The proportion of patients whose mobile telephone numbers are recorded by their general practice is low, limiting the power of our study to detect an effect.

Text messaging is already commonly used in general practice to remind patients about receiving their influenza vaccine, but this study is the first to estimate the effectiveness of this intervention.

An emerging trial methodology utilises routinely collected electronic health record sources to ascertain outcome data, at far lower cost than traditional methods of data collection. 10 The feasibility of this approach in primary care practices requires further examination. Therefore we performed a cluster randomised trial in English primary care, embedded within routine electronic health records, to determine the effectiveness of text message influenza vaccination reminders in increasing influenza vaccination uptake among the at-risk under 65s. Randomising individuals within practices would have been logistically complex and so a cluster design was chosen to reduce the burden to practices of participating in our trial. The aims of the study were (1) to develop the methods for conducting cluster randomised trials of text messaging interventions, and to evaluate feasibility of recruitment, identification of eligible practices and patients, and to follow-up through routine electronic health records at very low cost; and (2) to implement a cluster randomised trial to test the effectiveness of using a text messaging influenza vaccine reminder in achieving an increase in practice-level influenza vaccine uptake in patients aged 18–64 years with chronic conditions, compared to standard care.

Vaccination reminders by letter, postcard and telephone have shown to be effective, 6 , 7 but these interventions become costly when there are large numbers of at-risk patients. Text messaging is a useful tool because it is cheap, fast and personal, and because mobile phone ownership in the UK is widespread (94% of adults own a mobile phone). 8 Consequently, text messaging has been increasingly used in the National Health Service to contact patients for appointment reminders and other health-related issues. In the 2010–2011 influenza season, text messages were used in roughly a third of practices for influenza vaccination reminders 9 but there is no evidence that this strategy increases uptake in primary care.

Annual seasonal influenza vaccination is offered to certain patient groups in the UK, including patients aged 65 years and older, and those who are aged under 65 years with one or more chronic conditions (the ‘at-risk’ under 65s). The UK government sets targets for vaccine uptake in these groups; in the 2013/2014 influenza season the target was 75%. 1 While uptake in the over 65s reaches this target in many practices, barriers to vaccination result in unmet targets for the at-risk under 65s (uptake in 2012/2013 was 49.6%). 2 These barriers include failure of healthcare workers to recommend vaccination, low awareness of eligibility for vaccination, perceptions about severity of influenza, and personal beliefs about vaccine effectiveness, pain and side effects. 3–5 In the face of these barriers, strategies to improve vaccine uptake are required.

Methods

Further details of our methods have been described previously in our published protocol,11 and the formal study protocol is available in online supplementary appendix 1.

Trial design This was a two arm, pragmatic, cluster randomised trial, with general practices as clusters allocated in a 1:1 ratio to each arm.

General practice and patient eligibility The study took place in the 2013/2014 influenza season (vaccinations September–December) among English primary care practices in three settings: (1) the Clinical Practice Research Datalink (CPRD),12 a primary care database based on Vision software and covering 8% of the UK population; (2) TPP SystmOne software13 users; (3) iPLATO text messaging software users in London14 (initially only practices in Islington and Barnet were targeted, and this was later extended to all London boroughs to achieve the required sample size). Practices were recruited between July 2013 and October 2013. Practices were eligible if they already used text messaging software and did not use text messaging to contact patients about influenza vaccine in the 2012/2013 season. Eligibility of CPRD practices was determined based on the routinely recorded data and only practices meeting the criteria were invited. All users of iPLATO across London boroughs, and all practices using TPP SystmOne software were invited to the trial and eligibility was determined at the next stage. Within participating practices, eligible patients were those aged 18–64 years in at-risk groups due to chronic conditions at the start of the influenza vaccination season. This comprises patients who are under 65 with chronic heart disease, chronic neurological disease, diabetes, chronic kidney disease, chronic liver disease, chronic respiratory disease and immunosuppression, as set out by the Chief Medical Officer.15 Pregnant women and carers, who are also eligible to get the vaccine, were not targeted in our study to avoid ethics concerns (to avoid contacting patients by text message whose circumstances are time dependent and may have changed without being updated in the medical record). Practices are able to identify these individuals as part of routine care, as these patients are targeted for seasonal influenza vaccination. Patients were excluded from the study if they transferred out of the practice or died before the end of data collection (31 December 2013). Consent was gained from the general practitioner at each general practice, before randomisation had taken place. Patients did not provide consent to be part of this trial. The trial co-ordinator enrolled general practices and informed the practices of their allocation.

Interventions Practices were allocated to either a standard care arm or an intervention arm. In the standard care arm, practices were asked to continue with their seasonal influenza vaccination campaign as planned, typically using measures such as posters in the practice and letters to patients. In addition to their planned campaign, practices in the intervention arm were asked to send a text message influenza vaccination reminder using their in-practice text messaging software, to patients aged 18–64 years in at-risk groups. This required practices to perform two steps: (1) identify eligible patients: most clinical software systems have algorithms to identify eligible patients and practices were encouraged to use these; (2) send a tailored text message to these patients using software embedded in the electronic health record. The recommended text message content was “Hello PATIENT NAME, to reduce your risk of serious health problems from flu, we recommend vaccination. Call PRACTICE PHONE NUMBER to book. PRACTICE NAME.” The message content was designed by behavioural specialists and detail is given in the study protocol.11 Step by step instructions were provided (see online supplementary appendix 2) but no formal training was given to practices as they were included in the trial based on their experience in using text messaging for other purposes (eg, appointment reminders). Practices were instructed to send one text message to their patients at the start of their seasonal influenza campaign, but were free to send additional messages later in the season if they chose to. Details of the intervention delivery, including which member of staff sent the text message, the time of day the messages were sent, and whether messages were sent in bulk or individually were left to the discretion of the practice (though guidance was provided in the step by step instructions).

Randomisation, allocation concealment and blinding After enrolment to the trial, practices were randomly allocated (1:1) to the standard care arm or the intervention (influenza text messaging reminder) arm. Randomisation was performed within strata defined by the clinical software system or text messaging software used by each general practice (TPP SystmOne, CPRD, iPLATO), and was additionally stratified by region (for CPRD practices) and borough (for London-based practices). Block randomisation, using block sizes of 2, 4 and 6, was performed. The allocation sequence was generated by an independent statistician who was blind to practice name. The trial co-ordinator, who enrolled the practices and informed them of their allocation, was unaware of the allocation until a complete block had been randomised. It was not possible to blind practices to their allocation but data management and analysis were performed blind to allocation.

Outcomes The primary outcome, measured at the cluster level (GP practice), was influenza vaccination uptake among patients aged 18–64 years in the seven prespecified risk groups during the period between 1 September and 31 December 2013. This time period would have captured the majority of influenza vaccinations. Secondary outcomes were (1) the proportion of eligible patients sent a text message, and (2) outcomes used to establish the feasibility of our methods, including the rate of recruitment, adherence to the text messaging protocol, adverse events and problems with message delivery, availability of electronic health record data, and the time and cost required to gather data.

Data collection Data on vaccination uptake, overall and by clinical risk group, were obtained through extraction of routinely collected data. Additional consent from GPs was required for further demographic patient-level data (age and sex) and so were only available for a subset. In practices that contribute data to the CPRD or use TPP SystmOne software, relevant patient-level data were extracted from the database using prespecified Read codes (the standard clinical terminology system used in general practice in the UK). Within London practices using iPLATO software, data were extracted by the clinical software supplier of the practice or the practices themselves. Data extracted included patient registration details, age, sex, death date, clinical risk group, vaccination status and influenza vaccine invites. Codes indicating risk groups and vaccinations were identified using PRIMIS Plus Read code lists, available online.16 In one of the three settings, where text messaging was systematically recorded in the practice electronic health record, adherence to the intervention was assessed by study investigators. It was not possible to measure adherence to the exact wording of the text message. All data were stripped of personal identifiers before being supplied to the study team. Where patient-level data were not available, practice-level data were supplied by Immform, the UK's surveillance system for influenza vaccine uptake.17

Sample size To achieve 90% power with a 5% significance level, assuming an intracluster correlation of 0.024, 100 practices were required to identify a 7.5% relative increase in vaccine uptake from 54% to 58%.18 To account for differences in the number of eligible patients per practice and dilution of the intervention effect through potential contamination between arms, we chose to recruit and randomise 150 practices to the study.11

Statistical methods The primary analysis was by intent to treat, comparing the cluster-specific proportions of vaccine uptake between randomised arms using a two-sided t test, with cluster proportions weighted to account for differing numbers of at-risk patients across practices. Minimum-variance weights were used,19 which weight cluster i by where ICC is the intracluster correlation coefficient and n i is the number of at-risk patients in that cluster. Sensitivity and subgroup analyses We conducted a number of secondary analyses to account for contamination between the randomised arms. First, we used an instrumental variables approach to fit a structural mean model, using the minimum-variance weights from the primary analysis, to estimate the intervention effect under hypothetical full adherence,20 that is, assuming all eligible patients in the active arm received an influenza text message, and no eligible patients in the other arm received a text message. We additionally performed a per-protocol analysis which restricted analysis to the patients in each practice who actually received their allocated intervention. These analyses were restricted to practices for which data were available regarding text messaging. In a non-randomised preplanned comparison, we explored whether the intervention effect differed by time of day the text message was sent by using a multilevel logistic regression model including a random main effect for practice fitted to patient-level data. These analyses were also restricted to practices for which data were available regarding text messaging. A post hoc exploratory analysis was conducted using similar multilevel logistic regression models to assess the intervention effect within seven predefined at-risk patient subgroups. To assess whether participation in the trial changed vaccine uptake (Hawthorne effect), we compared practices taking part in the trial to those outside the trial using the ResearchOne primary care database (a research database holding over 400 TPP SystmOne practices). T tests were used to compare the at-risk population size and vaccine uptake of standard care practices taking part in the trial to practices outside the trial. Data were analysed using Stata V.14.1. Missing data Receipt of in-practice influenza vaccine was automatically recorded in the clinical system from which we extracted the data, so there was no missing outcome data. However, if patients received their influenza vaccine outside the practice, this would not have been recorded unless the patient informed their GP. Patients transferring out of the practice or dying during the study period were not included in the denominator. Adherence to the intervention could not be measured for all practices because specific vaccine reminder text messages were systematically recorded for practices using only TPP SystmOne clinical software.

Substudy A substudy was conducted among two intervention practices, whose administrative staff sent a short questionnaire (four questions) to patients who had received the text message. Questionnaires were sent after the data collection phase for the trial was complete (December 2013). Patients were asked to return anonymised responses directly to the study team by mail (postage pre-paid) via a self-completion questionnaire (free of any patient or practice identifiers) about their memory of having received the text message, any objections to the message, and whether it encouraged them to receive the vaccine (Questionnaire in online supplementary appendix 3).