Mobile.Net is a nationwide multicenter randomized controlled two-armed trial. The Mobile.Net trial evaluated the effects of tailored short message service (SMS) text messages constructed to encourage patient medication adherence and outpatient care for adult patients with psychosis [21]. Participants in the intervention group received semiautomatic text messages for 12 months (approximately 10 per month; range 2-25 text messages) based on their preferences [22]. They were able to decide the amount, timing, and frequency of the SMS text messages delivered. They were also able to change the content or timing of the messages during the trial. Treatment as usual was offered to all participants. The study was carried out according to the Declaration of Helsinki and approved by the Ethics Committee of the Hospital District of Southwest Finland. Written informed consent was obtained from the study participants after they were given a complete description of the study.

There were a total of 1139 participants, men and women, ranging in age from 18 to 65 years. Each participant had a continuing prescription for antipsychotic medication, access to a mobile phone, and the ability to use the Finnish language. After participants were recruited, they were then randomized. Forensic patients and those having a planned nonacute treatment period were excluded from the study [21].

Recruitment, including activities conducted before and during participant enrollment [23], occurred face-to-face in 45 psychiatric hospital wards in Finland (between September 5, 2011 and November 30, 2012). Research nurses in each study ward performed chart reviews to check the eligibility of each patient admitted to the study ward [24]. After completion of the baseline data, patients were allocated randomly (computer-based randomization with four block randomization) into two groups (SMS text message intervention group or control group) [21].

Attrition, including actions after enrollment in the study [23], was assessed at the 12-month follow-up period (between September 5, 2012 and December 31, 2013). Attrition was defined as the loss of eligible participants from the study groups [25]. To gather follow-up data, a postal survey, including the Quality of Life Enjoyment and Satisfaction Questionnaire (Q-LES-Q) by Endicott et al [26] and the Client Satisfaction Questionnaire (CSQ-8) by Atkisson and Greenfield [27], was carried out. Follow-up data were collected by members of the research group. During the study period and follow-up, each point of contact with participants was completely tracked and recorded. Attrition was assessed during the intervention and during the follow-up.

Data for this paper were divided into two categories: data concerning patient recruitment and data relating to attrition (Textbox 1).

Recruitment data were categorized into five groups: (1) patients screened for eligibility, (2) eligible participants, (3) eligible participants whose consent was not requested, (4) participants who refused to participate at the point of contact in the psychiatric ward, and (5) those who consented to participate. Outcomes were recorded as screened, eligible, consent not asked, refused, and enrolled. In addition, to track the pace of recruitment, a record of all identified, screened, eligible, unwilling, and successfully recruited patients was kept using a specific monitoring sheet developed for the trial. Patient flow was monitored and recorded daily on the study wards. Daily progress of patient recruitment was reported as “recruitment pace” (ie, how many new patients were recruited e each day) [28].

Attrition data were categorized into two groups: (1) attrition during SMS text message intervention and (2) attrition during the 12-month follow-up period [29]. Attrition during the follow-up was then divided into four categories, based on the follow-up data collection method: (1) telephone interview, (2) participants’ notification (ie, withdrew from the follow-up survey), (3) postal survey, or (4) register data retrieval from the Finnish National Care Register for Health Care [30]. Outcomes were reported as dropout rates. In our study, the dropout rates were calculated based on telephone or postal survey responses, or by whose data were not available in register data retrieval due to an incorrectly entered ID.

Descriptive statistics (frequency, percentage, mean, standard deviation) were used to describe participants’ demographic characteristics, recruitment, and attrition metrics (study participants lost in the follow-up). The demographic variables examined included age, gender, marital status (lives alone, ie, single, divorced, or widowed; lives with someone, ie, married), vocational education (none, vocational education), employment status (employed/self-employed, retired, student, job seeker), diagnosis (International Statistical Classification of Diseases and Related Health Problems, Tenth Revision [31]), and age at first contact with psychiatric services. To analyze possible differences between patients who participated in the study and those who dropped out, t tests and chi-square tests were used.

Multiple logistic regression analysis was used to determine predictors of dropping out of the 12-month postal survey follow-up. Participants’ demographic characteristics, including age, gender, marital status, vocational education, and employment status, were chosen as predictors and added to the analysis [8,32]. All data were analyzed using SPSS version 21. A P value <.05 was interpreted as a statistically significant difference.

A total of 11,530 patients admitted within psychiatric inpatient hospital wards were screened during the 15-month (453 days) recruitment period. There were 6565 who did not meet the eligibility criteria. A total of 779 patients dropped out before the eligibility assessment because they were transferred to another ward or rapidly discharged from hospital. Of the candidates who were screened, 36.31% (4186/11,530) appeared eligible.

Of the 4186 eligible patients, informed consent was asked from 3417 (81.63%). Informed consent was not asked in 18.37% (769/4186) of the cases because the person was quickly discharged from the ward, absconded from hospital, or the research nurses simply forgot to ask.

When age and gender of the screened noneligible and eligible patients were compared, it was found that the eligible patients were generally younger than the noneligible patients (P<.001). Men were more often noneligible than women (P<.001) (Table 1).

Out of the 3417 eligible participants whose consent was asked, 2278 patients (66.67%) refused to participate in the study. Although reasons for refusal were not asked due to ethical guideline requirements [33], some patients voluntarily offered explanations, such as they did not know how to use mobile phones or text messages, their mobile phone was broken, or they did not have a mobile subscription at the that time. Patients who refused to participate were older than consenting, randomized patients (P=.001) (Table 1).

The pace of recruitment was analyzed based on the number of new patients recruited each day. At the beginning of the study, recruitment was slow. The recruitment rate reached its peak 15 months after enrollment started. For every 10 screens completed, one person was successfully enrolled, at an average recruitment speed of 76 participants each month (2.5 participants per day).

Of the 1139 patients who were enrolled in the study, the data of 16 participants were excluded due to either the withdrawal of informed consent (n=10), the patient did not meet the inclusion criteria (n=5), or a recruitment error (n=1). This left us with a total of 1123 participants (intervention group: n=563; control group: n=560).

The results of this study demonstrate that it was challenging to recruit, and engage, participants in an SMS text message-based trial follow-up. One-third of patients (36.31%) appeared eligible, and two-thirds of eligible patients (66.67%) refused to participate. Participants were well engaged with the SMS text message intervention provided, but their engagement with the trial follow-up varied: the highest being with the register data retrieval (99.64%) and lowest with the postal survey (47.55%). Participants’ demographic characteristics (age, gender, vocational education, and employment status) were seen as dropout predictors.

In our study, within the context of psychiatric inpatient care, we were able to recruit 1139 individuals (33.33%) out of 3417 eligible participants, whose consent was requested. Age and gender tended to be factors influencing recruitment and refusal. Our refusal rate was 66.67% (2278/3417), which is in line with previous studies suggesting that high refusal rates are a major problem faced during the recruitment process [11,15]. Patients’ illness-related issues [14,19] and perceived stigma were seen as barriers to participation in previous studies [36,37]. This may also be the case in our study, which focused on people treated with antipsychotic medication recruited from hospital wards providing psychiatric care.

Lack of interest in the trial [11], lack of motivation to use mobile technology-based interventions [13], or a lack of capacities may also have affected the refusal rates in our trial. During the participant screening, some patients expressed their inability to use mobile phones or text messaging, stating that as the reason for their refusal. This was unexpected because mobile phone text messages have been proven to be feasible and acceptable among people with severe mental health problems [38], as well as easy to use [35]. Therefore, our findings may reveal that a digital divide may still exist in the area of mental health, although relevant literature identifies this concept as promising [14]. Previous literature also shows that people with mental health problems are able and willing to use mobile-based interventions when given the opportunity [38,39]. In our case, it might have been useful to train those who said they did not know how to use mobile phones, and to offer mobile phones to those who did not have them. This might have given us important knowledge about the digital divide, especially those who are not so familiar with the mobile technology.

The attrition rate during the SMS text message intervention was low (4.8%). This finding does not support previous findings stating that low engagement and discontinuation are major problems in intervention studies [11,14]. However, it is possible that participants who stopped using the SMS text message intervention did not notify the researchers. What we found here was that women dropped out from the SMS text message intervention more often than men did. This is contradictory to the results of Ben-Zeev and colleagues [8], who found that women were significantly more engaged with mHealth interventions than men were. Further, participants’ demographic characteristics (age, gender, vocational education, and employment status) predicted the risk of leaving the study early and, subsequently, not participating in the follow-up. This is in line with previous studies, which reported that participants who left the studies early differed from those who were retained [8,32]. Therefore, it is important to identify feasible and useful mHealth interventions targeting different patient groups and, further, to identify factors that facilitate or prevent patient engagement with mHealth interventions.

Patient engagement in the trial during the follow-up varied depending on the source of data collected, the highest being in the register data retrieval (99.64%) and the lowest in the postal survey (47.55%). Low engagement and discontinuation have been found as fundamental problems in technology-based intervention studies [9,20]. Participants’ age, gender, vocational education, and employment status seemed to predict their risk of dropping out from the postal survey. This group of older, women, and retired participants may also reveal the digital divide in this patient group. It may be that they just do not value technology as much as younger people do and, therefore, did not participate in the postal survey. Recently, however, technology usage has increased among older adults [40]. This is promising and may give researchers new hints on how to engage participants in trials. Given this, it is important to consider which data collection methods are appropriate to use among people with serious mental health problems, and to identify factors that facilitate patient engagement in mHealth-based trial follow-ups.

Our study has some limitations. The recruitment data concerning information about screened patients lacked some information, especially about patients’ ages. Therefore, results related to patient demographics concerning recruitment have to be handled with caution. More importantly, we did not gather knowledge about the participants’ actual SMS text message use. Therefore, we lack knowledge about participants’ true engagement with the SMS text message intervention. However, according to our findings before the study actually started, participants were very satisfied with the intervention [35].

A key strength of this study was in its large nationwide sample of people treated with antipsychotic medication. Another was that, to the best of our knowledge, this was the largest trial evaluating a text message system. Our findings regarding attrition are important for those conducting similar RCTs among people with severe mental health problems, although this group may well have different issues with the technology when compared with others [17,41].

Initial patient recruitment and then engagement in the 12-month postal survey follow-up were the most vulnerable phases in the SMS text message-based trial. This may indicate that people with serious mental health problems may need extra support during the recruitment process, and necessitate further support to engage in completion of these follow-up questionnaires—at least within SMS text message trials.

Researchers should acknowledge the possible digital divide for people with serious mental health problems, and choose convenient and efficient data collection methods for study follow-ups. At follow-up, for Mobile.Net, high-grade routine data were almost complete. Methods of trials should take much more consideration of the nature of the target group of participants; otherwise, evidence is dogged with high attrition with the accompanying speculation of researchers. No statistical technique or learned speculation can make up for loss to follow-up. The solutions are likely to vary for different client groups. We think more research is needed both to investigate the support of the recruitment process and methods of follow-up in technology-based RCTs. Asking people to complete forms that are likely to result in grossly incomplete datasets could be considered an unethical—and potentially dangerous—waste of time and resources.