Study design

A cluster-randomised controlled parallel-group trial design was employed for the TRAINS study using all GP practices in the CPRD Aurum database in England, as recorded in June 2021. The study included a total of 1389 GP practices. The practices were randomised into two groups: 693 practices in the intervention group and 695 practices in the control group, which continued with their usual care. This design was particularly suitable as it allowed for the use of routinely collected data, thereby ensuring that the study was grounded in real-world, practical conditions15.

This trial received ethical approval from the University of Sheffield Research Ethics Committee (Reference Number: 037412). Additionally, an Independent Scientific Advisory Committee (ISAC) approval was granted for Clinical Practice Research Datalink (CPRD) database research (Protocol Reference ID: 21_000436).

Participants

The target population for the intervention consisted of GP practices in England that were actively contributing to the CPRD Aurum database at that time of the study in June 2021. Data were extracted for school-age children who had been diagnosed with asthma, were registered with an eligible practice and had received preventive prescriptions in the year leading up to the study.

Inclusion criteria for practicesExclusion criteria for practices

General practices located outside of England.

Practices that left CPRD post-intervention (June 2021) and before the completion of the follow-up o the primary outcome (September 2021).

Practices that underwent a merger following the intervention and before the end of follow-up period, especially if the merged practices were initially part of different study arms.

Inclusion criteria for the data extraction from CPRD

School-age children, aged between 4 and 16 years old as of 1 September 2021, with a coded diagnosis of asthma, who had been prescribed asthma preventers medication during the past 12 months.

Randomisation and allocation concealment

GP practices were randomised 1:1 to either intervention or control group. The randomisation process was stratified based on practice size (decile) within the CPRD. The included practices were identified through the CPRD. The process of randomisation was performed by [SAJ] using a bespoke Excel spreadsheet and then reviewed by [RS]. A file with anonymous practice identification codes and the stratification number were provided to the CPRD.

Procedure

In this study, GP practices enrolled in the CPRD Aurum database and assigned to the intervention group received the intervention package twice—via both email and mail. The intervention package was specifically designed to guide and support GPs, comprising a letter to GPs that highlighted the decline in asthma preventer medication collection during summer holidays, emphasising its significance for children with asthma and presenting findings from the PLEASANT study along with key recommendations and available resources. It also included a leaflet detailing the PLEASANT study, and reminder templates for a SMS text message and GP letter to facilitate adaptation. Further details about the design of the intervention can be found in the protocol15. Please see Appendix A for the full the intervention. It is important to note that the letters sent to GP practices were advisory in nature, allowing each practice the flexibility to decide how to practically adopt the intervention as the wished.

The intervention was distributed in June 2021, with emails sent on 23 June and postal packages dispatched on 25 June. This process was coordinated by CPRD. In total, 693 postal packages and emails were sent to GP practices. In some cases, multiple contacts within a practice, such as the practice manager and lead GP, were included, resulting in a total of 1403 email contacts. To monitor engagement, we used CPRD’s email read receipt feature, which enabled us to track whether the CPRD contact lead at each practice opened our email.

Data were extracted from the CPRD Aurum database build in May 2022, focusing on patients aged 4–16 years as of 1 September 2021, who were registered with GP practices in England participating in the TRAINS study. To be eligible, patients needed to have an asthma diagnosis and a prescription for asthma medication within the past year (from 1 June 2020–31 May 2021), indicating active asthma, and must have been alive at the end of the primary analysis period on 30 September 2021.

The extracted records included comprehensive details on all medical contacts, including prescription requests and out-of-hours visits. CPRD provided anonymised data for each patient, which encompassed the General Practice identifier, year and month of birth (for patients under 16), sex, ethnic group, details of prescribed asthma medication, and medical contact data for each appointment, including dates. The research team accessed only fully anonymised data, ensuring that no patient-identifiable information was available.

Following the intervention distribution, data collection continued over a six-month period, ending in December 2021. Two baseline periods, 2019 and 2020, were used for comparison. As 2020 was not considered a standard year due to the impacts of the COVID-19 pandemic we used two years. The primary goal was to evaluate the real-world impact and effectiveness of the intervention during this phase.

OutcomesPrimary outcome

The primary outcome was the proportion of children diagnosed with asthma who collected a prescription for an asthma preventer medication within the time frame of 1st August 2021–30th September 2021.

Secondary Outcomes

The following is a list of the secondary outcomes:

1.

The total number of prescription collection of asthma preventer medication per patient from 1 August 2021–30 September 2021.

2.

The total number of prescription collection of asthma preventer medication per patient in August 2021 and September 2021.

3.

The proportion of patients who have collected an asthma preventer medication in August 2021 and September 2021.

4.

The total number of prescription collection for asthma preventer medication in the 6 months post-intervention starting 1 July 2021.

5.

The proportion of patients with unscheduled medical contacts from September 2021 through December 2021, also segmented by individual months.

6.

The total unscheduled medical contacts per patient from September 2021 to December 2021, and the individual months from 1 September 2021–31 December 2021.

7.

The proportion of patients with medical contacts (both scheduled and unscheduled) from September 2021 to December 2021, and the individual months from 1 September 2021–31 December 2021

8.

The total medical contacts (both unscheduled and scheduled) per patient from September 2021 to December 2021, and the individual months from 1 September 2021–31 December 2021.

Sample size

The target sample size for this study was determined based on feasibility and the expected number of GP practices providing data to the CPRD. We anticipated including approximately 1389 GP practices (693 in the intervention group and 695 in the control group). Drawing on data from the previous PLEASANT study, we expected around 85 school-age children with asthma per practice13. Assuming a 30% prescription collection rate and an intraclass correlation coefficient of 0.03, the study aimed to achieve a precision level estimated as a half-width of 1% for the 95% confidence interval.

This precision-based approach was selected due to feasibility constraints and the primary objective of the TRAINS trial: to accurately estimate the proportion of school-aged children with asthma who collect their prescriptions in each group. Precision-based calculations are particularly suited for trials focused on estimating rates or proportions with high accuracy. Although the study does not have a pre-specified effect size for power calculations, it is designed to provide an accurate estimate of any observed effect.

With an anticipated precision of approximately 1%, the study aimed to deliver a reliable estimate of prescription collection rates in the context of asthma management for school-aged children.

Statistical analysis

Baseline characteristics of both practices and patients were reported and compared between the treatment groups. At the patient level, descriptive statistics were provided, including gender, age group, and ethnicity (presented as frequencies and percentages), along with age (reported as mean, standard deviation, median, interquartile range, minimum, and maximum). At the practice level, we presented the practice size (deciles), Index of Multiple Deprivation (IMD) (as frequencies and percentages), and the number of children per practice (reported as mean, median, interquartile range, range, standard deviation, minimum, and maximum).

The primary analysis of effectiveness was conducted in the intention-to-treat (ITT) population, which included all practices with data collected during the study period. Sensitivity analyses were performed on a subset of ITT practices identified through email read receipt (RR) confirmations for the intervention practice only.

For the primary outcome, a mixed-effects logistic regression model was used to analyse the proportion of children collecting prescriptions in the primary study period, adjusting for gender, age group, ethnicity, and prescription collection in August–September 2019 and 2020 (included as binary covariates). We initially planned to use 2020 alone as the baseline year; however, because prescribing and consultation patterns in 2020 were disrupted by the COVID-19 pandemic, 2019 was also included as an additional baseline comparator. Binary outcomes were analysed using mixed-effects logistic regression, and count outcomes using mixed-effects negative binomial regression. To account for clustering within practices, random effects were applied based on GP practice ID (decile) and IMD levels. Results from logistic regression are reported as odds ratios (ORs) with 95% confidence intervals (CIs), and results from negative binomial models as incidence rate ratios (IRRs) with 95% CIs. Analyses of unscheduled and total medical contacts across different time periods were performed using the same modelling approach.

All analyses were performed in Stata (version 17), and R (version 4.2.3) was used for generating time-series visualisations. All analyses were conducted using a two-sided significance level of 5%.