Keywords
Randomised Controlled Trial, Embedded Trial, SWAT, Retention, text, notification, personalisation, SMS
Randomised Controlled Trial, Embedded Trial, SWAT, Retention, text, notification, personalisation, SMS
- Throughout the manuscript and the title any reference to 'retention rate' or 'response rate' has been replaced with 'questionnaire completion', for consistency.
- Figure 1 has been updated to ensure consistent terminology.
- Additional text has been included around the various data collection methods, and why telephone was the main interest
- Additional details were included around the money incentives.
- Sentence structure was altered to allow for the manuscript to be grammatically correct.
- The term 'text' was replaced with 'text message' throughout the manuscript.
- The abstract was altered to be clearer that the significant results were only related to telephone collection.
- Wording was altered for clarification around the sample size, and eligibility for the SWAT.
See the authors' detailed response to the review by Frances Shiely
See the authors' detailed response to the review by Phil J. Edwards
Randomised controlled trials (RCTs) are the ‘gold standard’ for evaluating healthcare treatments. However, it is well documented that retaining participants can be difficult and low response rates to questionnaires can compromise the reliability and generalisability of the results1,2. A study within a trial (SWAT) can be used to test interventions to improve retention of participants, via increasing questionnaire completion3.
There is research to support the concept that text messages are effective at improving questionnaire completion rates in trials4–7. There is insufficient evidence to determine if the timing of text messages improves questionnaire completion rates, and limited papers exploring if personalisation (inclusion of the participants name) impacts questionnaire completion rates8–11. This factorial SWAT aims to evaluate the effectiveness of the timing and personalisation of text messages within an RCT to add to the evidence base for both of these interventions.
This two-by-two factorial study was embedded within the MiQuit-3 RCT. MiQuit-3 (ClinicalTrials.gov NCT03231553) is an RCT evaluating the effectiveness of a text-message, smoking cessation self-help support programme for pregnant smokers (MiQuit), and the protocol has been published previously12. This factorial SWAT was embedded at the 36-week gestational time point. The approval for this factorial SWAT and the MiQuit-3 trial was granted by East Midlands–Nottingham 1 Research Ethics Committee (NRES reference 13/EM/0427 and 17/EM/0327). As the SWAT was considered low risk, informed consent was not obtained from participants, and they were unaware of the SWAT. However, as part of the MiQuit-3 trial all participants consented to their anonymised data being used for further research and being published. The SWATs that form the factorial design are also registered with the Northern Ireland Hub for Trial Methodology Research SWAT Repository (SWATs 35 and 44; both registered December 2015).
As with all SWATs, the sample size is limited by that of the host trial, and a formal power calculation has not been conducted. In total 1002 participants were randomised to the MiQuit-3 trial. As this SWAT was implemented mid-way through follow up for the host trial, all participants that had not yet had their 36-week gestational follow-up, approximately 200, were eligible to participate in the SWAT, and any that had already passed this follow-up time point were unable to be included in this SWAT.
Participants in MiQuit-3 were unaware of their participation in this SWAT, however, they could not be blinded to the contents or timing of the text message. Participants were randomised 1:1:1:1 to each of the four groups (see Table 1). The randomisation was undertaken by a statistician independent of the host trial, and of the staff involved in sending the text messages. Block randomisation was used with varying block sizes of 4, 8, 12 and 16, which was stratified by host trial allocation, and whether they had completed the previous follow-up or not. The randomisation sequence was generated in Stata v.15 (RRID:SCR_012763) and implemented using a remote computer system, independent of the researchers.
This SWAT explored two different interventions; personalisation and timing of text messages (early; one week before follow-up, or late; one day before follow-up). Details of the text message sent to participants can be found in Table one. As detailed in the MiQuit-3 protocol12, all participants were given a £5 voucher if they completed the 36-week follow-up. Those who provided a saliva sample to validate their smoking status were given an additional £30 voucher. This amount is more than is stated in the published protocol, but this change was made prior to the implementation of this SWAT, and as such all participants involved in this SWAT would have been eligible for this amount. These monetary incentives formed part of the host trial, and where not related to the factorial SWAT being undertaken.
The primary outcome was completion rate; defined as the proportion of the questionnaires completed over the telephone within the follow-up window (14 days).
The secondary outcome measures included:
Completion rate where the questionnaire was completed by any method (postal, telephone, email/web, or text message) within the follow-up window (14 days)
Time to completion, defined as the number of days between the due date of the 36-week gestation follow-up and the date the questionnaire was recorded as complete
Number of attempts to contact required before the questionnaire was complete, or the maximum number of attempts, six, is reached.
Both time to completion and number of attempts to contact were not restricted by method of data collection, and thus included participants who completed (or were being contacted) via any method.
The data were analysed in Stata v.15 (RRID:SCR_012763) on an intention-to-treat (ITT) basis, using two-sided tests at the 2.5% level. As this is a factorial design the Bonferroni correction was applied to allow for multiple testing13,14. Participants were excluded from the analysis if they had withdrawn prior to the time point.
The primary outcome and completion for all methods were compared using a logistic regression model. Time to completion (days between questionnaire due and complete) was analysed using a Cox Proportional Hazards regression. Participants who completed the questionnaire early had their time set to 0.1, those who did not complete it were censored at either last contact date or 120 days if not contacted, and those who withdrew in the course of the SWAT were set to their withdrawal date. The assumptions for this model were assessed using Schoenfeld residuals15. The number of attempts to contact was analysed using a negative binomial regression model, due to evidence of overdispersion. All models were adjusted for host trial allocation, whether the participant had completed the previous follow-up, age, and both SWAT intervention allocations separately. All models were repeated with the inclusion of an interaction term to explore any possible interactions between the two SWAT interventions. This was done using two-sided tests at a significance level of 5%.
Stata is proprietary software: a freely available alternative software that could be used to undertake this analysis is RStudio (RRID:SCR_000432)16.
In total, 194 participants were randomised into the SWAT; 50 received the personalised text message and early notification, 47 received the personalised text message and late notification, 50 received the non-personalised text message and early notification, and 47 received the non-personalised text message and late notification17. Five participants withdrew prior to the implementation of the SWAT and are not included in the analysis. Participants were only included in a model if all relevant covariates for that model were present. The number included in each of the analysis, by arm, is shown in the flow diagram – Figure 1. Three participants were not contacted due to difficulties/adverse events associated with their pregnancy, but are still included in the analysis under ITT principles. The flow of participants can be seen in Figure 1. Baseline characteristics by SWAT arm and overall, can be found in Table 2.
The main method of data collection for the MiQuit-3 trial was telephone collection. As such the primary outcome explores the completion rates where the data was collected via telephone calls only. The overall completion rate by telephone was 66.1% (125/189) within the follow-up window (14 days). There were similar completion rates of the questionnaire via telephone within three groups; 50.0% for personalised early (24/48), 52.3% (23/44) for personalised late, and 58.0% (29/50) of non-personalised early, and was slightly higher in the non-personalised late group, 66.0% (31/47).
There was no evidence for a difference in completion rate for the timing of the text message where data was collected via telephone calls; adjusted odds ratio (OR) 0.86 (95% CI 0.44–1.67, p=0.65). There was evidence to suggest a difference in completion rate (adjusted OR 0.44, 95%CI 0.22–0.87, p=0.02) which implies those who received the non-personalised text message were more likely to complete the questionnaire than those who received a personalised text message, when data was collected via the telephone. Full details can be found in Table 3.
Completion rates for all methods. Additional methods of data collection were used alongside telephone calls. For completion via any method the data could have been collected via post, telephone, email/web or text message. When looking at any method, there were similar completion rates of the questionnaire within each of the four groups; 64.6% for personalised early (31/48), 63.6% (28/44) for personalised late, 66.0% for early (33/50) and 70.2% (33/47) of non-personalised.
Primary Outcome | Group | Statistic* | 95% Confidence Interval | p-value |
---|---|---|---|---|
Completion rate for telephone only | Personalised versus non-personalised | OR = 0.44 | 0.22 to 0.87 | 0.02 |
Early versus Late | OR = 0.86 | 0.44 to 1.67 | 0.65 | |
Host trial allocation (Intervention versus Control) | OR = 0.63 | 0.32 to 1.22 | 0.17 | |
Completed previous follow-up (Yes versus No) | OR = 9.90 | 3.87 to 25.35 | >0.001 | |
Age (years) | OR = 1.02 | 0.96 to 1.07 | 0.60 |
Secondary outcomes:
Full details for all secondary outcomes can be found in Table 4.
Secondary Outcome | Group | Statistic* | 95% Confidence Interval | p-value |
---|---|---|---|---|
Completion rate for all methods | Personalised versus non-personalised | OR = 0.61 | 0.30 to 1.24 | 0.17 |
Early versus Late | OR = 1.06 | 0.52 to 2.15 | 0.87 | |
Host trial allocation (Intervention versus Control) | OR = 0.79 | 0.39 to 1.60 | 0.51 | |
Completed previous follow-up (Yes versus No) | OR = 8.45 | 3.60 to 19.86 | >0.001 | |
Age (years) | OR = 1.05 | 0.99 to 1.11 | 0.12 | |
Number of attempts to contact required | Personalised versus non-personalised | IRR = 1.14 | 0.92 to 1.41 | 0.23 |
Early versus Late | IRR = 1.08 | 0.88 to 1.33 | 0.45 | |
Host trial allocation (Intervention versus Control) | IRR = 1.11 | 0.90 to 1.37 | 0.33 | |
Completed previous follow-up (Yes versus No) | IRR = 0.64 | 0.50 to 0.82 | >0.001 | |
Age (years) | IRR = 1.00 | 0.98 to 1.02 | 0.79 | |
Time to completion | Personalised versus non-personalised | HR = 0.76 | 0.54 to 1.07 | 0.12 |
Early versus Late | HR = 1.00 | 0.71 to 1.40 | 0.99 | |
Host trial allocation (Intervention versus Control) | HR = 0.87 | 0.62 to 1.21 | 0.40 | |
Completed previous follow-up (Yes versus No) | HR = 3.42 | 1.95 to 5.99 | >0.001 | |
Age (years) | HR = 1.01 | 0.98 to 1.04 | 0.51 |
There is no evidence to suggest that there is a difference in completion rate for personalised versus non-personalised text messages; adjusted OR 0.61 (95% CI 0.30–1.24, p=0.17). Additionally, there was no evidence to suggest there was a difference in completion rates in participants who received an early or late text message; adjusted OR 1.06 (95% CI 0.52–2.15, p=0.87).
Number of attempts to contact required. The average number of attempts to contact required was 3.0 for all participants, with the average similar for each group (3.3 for both personalised early, 3.2 for personalised late, 3.1 for non-personalised early and 2.7 for non-personalised late). Researchers attempted to contact a participant a maximum of six times. The maximum number of attempts to contact was reached for 55 of the 174 participants (31.3%) and was similar across three groups (38.6% for personalised and early, 31.7% for personalised and late, 31.1% for non-personalised early) and slightly lower in the non-personalised late group, 25%.
There was no evidence of a difference in number attempts to contacts required between those who received an early text message or a late text message (p=0.45). There is also no evidence to suggest a difference between those who received a personalised or non-personalised text message (p=0.23); adjusted incidence rate ratio (IRR) 1.14.
Time to completion. The average time to completion of the questionnaire was 6.2 days (ranging from 5 days early to 103 days late). The time to completion was similar between those who received an early or late personalised text message text message (8.2 days for early versus 7.1 days for late) and was similar for those who received an early or late non-personalised text message (4.9 days for early versus 4.7 days for late). However, there was a slight difference in time to completion between those who received personalised or non-personalised text message.
There was no evidence of a difference in time to completion between those who received the text message early or late (p=0.99) or those who received a personalised or non-personalised text message (p=0.12). This suggest that neither timing nor personalisation of the text message reminder affect the time taken to complete the questionnaire. The assumptions for the model held when examined using Schoenfeld residuals (p=0.66).
Interaction terms. All of the models were re-run with the inclusion of any interaction term between the two SWAT allocations. There was no evidence of an interaction for the completion rate, both by phone only (p=0.57) and all methods (p=0.54). There was also no evidence of an interaction for the number of contacts required (p=0.69), or the time to completion (p=0.88).
Comparison with the whole RCT. There were 1002 participants who were randomised into the MiQuit-3 trial. Of the 777 who were not included in the SWAT, and were due a 36-week follow-up, 499 completed the questionnaire (64.2%). This is similar to the completion rate for the participants in the SWAT (overall 66.1%).
This factorial SWAT showed that the timing of the text message reminder had no effect on the questionnaire completion rate, the time to complete, or the number of attempts to contact required; these results mirror what Partha et al. reported in their work8. It also showed that personalised text messages have no effect on completion time, or number of attempts to contact required. However, it did show that there was some evidence that sending a non-personalised text message reminder would have a larger increase in response than sending personalised text messages did, but these finding were only significant when exploring telephone data collection. Cochrane et al. found no statistically significant difference in their study, but results favoured the non-personalised text messages11. As our work was conducted in a female-only population, who were between 17 and 41 years of age, the results here are only directly related to this population. Equally, as the SWAT was not powered to detect a difference, more SWATs should be undertaken in this area to allow the results to be combined in a pooled analysis to determine the true effect of the interventions and consider the effects on a wider population.
Figshare: Underlying data for ‘Pre-notification and personalisation of text-messages to retain participants in a smoking cessation pregnancy RCT: an embedded randomised factorial trial’. https://doi.org/10.6084/m9.figshare.14224319.v117
Data are available under the terms of the Creative Commons Attribution 4.0 International license (CC-BY 4.0).
Figshare: CONSORT checklist for ‘Pre-notification and personalisation of text-messages to retain participants in a smoking cessation pregnancy RCT: an embedded randomised factorial trial’. https://doi.org/10.6084/m9.figshare.14229647.v118
The authors would like to acknowledge Professor Tim Coleman, the Chief Investigator for the MiQuit-3 trial, and members of the MiQuit-3 trial involved in implementing the SWAT into the trial.
Views | Downloads | |
---|---|---|
F1000Research | - | - |
PubMed Central
Data from PMC are received and updated monthly.
|
- | - |
Competing Interests: No competing interests were disclosed.
Reviewer Expertise: Epidemiology; Trial Methodology; SWATs
Is the work clearly and accurately presented and does it cite the current literature?
Partly
Is the study design appropriate and is the work technically sound?
Yes
Are sufficient details of methods and analysis provided to allow replication by others?
Partly
If applicable, is the statistical analysis and its interpretation appropriate?
Partly
Are all the source data underlying the results available to ensure full reproducibility?
Yes
Are the conclusions drawn adequately supported by the results?
No
Competing Interests: No competing interests were disclosed.
Reviewer Expertise: Epidemiology; Trial Methodology; SWATs
Is the work clearly and accurately presented and does it cite the current literature?
Yes
Is the study design appropriate and is the work technically sound?
Yes
Are sufficient details of methods and analysis provided to allow replication by others?
Yes
If applicable, is the statistical analysis and its interpretation appropriate?
Yes
Are all the source data underlying the results available to ensure full reproducibility?
Yes
Are the conclusions drawn adequately supported by the results?
Yes
Competing Interests: No competing interests were disclosed.
Reviewer Expertise: Evidence-based Data Collection
Alongside their report, reviewers assign a status to the article:
Invited Reviewers | ||
---|---|---|
1 | 2 | |
Version 2 (revision) 30 Sep 21 |
read | |
Version 1 22 Jul 21 |
read | read |
Provide sufficient details of any financial or non-financial competing interests to enable users to assess whether your comments might lead a reasonable person to question your impartiality. Consider the following examples, but note that this is not an exhaustive list:
Sign up for content alerts and receive a weekly or monthly email with all newly published articles
Already registered? Sign in
The email address should be the one you originally registered with F1000.
You registered with F1000 via Google, so we cannot reset your password.
To sign in, please click here.
If you still need help with your Google account password, please click here.
You registered with F1000 via Facebook, so we cannot reset your password.
To sign in, please click here.
If you still need help with your Facebook account password, please click here.
If your email address is registered with us, we will email you instructions to reset your password.
If you think you should have received this email but it has not arrived, please check your spam filters and/or contact for further assistance.
Comments on this article Comments (0)