Dissemination of Findings Fast Using Online-videos Trial

[Statistical analysis plan for the DIFFUSION trial]

SETTING

Two videos will be uploaded on YouTube through a new account created for this study. The uploaded videos will be embedded in two independent websites with white background. A computer programme has been created to collect the data of access to the videos.

International journals in obstetrics and gynaecology published in 2013 and 2014 will be sought for participant recruitment. The authors of the articles published in the journals who met the inclusion criteria will be assigned ID numbers and randomised to receive either the intervention or control video. Randomised participants will be sent a unique web address (URL) of the video that they are allocated to by e-mail via a mass e-mailing service, Campaign Monitor. Each URL will be linked to the participant's ID number in the database. They will be asked to watch the video and forward it to their colleagues if they find it helpful. If a randomised participant forwards their URL to another person, and that person clicks on it, the click will be recorded against the participant's ID number in the computer programme. Therefore, if there are multiple accesses from the same ID number, it is either because the participant watched the video more than once or because the participant forwarded the URL and other people watched the video.

This trial will be coordinated from the London School of Hygiene \& Tropical Medicine (University of London) and conducted worldwide online.

SAMPLE SIZE AND STATISTICAL POWER

Two main factors determine the number of patients needed in a trial. These are the estimated event rate and size of the treatment effect. The investigators will power the trial for our main outcome: percentage of participants that forwarded the video at least once in each arm.

ESTIMATED EVENT RATE: Based on other studies, baseline e-mail forward rate was assumed to be 10%.

SAMPLE SIZE AND SIZE OF INTERVENTION EFFECT THAT SHOULD BE DETECTABLE: Assuming the change from baseline is 7.5% (75% increase from 10% to 17.5%), the sample size required for this study was calculated. The calculation suggested 1000 participants, 500 per group, are required to test the null hypothesis at the 5% significance level with 90% power. We learnt from the pilot trial that around 14% of e-mail recipients click on the link and watch the video. In order to get 1000 participants to watch the videos, 7000 participants will be required given the video viewing rate of 14%.

CRITERIA FOR EVALUATION

All participants randomly assigned to one of the videos will be analysed together, regardless of whether or not they watched the video, on an intention to treat basis. This analysis is to examine the effect of the intervention in a practical scenario. In addition, only those who watched the videos will be analysed on a per-protocol basis to estimate the efficacy of the intervention.

RANDOMISATION

An independent statistician will randomly allocate the ID numbers of the participants to either the intervention or the control groups using a computer programme. Allocation will be concealed to a person who will recruit participants. Randomisations were balanced 1:1.

BLINDING AND UNBLINDING

As two videos are apparently different, participants cannot be blinded in this trial. However, each participant is not sent the other video so in principle they cannot tell if they received a video with more or less emotional content. The study coordinator who will conduct data analyses is masked to the allocation. The data will be unblinded after this SAP is submitted to the ClinicalTrial.gov and all data analyses have been completed. The data are kept in a password-secured online archive (Google drive). Access to the data is limited to the trial coordinator and technical assistant who developed the computer programme for data collection.

ANALYSIS PRINCIPLES FOR FINAL ANALYSIS

• Two different analyses will be conducted for the primary outcome on an intention-to-treat basis and a per-protocol basis.
• All tests will be two-sided, and the nominal level of α will be 5% for the main analysis.
• All statistical analyses will be unadjusted except where indicated.
• Missing values will not be imputed unless specified otherwise. Where the number of missing observations is substantial, the number of observations used in the analysis will be reported.
• P values will not be adjusted for multiplicity. However, the outcomes are clearly categorised by degree of importance (primary and secondary).

DATA COLLECTION AND OUTCOME ASSESSMENT

Every time someone clicks on a URL to access a video our computer program collects four variables: (a) ID number (b) IP address from which the URL is clicked, (c) date and time of access (recorded in Great British Time regardless of the country of the IP address) and (d) type of device (web browser, phone or computer device)

Definition of access by the same person to a video:

As there is no way to detect forwarding, accesses to the video by unique individuals will be counted as the closest indicator of forwarding. It is also impossible to precisely distinguish an access by a unique individual, in other words, to tell if the video was accessed more than once by the same person or by different people. Therefore, the investigators need to make assumptions based on the data collected by our programme. In the main analysis, the investigators will consider that accesses with the same ID number from the same IP address mean that a person watched the video more than once. Accesses with the same ID number from different IP addresses will be considered as accesses by unique individuals. The data of the other variables (time and type of device) will be considered in the sensitivity analysis. Multiple accesses with the same ID from the same IP address within five minutes will be counted as one access, allowing for the possibility that a person watched the video more than once because the video did not play the first time or the parson wanted to confirm the information in the video.

CHARACTERISTICS OF PARTICIPANTS AND BASELINE COMPARISONS

Description of the baseline characteristics listed below will be presented by intervention group. The countries which the participants are based in were categorised into low, middle and high income countries based on the classification by the World Bank. Discrete variables will be summarised by frequencies and percentages. Percentages will be calculated according to the number of participants for whom data are available. Where values are missing, the denominator (which is less than the number of patients assigned to the treatment group) will be stated in either the body or footnote of the corresponding summary table.

Continuous variables will be summarised using standard measures of central tendency and dispersion, i.e. either a mean or a median with SD.

DESCRIPTION OF ANALYSES

The outcome assessor will be masked to the identity of each participant by using ID numbers instead of individual names or e-mail addresses.

Accesses to the videos will be counted as a substitute for the measure of forwarding. Therefore, even if a participant forwarded the URL to another person, if the person does not click on the link, that forwarding will not be counted. This leads to the underestimate of the effect of the intervention.

PRIMARY OUTCOME: A standard chi-square test will be used as the primary test of statistical significance of the effect of intervention allocation on the proportion of video forwarding. Frequencies and percentages per arm, and a risk ratio measuring the intervention effect and its 95% confidence interval (CI) will also be reported.

SECONDARY OUTCOMES: Distribution of number of accesses generated by each participant will be compared. The difference in the means (SD) in each arm will be estimated in order to assess the effect of the intervention along with the P value. To test the statistical difference in distribution, Wilcoxon signed-rank test will be conducted. In case the distributions are complex, the data will be stratified into a few groups and chi-square test will be conducted to test the statistical difference in distribution.

After assessing the distributions, a t-test and Mood's median test will be conducted to test the statistical difference in the mean and median of the number of access to the videos that each participant generated respectively.

SENSITIVITY ANALYSES: The effect of the intervention may vary according to the definition of an access by a unique person. Therefore, the investigators will conduct sensitivity analyses for all outcomes using two different definitions of an access by a unique person: most conservative definition and most liberal definition.

The possible largest effect of the intervention and the difference in all the outcomes between the two groups will be estimated using the most liberal definition of forwarding. Likewise, the possible smallest effect of the intervention and the difference in all the outcomes between the two groups will be estimated using the most conservative definition. With the most liberal definition, the investigators will count all accesses as those by different persons. With the most conservative definition, the investigators will count accesses as those by different persons only if the data from all the categories are different.