The study was based on data from the PERSUADE study [16]. This was a randomized double-blind placebo-controlled trial (clinicaltrials.gov; NCT02716285) conducted in 4 hospitals located throughout the Netherlands (Multimedia Appendix 1; Figure S1). The study protocol was approved by the Maastricht University Medical Center+ Ethics Committee. All study procedures were performed in compliance with Good Clinical Practice Guidelines and according to the revised Declaration of Helsinki [19]. All participants gave written informed consent prior to participation.

The study design of the PERSUADE has been described in detail elsewhere [16]. In brief, the primary aim was to investigate the efficacy of peppermint oil—a conventional small-intestinal release formulation and a novel ileocolonic release formulation—in patients with IBS. To this end, patients between 18-75 years of age, who fulfilled the Rome IV criteria for IBS and had a mean worst abdominal pain score of at least 3 on an 11-point rating scale (0, no pain; 10, worst possible pain) during a 14-day pretreatment period were included. Participants were randomized to placebo, small-intestinal release peppermint oil, or ileocolonic release peppermint oil for an 8-week treatment period.

Data were collected using a customized framework for digital data collection, specifically designed and developed for the trial, consisting of (1) a digital symptom diary (smartphone app); (2) an electronic case report file (eCRF, Castor EDC); (3) web-based patient questionnaires (Castor EDC); and (4) a planning tool (Ldot). During the 14-day pretreatment and the 8-week treatment period, patients were instructed to register symptoms daily in the digital symptom diary. Study visits and telephone follow-up telephone interviews were documented in the eCRF. Patients were asked to complete several web-based questionnaires at different time-points within the study duration. The complete list of inclusion criteria and study overview with timing of the questionnaires is given in the Multimedia Appendix 1. Primary efficacy results of the PERSUADE study have been described elsewhere [16].

For the digital symptom diary, an electronic smartphone app was developed by the Center for Data and Information Management at Maastricht University (MEMIC), in close collaboration with the investigators. The app was programmed using Xamarin, a framework to develop cross-platform apps using C sharp programming. The PERSUADE app supports Android and iOS devices. A Maastricht University industrial designer designed the visual content. A MEMIC team of data managers and researchers of the Maastricht University Medical Center+ Neurogastroenterology group tested the app and provided feedback throughout several phases of development. Additionally, a patient was asked to use the diary and provide feedback regarding its user friendliness. Patient inclusion commenced once a version was reached that all agreed on.

The app’s home screen consisted of 3 main elements: the daily end-of-day symptom questionnaire, a medication list, and the Bristol stool chart questionnaire (Figure 1). The end-of-day symptom questionnaire included one main question to assess the primary outcome (in accordance with FDA guidelines): “How would you rate your abdominal pain today? Think about the worst abdominal pain today” (11-point numerical rating scale) (Figure 2). The daily symptom questionnaire was accessible between 6 PM and 12 PM and was unavailable outside this time window, to avoid premature completion. Other daily questions were related to “need of rescue medication” and “adverse events experienced.” If a patient had not completed the daily entry before 10 PM of that particular day, one push notification was sent. At the end of each week, the end-of-day questionnaire consisted of additional questions regarding abdominal discomfort, abdominal bloating, abdominal cramping, belching, nausea, and urgency during the last week (11-point scale numerical rating scale). It was not possible to enter data for previous days, and participants could not review prior entries. Automated routing, response requirements, and real-time data verification were built in to increase data quality and completeness.

The medication list was used once to register all regular medications. Patients were asked to keep their concomitant medication use as stable as possible. However, if alterations were needed, they were able to delete, add, or change the dosage of nongastrointestinal drugs.

The Bristol stool form scale was used to register all bowel movements (Figure 3). There was no minimum or maximum number of registrations per day.

All patients received extensive verbal and written instructions during the screening visits on how to use the app and were encouraged to contact the researchers if the app crashed or otherwise did not function properly. A personalized username and password were provided for access to the app. Patients were instructed to enable automatic updates of the app to ensure the most recent version was used. If a patient did not own a smartphone or tablet, a device was provided. During the complete pretreatment and treatment periods, an alert system in the planning tool notified the investigators when a patient had failed to submit 3 or more daily entries. In addition, the development team received automated notifications of app crashes.

At randomization, at 2, 4, 6, and 8 weeks of treatment, and at 3 and 6 months of follow-up after treatment ended, patients were asked to fill out web-based questionnaires. We chose not to implement these into the digital diary because of the large number of questions. Included were a questionnaire regarding demographics and lifestyle and validated questionnaires regarding symptom severity (IBS Symptom Severity System), quality of life (IBS Quality of Life, EuroQoL EQ-5D-5L), comorbid symptoms of anxiety and depression (General Anxiety Disorder 7, Patient Health Questionnaire 9), and health care utilization and productivity loss (Medical Consumption Questionnaire, Productivity Cost Questionnaire). Patients received invitations via email containing an HTML-link to the electronic environment. If a patient had not completed the questionnaire within 2 days, 2 automatic reminders were sent via email. Automated routing, response requirements, and real-time data verification were built in to increase data-quality and completeness.

During the study visits and telephone follow-up calls, investigators documented all findings in a cloud-based eCRF. The eCRF forms were built by the first author with input from the other authors and contained items regarding demographics, Rome IV diagnostic criteria for IBS, history and physical examination, adverse events, and general wellbeing. Investigators were given unique usernames and passwords to view and add data for their respective inclusion centers. To achieve registration uniformity, the investigators were trained on how to enter data, and additional step-by-step instructions were given in a standard operating procedure document. Real-time automated data verification and corresponding pop-up notifications were built in to prevent typing errors or other erroneous entries. Automated routing of questions and response requirements ensured that correct items were displayed and filled in. An audit trail enabled tracking of all data changes.

Ldot is a web-based tool developed by the Center for Data and Information Management at Maastricht University and was used to monitor study logistics. All personal patient data were entered into Ldot, and the app supported the study workflow by indicating when each study event (eg, randomization, follow-up call, etc) needed to take place for each patient. Ldot was able to communicate with the digital diary and the web-based questionnaires. For example, all email invitations for the questionnaires were sent automatically via Ldot. Patients’ adherence to the diary and web-based questionnaires could be monitored within Ldot and investigators were notified if patients failed to complete 3 consecutive days in the diary. Investigators were also notified if patients failed to complete a web-based questionnaire after a reminder was given. To guarantee the anonymity and quality of research data, no research data could be entered into Ldot. Investigators could view and add personal data for their respective inclusion centers. There was no possibility of viewing data from other inclusion centers, except for the coordinating investigator (first author) who had access to all data. An audit layer of the app tracked and stored information of all changes.

All software and data storage complied with international ISO27001, ISO9001, good clinical practice guidelines, and Dutch NEN7510 guidelines. Electronic diary data, web-based-questionnaire data, eCRF data, and sensitive personal data (Ldot) were all stored on different (nonconnected) servers. Several back-ups were made per day. Access to the servers was and will be restricted, with 24-hour on-site surveillance. Data will be stored for 15 years after study completion.

The primary outcome of the current study was patients’ adherence to the digital symptom diary, defined as the mean percentage of entries and calculated by dividing the number of completed entries by the number of minimal requested entries (total number of days in study). Patients were instructed to complete a diary entry on all consecutive days during the 14-day pretreatment and 56-day treatment period, or all days until discontinuation with the study.

Secondary outcomes were change in mean adherence per week over time, sociodemographic characteristics, clinical patient characteristics associated with adherence, time of diary completion, and difference in adherence between patients who were defined as responders to treatment versus nonresponders. Potential data loss and critical evaluation points were considered to explore the overall feasibility of a smartphone app as a primary data-collection tool in a randomized controlled trial. Other secondary outcomes were patients’ adherence to and completeness of the additional web-based questionnaires and investigators’ adherence to and completeness of the eCRF.

Statistical analyses were carried out using SPSS statistical software (version 25.0 for Macintosh; IBM Corp). Data are expressed as mean and standard deviation or as number and percentage. Multivariable linear regression analysis was used to investigate the association between baseline patient characteristics and adherence to the digital diary, adjusting for minimization variables (age, gender, IBS subtype, inclusion center, and treatment group). A repeated measures analysis of variance was performed to assess the influence of time (weeks) on adherence. If the Mauchly test indicated that the sphericity assumption was not met, Greenhouse-Geisser corrected results were reported. A P<.05 (2-sided) was considered statistically significant.

Overall, 190 patients were randomized. One patient was randomized erroneously (ie, without fulfilling all inclusion criteria). Therefore, 189 patients (age: mean 34.0, SD 13.3 years; female: 147/189, 77.8%; male: 42/189, 22.2%) were analyzed (n=64 in the placebo group, n=62 in the small intestinal release peppermint group, n=63 in the ileocolonic release peppermint oil group). Of the 189 patients, 95.8% (181) were Caucasian and 4.2% (8) were of mixed descent. Most patients (109/189, 57.7%) were recruited from a primary care setting. Eleven patients withdrew from the study during the treatment period (data until discontinuation were included in the analyses). Baseline characteristics are presented in Table 1. During recruitment, only a single patient stated the digital data collection as a reason not to participate.

Most patients used their own smartphones, but 4 out of 189 patients needed a device provided by the investigators. Patient adherence to the daily digital symptom diary was excellent during the entire study period, reflected by a mean completion rate of 87.9% (SD 9.4%), 91.5% (SD 9.2%), and 86.9% (SD 10.8%), during all 70 days of study duration, the 14-day pretreatment period, and the 8-week treatment period, respectively. Adherence during the treatment period did not differ significantly for treatment groups compared to that of the placebo (placebo: mean 87.2%; small-intestinal release: 88.3%; P=.67), and for the ileocolonic release peppermint oil (mean 87.2%; P=.33). Adherence did not differ between patients that were clinical responders to treatment (mean 88.0%) versus patients who were nonresponders (mean 86.2%). Over the complete study period of 70 days, a significant decrease in mean weekly patient adherence to the end-of-day questionnaire was found (F_{5.9, 1114.9}=15.5, P<.001) (Figure 4). Nevertheless, adherence was still good at the end of the study (mean 79.6%, SD 26.6%) (Figure 4).

When exploring independent baseline predictors for adherence, the combined regression model that included all minimization variables (age, gender, IBS subtype, inclusion center), treatment group, baseline IBS symptom severity, anxiety and depression scores, and education level showed that 1 of the 4 inclusion centers (center C; Multimedia Appendix 1, Figure S2), (regression coefficient B –10.04, 95% CI –19.51 to –0.56, P=.04) and anxiety scores at baseline (B –0.59, 95% CI –1.12 to –0.06, P=.03) were negatively associated with adherence throughout the study. Indeed, when comparing adherence between the different inclusion centers, adherence in center C was lowest (mean 82.3%, SD 12.5%), compared with adherence in centers A (mean 88.3%, SD 9.2%), B (mean 84.7, SD 14.0%), and D (mean 91.4%, SD 7.7%). There was no statistically significant effect of gender (P=.84), age (P=.22), or education level (lower education level: P=.58, middle education level: P=.46, versus high education level) on adherence. Mean time of completing the end-of-day symptom diary was 9:46 PM (ie, 14 minutes before receiving the push notification).

Several technical issues were noted by the investigators or reported by patients. In most cases, the cause was found, and the issue was resolved by the app development team without data loss. Encountered hurdles included difficulties installing the app during the screening visit due to connectivity failure, not receiving reminder notifications, inaccurate visual scaling of questionnaires on smaller smartphone screens, updates of Android or iOS operating systems that interfered with prior processes, and connectivity failure due to server maintenance. All documented technical issues and their short-term consequences are presented in Table 2.

Adherence to the web-based questionnaires was also excellent. One patient did not complete the questionnaires at the end of the treatment period; all others completed all questionnaires until the end of the study or until discontinuation (n=11 discontinued the study). Halfway through the study duration, however, a routing error in one questionnaire became apparent. Although this mistake was corrected immediately, the error had already led to missing data for that particular question in 23.3%-54.0% of all patients, depending on measurement moment (Table S1 in Multimedia Appendix 1). No missing items were found in other questionnaire items.

Adherence of the investigators to the eCRF was excellent with a completion rate of more than 99%. In total, there were 27 patients with at least 1 missing variable in the case report file, 11 of whom discontinued the study during the treatment period (the missing values comprehend follow-up calls that were not conducted). The remaining 17 cases with missing data were because of missed follow-up calls (in 11 cases, 1 follow-up (out of 3) was missed), not registering if additional information about the 6-month follow-up period was given, not registering the date of the last menstruation, not registering if the general practitioner was informed about participation in the study, or not registering the number of capsules that were reported not to be taken during one of the follow-up calls.