3.1 Study design and setting

The Untire app study is an international waiting-list randomized controlled trial (RCT) aimed to examine the effectiveness of the Untire app in improving levels of fatigue and QoL in patients and survivors of cancer in four countries (ie, Australia, Canada, the United Kingdom, and the United States). Details are provided in the protocol.¹⁶ The trial has been registered in the Trial Registry on November 29, 2017 (https://www.trialregister.nl/trial/6642).

3.2 Study population

The Untire app study included cancer patients and survivors of ≥18 years, who experienced fatigue at moderate/severe levels (ie, having an average composite score of ≥3¹⁷ on items 1-3 of the FSI^12-14) and owned a smartphone, tablet, or iPad (ie, iOS/Android). Exclusion criteria were a diagnosis of and receiving treatment for a severe mental disorder since these persons may need more intensive treatment than offered by the app. Also excluded were persons having a diagnosis of myalgic encephalomyelitis/chronic fatigue syndrome or fibromyalgia, since the app focusses on improving physical activity, which could be potentially harmful to these patients.¹⁸

3.3 Procedure and recruitment

From March to October 2018, we recruited potential participants online via multiple paid social media advertising campaigns on Facebook using Facebook Ads Manager, directing persons to the study landing page of the online survey tool (ie, Questback). These campaigns match to female and male Facebook users who have shown interest in topics related to cancer, fatigue, and cancer survivorship. Based on participant characteristics of those who showed interest (ie, clicked on the advertisement), comparable participants (ie, lookalike audiences) were approached in subsequent advertisement campaigns.

On the study landing page, interested persons were screened for eligibility.¹⁶ Based on the e-mail address and/or name, we checked for multiple registration attempts at the eligibility screening and removed duplicate entries. Eligible persons received the information letter (online) and gave electronic informed consent. After completing a baseline questionnaire, participants were randomized into the intervention or control group. Intervention participants received a personal access code that could be used to activate the Untire app after downloading it from the app store, enabling free usage. All participants received e-mail invitations and reminders at 4, 8, 12, and 24-weeks to complete the questionnaires. This article focusses on the outcomes of the baseline (T0)-T12. Once control participants completed the 12-week questionnaire, they also received an access code that could be used to activate the app for free. All participants were allowed to seek or accept other treatments for their fatigue during the study.

3.4 Randomization and blinding

Participants who completed the baseline were automatically randomized 2:1 by the online survey tool into the intervention and control groups. Since intervention participants received the incentive (app access) right away, we expected more dropout in the intervention group at 12-weeks.^{19, 20} The randomization ratio was chosen to balance the number of 12-week assessments between intervention and control, and enabling a sufficient number of assessments for in-depth analyses within the intervention group. Due to the character of the intervention, it was not possible to blind participants and researchers to the intervention or control allocation.

3.5 Intervention

The Untire app intervention is based on evidence-based methods for patients with CRF in clinical practice and comprises four modules (ie, My themes, My exercises, Physical activity, and Tips), see Appendix for more information. It is hypothesized that the app addresses dysfunctional thoughts via CBT and psycho-education (My themes), reduces stress and improves sleep via MBSR (My exercises), helps to improve physical fitness through exercise instructions (Physical Activity), and empowers via positive psychology (Tips).^{3, 5, 21-27} In addition, through Quick scans, participants get weekly insight into their levels of fatigue, burden, happiness, satisfaction, and energy leaks over time (see Appendix). Participants in the intervention group were instructed to use the app at their own pace and could freely choose modules to work on. While daily app use was recommended, participants received instructions to use the app at least once a week. We provided automatic push notifications within the first 2 weeks of app use to support participants to activate the app. In case that participants did not use the app, reminders where send.

3.6 Outcome assessment

3.7 Statistical analysis

3.8 Sample size

Sample size calculations¹⁶ showed that we needed 164 participants with complete 12-week measures in the intervention and control group (total N = 328) to detect an improvement in the primary outcome with a minimal effect size of η² = 0.10 (α = .05, 1-β = 0.95), using the simplest between/within-group comparison (F tests - RM-ANOVA with within-between interaction) in G*Power 3.1.

4.1 Study sample

From the 3.060 persons who showed interest in the Untire app study, 1137 met the study criteria and gave informed consent, and 799 participants completed the primary outcome at baseline (Figure 1). As shown in Table 1, the vast majority of the sample was female (92%), middle-aged (55.5 ± 9.8 years), and moderate to severely fatigued (severity: 6.5 ± 1.4; interference: 5.8 ± 2.0). Most participants reported being a cancer survivor. The baseline characteristics of T12-completers and T12-dropouts differed concerning the percentage of participants who received help to manage their fatigue before (T12-completers: 13%, n = 44; T12-noncompleters: 8%, n = 35). The intervention participants who activated the app had a slightly higher level of education than those who did not activate the app (P < .02).

4.2 Does fatigue improve significantly after 12-weeks of app access?

After 12-weeks, participants in the intervention group showed significantly reduced levels of fatigue severity and interference as compared with participants in the control group (F(3,1912) = 4.55, P < .01, d = 0.40, and F(3,1912) = 4.10, P < .01, d = 0.35), see Table 2 and Figure 2. These results were confirmed by the T12-completers analysis (Appendix S1, Supporting Information 1). Concerning the degree of app use, “high users” used the app ≥9 days, “medium users” used the app ≥3 days, “low users” used the app ≥1 day. The reduction in fatigue severity and interference was seen, especially in participants who showed high and medium app use, in comparison to nonusers and participants in the control group (P's < .02; see Figure 2, and Supporting Information 2). The levels of fatigue in low app users, nonusers, and control participants did not statistically differ after 12-weeks (P's > .05).

The clinical relevance analysis for fatigue interference showed differences in the number of people who recovered, improved, not-changed, and deteriorated, favoring the intervention group (P = .04), whereas for fatigue severity, no statistically significant difference was found (P = .05), see Supporting Information 3. With regard to both fatigue severity and interference, significantly more people recovered in the intervention group vs the control group after 12-weeks (P's = .02).

4.3 Does QoL improve significantly after 12-weeks of app access?

QoL during the past week (EORTC-QLQ-30) did not differ significantly between conditions after 12-weeks (F(3,1897) = 2.34, P = .07, d = 0.15), whereas levels of QoL on average did improve significantly in the intervention group from T0 to T12 as compared with the control group (F(1,334) = 4.67, P < .01, d = 0.32), see Figure 2 and Supporting Information 1.

4.4 Moderation of intervention effects

The variables age, education, and cancer status were taken as moderators into account. The significant intervention effect on fatigue interference was moderated by age (F(3,1887) = 2.67, P = .046), with slightly stronger improvements in younger intervention participants (≤56 vs. >56). Education and cancer status did not moderate any of the intervention effects (all P's > .05; see Supporting Information 4 and 5).

5.1 Clinical implications

Our study demonstrated small-to-moderate treatment effects (d = 0.35-0.40) of the Untire intervention, which is larger than other self-management eHealth interventions (ie, d = 0.21¹⁰) but smaller than the effects of therapist-guided online interventions (ie, d = 0.71^{3, 36}) and face-to-face therapy (ie, d = 0.45³⁶; or d = 0.74²⁴) targeted at CRF. However, it is important to note that with self-management (ie, not therapist guided) mHealth interventions, a higher number of patients and survivors can be reached and helped simultaneously at much lower costs. Although the high attrition rates in the intervention group reflect that a self-management mHealth intervention might not be suitable for everyone, we assume that an app due to its low-threshold supply could be especially helpful for patients who would otherwise not engage in any form of treatment. In line with this, we appeared to have attracted participants who did not receive any help previously.

5.2 Study limitations

In addition to our study's international focus and large-scale set-up, we addressed the issue of widespread CRF by an innovative low-threshold mHealth intervention. A specific advantage of the Untire app itself is its flexibility, as the patients are in the lead to select the content or exercises that match their interests and needs instead of following a predetermined week-by-week program. Current in-app engagement strategies include different modes of content delivery, persuasive elements, reminders, and social support. Increasing personalized feedback may further engage participants and foster retention.

Carrying out a waiting-list RCT helped us to demonstrate causality, and to control for potential spontaneous improvements during the study period. A drawback of this design is that it might artificially inflate intervention effect estimates since participants in the control group could have been influenced by design to literally “wait-to-change” and thus do not improve.³⁷ However, this does not seem to be the case in our study, since control participants also showed improvements. Another limitation is that, in order to limit patient burden, QoL was assessed with two questions instead of using a complete, validated QoL questionnaire.

A drawback of using an online recruitment procedure is that multiple registrations by one person are possible. We removed duplicate registrations based on email address and/or name. In the study sample, the percentage of persons who completed the registration more than once did not differ between conditions, assuming that those attempts were not motivated by the idea of receiving immediate app access while being initially allocated to the waitinglist-control group (ie, no evidence for a broken randomization). Another drawback of online recruitment is the lack of access to hospital records, meaning that we cannot validate medical information (eg, cancer type and status). However, participants had nothing to gain by providing invalid medical information. Another limitation is that remotely conducted mHealth studies regularly show high attrition rates concerning study assessments and intervention participation (in our case, app use). High dropout rates and decreasing app use may be common,^{28, 30, 38} as the choice to quit is uncomplicated owing to the impersonal character of the online study and intervention.³⁹ We intended to counteract the law of attrition³⁹ by sending reminder e-mails supporting participants to complete study assessments. Due to the design and the methodology used, it was not possible to have personal contact with the participants, and increasing adherence by sending reminders was partially successful. Low retention rates prohibit any inferences about participants who dropped out and might introduce selection bias. It is important to realize that unequal dropout rates in intervention and control groups do not necessarily imply that effect estimates are biased.⁴⁰ As aforementioned, our data analysis did not find any substantial differences when comparing baseline characteristics (Table 1) of T12-completers with T12-dropouts, and the results were consistent between analytic strategies (ie, ITT and T12-completers). Furthermore, we acknowledge that blind analyses would have been the optimal way to eliminate the experimenter's bias. We aimed to reduce experimenter bias by publishing the methodology and the statistical analysis plan in the Trial Registry prior to analyzing our data.¹⁶

We should be aware that the recruitment strategy may have influenced our study sample, and that our findings might be limited to middle-aged female breast cancer patients who are using Facebook. However, our sample is likely to represent future users who might find the Untire app in a similar way, that is, through finding information on the internet or Facebook. This idea is supported by data from users who downloaded and used the Untire app after the study had been completed, which showed that most users were indeed middle-aged and female.

5.3 Future research

Future research could explore the mechanisms behind the app (ie, possible mediators of the treatment effect, such as fatigue catastrophizing, mindfulness, physical activity, or sleep) to understand the underlying processes determining the reduction of fatigue. To further validate the study results, also participants in non-English speaking countries should be targeted. Besides online recruitment, also other pathways of introducing the app to the end-user (ie, introduction to the patient by a healthcare professional) could be explored concerning differences in sample characteristics, dropout, adherence, and effectiveness.