Code for Max Verfuerden's PhD linking dormant trials to the National Pupil Database
Background Due to high attrition in randomised controlled trials (RCTs), cognitive effects of infant formula modifications remain uncertain. The aim of this thesis was to test a new method to minimise attrition and, through doing so, to compare differences in academic performance between children previously randomised to either nutritionally modified or standard infant formula.
Methods Nine dormant infant formula RCTs conducted in England (1982-2001) were available for linkage to the National Pupil Database. Linkage was based on legal exemption from the need for participant consent. A trusted third party provided de-identified data for up to four candidate pupil matches per participant and agreement-metrics for all shared linkage variables. I completed the linkage of de-identified data using auxiliary RCT variables and probabilistic methods. Six RCTs (n=1,563) were eligible for analysis, and a further three RCTs were used to assess linkage success and improve multiple imputation. Participant academic performance was measured using exam grades, with the primary outcome being General Certificate of Secondary Education (GCSE) Maths grades at age 16. Modified formula and standard formula groups were compared on an intention-to-treat basis, stratified by trial.
Results Within the six trials eligible for analysis, primary outcome data was available for 86% of all participants. This is substantially higher than the average of 22% above age 2 years in previous consent-based cognitive follow-ups of the trials. I found no benefit for GCSE Maths performance for any type of modified formula. Secondary academic outcomes provided weak evidence of harm for one of the formula modifications.
Conclusions Unconsented linkage of dormant trials to administrative education data is feasible and leads to higher follow-up rates compared to traditional consented follow-up methods. None of the investigated nutritionally modified formula interventions improved academic performance.
This code was developed using Stata.
Verfürden M, Harron K, Jerrim J, et al Infant formula composition and educational performance: a protocol to extend follow-up for a set of randomised controlled trials using linked administrative education records BMJ Open 2020;10:e035968. doi: 10.1136/bmjopen-2019-035968 https://bmjopen.bmj.com/content/10/7/e035968
The purpose of this repository is to make my PhD data analysis transparent and reproducible. Unfortunately, the data cannot be shared because it is subject to a data sharing agreement which allows only identified and information-governance trained users to process it in a data safe haven which needs to have passed the latest IG toolkit standards. The code was subject to output control and was therefore exported and uploaded manually.