Design

As detailed in our protocol [13], this was an efficient pragmatic two-arm (1:1 ratio intervention versus usual care) cluster randomised trial, disseminating Germ Defence to all GP practices in England to reduce respiratory tract infections (RTI). Randomisation was conducted by the independent Bristol Trials Centre (BTC). The 133 NHS Clinical Commission Groups (CCGs: NHS bodies responsible for the planning health care services for their local area) in England were divided into blocks according to region, and equal numbers in each block were randomly allocated to intervention or usual care. The randomisation schedule was generated in Stata statistical software by a BTC statistician not otherwise involved in the enrolment of general practices into the study. The principal investigators, the study statistician and research team remained blinded to the identity of randomised practices until the end of the study.

Setting and participants

All GP practices in England registered with NHS Digital (N = 6579) were included to ensure that the intervention was rolled out across demographically and geographically diverse regions.

Sample size considerations

To detect a relative risk reduction of 0.14 with 90% power (alpha 0.05), based on the previous Germ Defence implementation, PRIMIT trial [8] was calculated to require 11,124,176 participants from approximately 1484 practices (accounting for clustering). We randomised all GP practices in England, aiming for at least 25% of GP practices of those contacted successfully disseminating the intervention to their patients.

Intervention

Germ Defence content was rapidly adapted throughout the pandemic using state-of-the-art evidence, theory and the person-based approach [14], in order to ensure the advice remained up to date and appropriate. Content, design and structure were iteratively optimised via co-participatory approaches with the general public in order to ensure the intervention was as accessible, credible and motivating as possible [9]. On the first page participants reached, they could access content in 25 languages as well as infographics (which were also translated into other languages) that they could share with people who were not able to access digital content.

The original Germ Defence intervention drew on the theory of planned behaviour [15] and protection motivation theory [16] to address user motivations and intentions, employing additional theory-based behaviour change techniques such as an if–then planning and self-monitoring to help users implement their handwashing intentions. Drawing on the RE-AIM implementation framework [17], data from the original RCT of the intervention was analysed to examine intervention reach and showed that the intervention was equally effective across gender and age and was particularly effective for people with low and high levels of education [18].

The single-session intervention sought to improve users’ awareness of risks of infection and transmission, increase skills and confidence to reduce risks and use behaviour change techniques (such as making if–then plans) to support behaviours. The Germ Defence content was tailored such that a user selected one of four streams that was relevant to the user’s situation:

1.

To protect themselves generally

2.

To protect others if the user was showing symptoms

3.

To protect themselves if household member(s) showed symptoms

4.

To protect a household member who is at high risk

Clear and detailed advice was provided for self-isolating, social distancing, cleaning, wearing face coverings, ventilation and handwashing.

Intervention implementation

An initial email to practices was drafted by the research team that contained a unique weblink to the Germ Defence website and asked practices to disseminate this to all their adult patients (aged 16 +) via mobile phone text, email or social media. This email was iteratively optimised in pilot interviews with nurses, GPs and administrative staff from six practices to ensure it was acceptable and engaging. Reasons that practices might not engage were discussed (e.g. not enough time, did not perceive benefits, did not typically engage in research, concerns around privacy), and email content was refined to address these barriers.

To further support engagement, the email linked to a trial information website that addressed key concerns and frequently asked questions in more detail [19]. Practices did not need be a research active practice or to sign up to take part in the study; the practice-unique Germ Defence weblink allowed the study team to detect their involvement once patients accessed the Germ Defence intervention. The email also contained suggested text for patient mobile phone message and email. This was also made available in Bengali, French, Polish, Portuguese, Punjabi and Urdu.

On 10 November 2020, intervention arm practices were emailed (see Supplementary file 1) with a practice-specific weblink to the Germ Defence website and asked to disseminate this to all their adult patients (aged 16 +) via mobile phone text, email or social media. Two reminder emails were sent on 25 November and 10 December 2020 to intervention arm practices (see Supplementary file 2).

Data suggest that 16% of the GP practice email addresses forwarded by NHS Digital to the study team did not work, with a total of 613 ‘undelivered’ emails recorded in response to Germ Defence’s initial approach to intervention practices in England. This was usually because registered email addresses were out of date. During the intervention delivery phase, all invalid email addresses were investigated further via a series of manual Internet searches and telephone calls to practices, replacing invalid emails with new information as appropriate. This follow-up effort improved the data quality by around a third.

Patients at GP practices randomised to the usual care arm received standard management for the 4-month (17 weeks) trial period. On 10 March 2021, usual care arm practices were emailed a generic weblink to Germ Defence and asked to disseminate it to all their adult patients.

Measures and outcomes

The primary outcome was the rate of GP presentations for respiratory tract infections (RTI) per registered patient. Secondary outcomes comprised the rates of acute RTIs, COVID-19 diagnoses, COVID-19 symptoms, gastrointestinal infection diagnoses, antibiotic prescriptions and hospital admissions. COVID-19 symptoms were defined using two different code lists: one designed for high sensitivity and the other for high specificity. Each outcome was defined using SNOMED-CT codelists (see Supplementary file 3 and GitHub repository). A consultation for a specific outcome was identified if a patient had a code from the codelist recorded on a given day. If a patient had multiple codes from the same codelist on the same day, this was counted as one consultation. The number of such consultations divided by the number of patients formed the consultation rate.

All health outcomes were analysed using routinely recorded clinical and patient information in GP practice data. All data were linked, stored and analysed securely within the OpenSAFELY platform, https://opensafely.org/, a trusted research environment (TRE) enabling secure, transparent analysis of electronic health records. Data included pseudonymised fields such as coded diagnoses, medications, physiological parameters, patient age, patient ethnicity and deprivation score of the practice area. No free text data were included. All code used in this study is shared openly for review and re-use under MIT open licence: https://github.com/opensafely/GermDefence. Detailed pseudonymised patient data is potentially re-identifiable and therefore not shared. Primary care records managed by the GP software provider, TPP, were linked to admitted patient care (APC) data through OpenSAFELY. Practice allocations were ingested into the OpenSAFELY platform and linked to pseudonymised practice IDs by TPP and made accessible to the study team by OpenSAFELY.

A further secondary outcome, uptake of the intervention by GP practices, was monitored using embedded code in a unique Germ Defence website link given to each practice. When practices communicated the unique weblink to their patients, the study team were able to record usage of the weblink. Uptake was measured using website analytics such as number of users per practice, average time spent on the Germ Defence website and pages visited, monitored using Matomo to ensure privacy [13, 20]. In line with MRC (Medical Research Council) guidelines for evaluating complex interventions [21], we also sought to understand mechanisms of action by aggregating individual self-report measures of infection control behaviours (social distancing, self-isolation, wearing masks, handwashing, cleaning/disinfecting, ventilation) collected by the Germ Defence website and combined this with metrics of engagement with key intervention behavioural components (e.g. pages viewed, amount of time spent on intervention).

Patient and public involvement

Patient and public involvement (PPI) feedback was a key part of the co-participatory approach of the development of Germ Defence, in which members of the public were invited to feed back about the website and study in order to optimise and update it. A public contributor (C. R.) was a coinvestigator on the study team and contributed to writing the research proposal, updating and optimising the content of the intervention (including optimising intervention communications sent to patients by practices) and co-authoring the papers. Study materials were also reviewed by PPI representatives from the NIHR (National Institute of Health Research) Clinical Research Network (CRN).

Data analysisSummary of baseline data

This cluster randomised controlled trial was analysed at the practice level. Randomisation was carried out at practice level, and we did not have direct feedback on whether practices distributed the Germ Defence information to all, some or potentially no patients, nor whether individual patients were offered the information and made use of it. We, therefore, conducted all analyses using aggregated data at the practice level and considered each practice as a unit for the purpose of analysis. Outcome (consultations) and covariate data (median age, proportion of females, proportion from an ethnic minority, deprivation of practice area) from patient-level records were aggregated into weekly practice-level time-series data prior to analysis, covering the period from 17 weeks prior to randomisation until 17 weeks after randomisation (14th July 2020 to 15th March 2021) to achieve a target minimum of 15% infection rate.

Primary care data in the OpenSAFELY system at the time of analysis represented approximately 40% of practices in England [22]. Analyses of health outcomes were applied to 2498 practices.

Intention-to-treat analyses

The primary analysis used a standard intention-to-treat approach. For each of the eight health outcomes, rates of consultations per registered patient were compared at practice level between intervention and control groups for the 17-week post-intervention period. This was done using negative binomial regression with the consultation count as the outcome, the number of registered patients as the offset and the binary indicator of intervention/control group as the only independent variable.

Controlled interrupted time-series analyses

An additional analysis was performed for the same eight health outcomes using a controlled interrupted time-series (CITS) approach to understand temporal changes related to the intervention as distinct from the time-agnostic intention-to-treat approach. This was implemented within a generalised linear-mixed modelling framework by applying negative binomial regression to weekly level data spanning pre- and post-intervention periods for both the intervention and control groups. Data was also disaggregated by practice, allowing random intercepts at practice level. Variables included in the model were as follows: consecutively numbered weeks to capture a log-linear trend, intervention-control indicator, pre-post-intervention indicator and all two- and three-way interactions between these. Additional covariates included calendar month to capture seasonal effects and practice-level indicators such as area-level deprivation, median patient age and sex distribution represented as the proportion of females.

Process analysisImplementation process

Germ Defence website usage recorded from the unique identifying website links sent by each practice was used to examine whether intervention engagement (i.e. a practice effectively communicating link to patients) was predicted by practice characteristics (such as indices of deprivation, NHS Quality and Outcomes Frameworks).

Individual intervention usage

A range of additional behavioural mechanisms, overall patterns of practice and user engagement were described using website analytics. Analytics included number of users per practice, average time spent on the Germ Defence website and pages visited.

Association with health outcomes

To understand the mechanisms of action in the intervention, we examined the association between the rate of website usage within a practice (number of users divided by number of registered patients) and the rate of each health outcome (consultations per registered patient). A negative binomial model was applied to practice level data, and the association of interest was adjusted for decile of deprivation, proportion of patients from an ethnic minority and median age. This was done for all practices and then separately for a subset of practices that had greater than 1% uptake.

Information governance and ethical approval

NHS England is the data controller for OpenSAFELY-TPP, TPP is the data processor and all study authors using OpenSAFELY have the approval of NHS England. This implementation of OpenSAFELY is hosted within the TPP environment which is accredited to the ISO 27001 information security standard and is NHS IG (information governance) Toolkit compliant [23, 24].

Patient data has been pseudonymised for analysis and linkage using industry standard cryptographic hashing techniques; all pseudonymised datasets transmitted for linkage onto OpenSAFELY are encrypted; access to the platform is via a virtual private network (VPN) connection, restricted to a small group of researchers; the researchers hold contracts with NHS England and only access the platform to initiate database queries and statistical models; all database activity is logged; and only aggregate statistical outputs leave the platform environment following best practice for anonymisation of results such as statistical disclosure control for low cell counts [25].

The OpenSAFELY research platform adheres to the obligations of the UK General Data Protection Regulation (GDPR) and the Data Protection Act 2018. In March 2020, the Secretary of State for Health and Social Care used powers under the UK Health Service (Control of Patient Information) Regulations 2002 (COPI) to require organisations to process confidential patient information for the purposes of protecting public health, providing healthcare services to the public and monitoring and managing the COVID-19 outbreak and incidents of exposure; this sets aside the requirement for patient consent [26]. This was extended in November 2022 for the NHS England OpenSAFELY COVID-19 research platform. In some cases of data sharing, the common law duty of confidence is met using, for example, patient consent or support from the Health Research Authority Confidentiality Advisory Group [27].

Taken together, these provide the legal bases to link patient datasets on the OpenSAFELY platform. GP practices, from which the primary care data are obtained, are required to share relevant health information to support the public health response to the pandemic and have been informed of the OpenSAFELY analytics platform.

This study was approved by the Health Research Authority, Yorkshire & The Humber—Leeds West Research Ethics Committee (Ref.: 20/YH/0261).