Study design

The trial was conducted in the area of the Lille-Douai Health Insurance district (Northern France) during the institutional seasonal influenza vaccination campaign of 2014–2015.

It was a single blinded 2/1 registry-based cluster randomized controlled trial design. A computerized random draw was used to allocate GPs in each group. In the intervention group, 25 GPs received and were supposed to expose in their waiting rooms 135 pamphlets and one poster (added to the usual mandatory information) withdrawing all the other posters. In the control group, waiting rooms were kept in their usual state. Fifty GPs in the control group were not aware of the intervention but knew that the seasonal influenza vaccination uptake of their patients would be measured. The outcome was the number of seasonal influenza vaccination units released in community pharmacies. Baseline was defined as the 2013–2014 seasonal influenza vaccination campaign. Data were extracted between October 15, 2014 and February 28, 2015 from the SIAM-ERASME claim database of the Lille-Douai district Health Insurance Fund on patient level. The study was lengthened by one month (February 2015) following the extension of the vaccination campaign duration by health authorities. The protocol had therefore been amended.

The prescription of the seasonal influenza vaccine units was delivered by an individual GP depending on the skills, training and motivation of the latter. This may lead to a potential clustering of the outcome for patients treated by the same GP. To provide valid results, considering the cluster allocation by GP of patients, a correction has to be introduced in computing the number of GPs needed for the trial, called intra-cluster correlation coefficient (ICC). According to Austin PC, the largest mean difference in power for the analysis of a cluster randomized trial with binary outcomes was used, with an ICC of 0.02, for α = 5% and β = 20%.[13] With a predicted rate of influenza vaccination delivery of 0.65 in the intervention group and 0.60 in the control group, and a target size by GP of 400 patients, 75 GPs had to be enrolled (50 in the control group and 25 in the intervention group).[1]

Target population

The GPs participating in the study had previously been contacted by telephone and had given their written consent. GPs without waiting room, or sharing one single waiting room with several other GPs, or participating in another ongoing study, were not eligible to participate in this study. Language barriers in patients were not considered.

The study population existed of patients over the age of 16, who were registered by the Health Insurance on the participating GPs’ patients lists. The target population were patients over the age of 65 or having a chronic disease requiring seasonal influenza immunization (like COPD or diabetes). Patients were informed about the anonymous use of their data and could refuse to participate at any time.

Statistical analysis

A univariate analysis was performed first, to present the baseline characteristics of patients after randomization, respectively in the intervention and control groups. Quantitative variables were expressed as mean with a 95% Confidence Interval (95% CI) of the mean. Categorical variables were expressed as percentages and 95% CI of the percentage. The clustering variable GP was taken into account through the svydesign function of the package survey [14].

To assess the association between the vaccination status (dependent variable) and the group intervention/control, because of the hierarchical structure of the data and the high incidence rate of the main outcome, we used a generalized estimating equation (GEE) Poisson regression clustering by GP [15]; an exchangeable working correlation matrix was used. Although the trial is randomized and therefore the confounding factors are already balanced between the yes/no intervention groups, in order to ensure that the "previous influenza vaccination" factor was fully controlled, the variable “Previous influenza vaccination” was used as an adjustment variable in the model.

The analysis was carried out using R software (version 3.3.1) and the package Geepack [16].

Ethics, regulation and redaction

The study protocol passed the Ethics Committee of the Lille University Hospital (CPP Nord Ouest IV, advice #: HP 14/51) and the National Electronic Data and Liberty Commission (CNIL, advice 1783641 v 0). It is registered on ClinicalTrials.gov (registration #: NCT03239795).

The CONSORT statement has been used to complete this article.

Limits

The perspective of the influenza vaccination campaign was considered from a primary care point of view as opposed to the public health perspective previously accomplished by the French Health Insurance. The robust design of this study included a sufficient number of patients therefore the lack of demonstrated effect is unlikely correlated to an insufficient power of the trial: if an effect exists, it is insufficient to be considered as clinically relevant, i.e. able to enhance substantially the seasonal influenza vaccination uptake. Through randomization groups were comparable from baseline to the end of the study. The generalized mixed linear statistical model with maximized likelihood adjustment gave a precise evaluation of the effect on the main outcomes.

A possible bias could involve the share of patients vaccinated directly by a nurse without being exposed to the GP intervention as some patients may have received the invitation to be vaccinated directly from the Health Insurance. Nonetheless, the present study has been randomized and the groups are being comparable leading to conclude that this lack of exposure to the intervention was comparable in both groups, without other influence on the result than limiting the impact of the intervention. Some GPs of the intervention group refused partially or totally to redesign their waiting rooms when the intervention was positioned. This barrier signs that the trial was performed in real live conditions. The intention to treat analysis of the outcomes considers this limit and no per-protocol analysis has been conducted. Another limitation is that vaccines provided by pharmacies have been considered as full vaccinations. The research group stated in the study protocol that the number of vaccines provided by pharmacies that were not injected to patients was negligible: this hypothesis is the usual one used by health authorities to investigate the vaccination coverage rate but has not been investigated.[1] However, the number of delivered and not injected vaccines is probably comparable in both arms of the trial. As patients under the age of 16 were not enlisted with a GP by the health insurance before 2017, no data were collected in patients under the age of 16 with a chronic disease (like asthma). Encounters with a different GP than the one by whom patients were enlisted was not taken into account.

The influenza immunization rate increased in both arms of the study compared to the prior year. Contemporarily, at regional and national level, the number of vaccinated people decreased. The Hawthorne effect, i.e. the trend of people to change their behaviour when they know that they are under observation, might have masked the effect of posters and pamphlets. [29] The study group carries out a new study to compare the control group in this study with a new control group enrolled a posteriori and measure the Hawthorne effect (Zelen Design [30]).