were clear dose-response relations in risk-adjusted negative binomial, logistic regression, and survival analyses—eg, more than two or three house moves versus none (incidence rate ratio 1·40, 95% CI 1·08–1·80, for injury admissions and 1·18, 1·09–1·29, for respiratory admissions). For more than three moves at age 4–6 years, risk of educational underachievement increased to 1·79 (95% CI 1·48–2·16). Head injuries were also associated with significant underachievement (HR 1·46, 1·11–1·93). Interpretation The SAIL system provides a research efficient e-cohort platform on which it is possible to embed intervention trials and assess the effect of natural experiments and multisectoral policy initiatives on health, educational, and other population outcomes.