People living in urban areas have prevailing complaints of stress and related health concerns. For instance, the general population of Hong Kong has a stressful life, with an estimation of 13.3% having common mental disorders [1]. Stress was found to be correlated with multiple health issues, including insomnia, chronic pain, and unhealthy eating behaviors [2-4]. In the United Kingdom, the economic burden of low back pain, insomnia, and eating-related conditions was estimated to be £2.79 billion, £46.3 billion, and £8.5 billion, respectively (US $1=£0.77), projected into 2018 costs [5-7]. In Hong Kong, insomnia and chronic pain affect 35.2% and 39.4% of the general population, respectively, and close to half (46.9%) of the general population was found to have unhealthy eating habits [8-10].

To tackle these health concerns, mindfulness-based interventions (MBIs) have been applied to alleviate stress and enhance well-being. Mindfulness is defined as “paying attention in a particular way on purpose in the present moment and nonjudgmentally” [11]. In particular, MBIs were found to be effective in improving insomnia [12,13], pain [14-16], binge eating, and emotional eating [17,18]. MBIs have not only been applied to clinical populations, but they have also been demonstrated to be beneficial to the well-being of nonclinical populations. Meta-analyses showed that MBIs have moderate effect size in reducing stress, psychological distress, depression, and anxiety among healthy individuals [19,20]. Internet-based MBIs are gaining more evidence as well. According to another review and meta-analysis, web-based MBIs are effective in reducing depressive symptoms, anxiety symptoms, stress, and improving well-being and mindfulness with small-to-medium effect size [21]. The above evidence provides the foundation to further develop internet-based MBIs for both clinical and healthy individuals.

Regarding the mechanisms of MBIs in the promotion of well-being and the reduction of distress, the cultivation of mindfulness is found to mediate the relationship between MBIs and various outcome variables. For instance, Nyklíček and Kuijpers [22] found that changes in mindfulness partially mediated the relationship between mindfulness-based stress reduction intervention and its positive effects among people with distress symptoms. In addition to changes in mindfulness, according to Shapiro et al [23], the cultivation of mindfulness may also facilitate “reperceiving,” which is a shift of perspective that leads to an increased capacity for relating to one’s internal or external experiences objectively. For insomnia, mindfulness allows people to respond to stressors more skillfully by changing the patterns of worry and rumination that improve sleep quality [24]. For pain, preliminary evidence suggested that mindfulness and pain catastrophizing mediated the relationship between interventions and reduced perceived stress and improved quality of life [22,25]. For dysfunctional eating, MBI regulates appetitive and emotional processes by increasing both awareness and sensitivity to the eating process such that people can disengage themselves from the reactive eating habits [26]. To further test these mechanisms in this study, we hypothesized that text messaging–based MBIs would promote better well-being and fewer symptoms of insomnia, pain, and unhealthy eating through changes in both mindfulness and reperceiving.

Although internet-based MBIs were effective in reducing distress and promoting well-being [21], users’ retention and engagement remain challenging for digital health interventions [27,28]. A median of 56% retention was found, and the attrition rate can be as high as 75% for internet-based interventions [29,30]. Text messaging, considered as one of the future trends of internet-based interventions, may alleviate this problem, given the temporal synchronization feature of instant messaging that allows the intervention to catch people’s attention directly [31,32]. A systematic review showed that text messaging improves treatment adherence and reduces social isolation [33]. Up till now, to the best of our knowledge, text messaging is mostly used as a reminder, self-monitoring tool, or as an adjunct to face-to-face intervention instead of being utilized as the core treatment modality, except in a few physical health intervention studies [29,34-38]. Thus, the potential for text messaging to offer mental health intervention that improves users’ engagement and retention is yet to be investigated.

The aim of this study was to test whether a text messaging–based MBI is effective in reducing distress related to insomnia, pain, and dysfunctional eating in 3 randomized controlled trials. WhatsApp [39] was chosen as the messaging tool because it is the most popular in the world with 2000 million monthly active users and is the most accessible option for the community [40]. Specifically, we aimed to investigate whether text messaging–based MBI can result in improvement in primary outcomes (ie, insomnia severity, pain severity, dysregulated eating) and secondary outcomes (ie, depressive and anxiety symptoms, mental well-being, and functional impairment). We also aimed to examine mindfulness, dysfunctional beliefs (for insomnia), pain catastrophizing (for chronic pain), and power of food (for dysregulated eating) as possible mechanisms of change that mediate the relationship between text messaging–based MBI with both primary and secondary outcomes.

WhatsApp numbers for each trial were disseminated in the recruitment materials, and interested participants enrolled in the relevant trial via the instant messenger app, that is, WhatsApp. The nature and procedure of the studies were then explained to the participants through a WhatsApp message. Participants then completed the self-assessment web-based questionnaire together with a written consent page. Screening for eligibility of the participants was done upon completion of the preintervention questionnaire. Eligible participants were grouped into monthly batches, and the second author handled the enrollment, randomization, and intervention assignment of the participants. The randomization was conducted with Microsoft Office Excel 2010 [41] on an individual basis and 1:1 ratio between the intervention or wait-list control conditions with stratification on age (cutoff at 45 years old) and prior experience of mindfulness practice (no experience at all vs not novice). Participants were informed about their allocation of condition via WhatsApp by the same person who did the randomization. No blinding was feasible given the use of wait-list control in the design. A feasibility study on text messaging–based MBI had been done prior to this study. The intervention had a similar structure with this study. It was an uncontrolled study aiming at examining the impact of a general MBI, and the results revealed an improvement in the well-being and mindfulness of the participants. Therefore, the research team proceeded to these randomized controlled trial studies. Preintervention, postintervention, 1-month, and 3-month follow-up questionnaires were distributed via WhatsApp using Qualtrics [42]. Pretesting of the questionnaire before launch was done by the second author. The questionnaires were divided into 4 pages, namely, the consent form, demographic and screening items, primary outcomes, and secondary outcomes. The questionnaire pages were limited to around 50 items per page. Cookies were enabled, and participants could resume their questionnaire on the same device. Participants were required to input their research ID in the questionnaire, and duplicated data with the same research ID were eliminated. The postintervention questionnaires were sent to participants in both wait-list control and intervention conditions on the 22nd day. To adopt an intent-to-treat approach, all participants were invited to complete the self-report questionnaires at every time point regardless of whether they had completed the questionnaires at previous time points. Wait-list control participants received the intervention after completion of the 3-month follow-up questionnaire. Data were exported and stored as a password-protected Excel file in an encrypted flash drive.

These studies were conducted in accordance with the ethical codes of the Declaration of Helsinki. Ethics approval was obtained from the Joint Chinese University of Hong Kong-New Territories East Cluster Clinical Research Ethics Committee (2017.203-T). Findings in this paper are reported in accordance with the recommendations of the Consolidated Standards of Reporting Trials (CONSORT).

Participants were recruited from the general public via social networking platforms, a health-related information website, a local magazine, mass emails, and announcement postings at local tertiary institutions, collaborating nongovernmental organizations, and mutual-aid groups. Participants were quasi-anonymous. Multiple registrations to different trials were eliminated. Participants were only allowed to join one of the trials. The recruitment and follow-up lasted from July 2017 to October 2018; HKD 100 (US $12.87) was offered to 10 participants from each trial as an incentive via random draw. Eligibility criteria for participation included (1) age of 18 years or older, (2) ability to understand Cantonese and give consent, and (3) adequate level of computer literacy to follow the web-based instructions independently, together with daily access to the internet. Participants who self-reported receiving psychiatric services or active suicidality were excluded from this study; hotline and related resources were provided to those participants. No harm or other unintended effects were noticed in both conditions across all 3 trials. In the 3 trials, 364, 264, and 371 eligible participants were recruited for insomnia, pain, and dysregulated eating trials, respectively. Some of the participants withdrew from the studies or became out of reach after the assignment of condition. See Figure 1, Figure 2, and Figure 3 for the CONSORT diagrams of the 3 randomized controlled trials.

All MBIs were delivered in 3-week packages, with 21 days of daily mindfulness exercises that consist of an audio clip and a psychoeducational article delivered via WhatsApp. Participants were grouped into monthly batches, and they received the same messages on a fixed schedule. The intervention was technician-assisted when participants had any specific questions. One technician was involved, and the interaction was confined to confirmation of enrollment, explanation of the study procedures, general enquiry, and reminder for data collection as well as monitoring adherence. Standardized materials were delivered with broadcast function without other personalized messages. To enhance adherence, brief mindfulness exercises were used, which took around 10-15 minutes to complete. Participants were invited to indicate their adherence by replying on WhatsApp after finishing each exercise. Reminder messages were sent to those participants who had not completed any exercise within a week. See Multimedia Appendix 1 and Multimedia Appendix 2 for the outlines of the interventions. The delivery format referenced a WhatsApp-based intervention of Acceptance and Commitment Therapy with daily passage and audio clip [43]. Participants received text and audio from the research team through WhatsApp. Notifications were pushed. Progress could be tracked by checking if the participants had seen the message and listened to the audio in WhatsApp. See Multimedia Appendix 3 for the screenshots.

For insomnia and pain, the intervention content was designed by clinical psychologists referencing the mindfulness-based stress reduction program [44]. Initially, the intervention cultivates participants’ focused observation and awareness on bodily sensations, emotions, as well as thoughts. Practices included mindful breathing, mindful stretching, and body scan. Examples that are specific to insomnia and pain were included to increase relevance. Participants were encouraged to stay with aversive experiences with an open attitude. Lastly, suggestions were introduced to participants for integrating mindfulness practices in their daily life. For dysregulated eating, the MBI for eating incorporated 2 components running in parallel: (1) general mindfulness exercises and (2) specific eating-related components drawn from the Mindfulness-based Eating Awareness Training [26]. General mindfulness exercises started with observation and awareness of bodily sensations, emotions, and thoughts. Exercises on responding to unpleasant experiences were introduced. Meanwhile, eating-specific exercises were introduced to facilitate awareness of the eating experience. Specific exercises covered physical sensations of hunger and fullness, satisfaction of taste, changes in overall physical status, emotional triggers of eating, as well as choosing food with mindful awareness.

Three sets of questionnaires were designed for each trial. Primary outcomes were used only in specific trials that included the severity of the targeted health concern itself and the attitude toward the health concern. Secondary outcomes were used across 3 trials that covered functional impairment, emotional disturbance, and mental well-being, while mindfulness and specific reperceiving variables were measured as potential mediators. Participants were also invited to provide demographic information such as age, gender, religious belief, education level, employment status, personal income, and prior experience in meditation. Usage of medication to cope with sleep disturbance or pain was also measured. Use of mental health services was also recorded.

Significance tests such as the independent 2-sided t tests and chi-square test of independence were employed to explore any baseline difference between completers and dropout participants. Intent-to-treat analyses were used in this study. Data of all the participants were included in the analysis regardless of their treatment adherence or attrition. Missing data were treated using multiple imputations, and 100 imputed data sets were generated. van Ginkel and Kroonenberg’s [76] method of repeated measures analysis of variance (ANOVA) using imputed data was used in this study. Specifically, missing values were estimated and 100 plausible complete versions of the data sets were created. Results from these 100 data sets were then pooled into 1 analysis by applying Rubin’s [77] pooling procedures. The pooling procedure was carried out using an SPSS macro by van Ginkel [78]. Treatment condition (intervention condition vs wait-list control) and time were entered as the fixed between-group and within-subject factors, respectively. To examine the mediating effects of mindfulness and reperceiving (ie, dysfunctional beliefs and attitudes about sleep, pain catastrophizing, power of food) on the relationship between condition and the primary and secondary outcomes, path analyses were conducted using Mplus 7 [79]. In the path analyses, condition was dummy coded (with the wait-list control condition coded as the reference group) and was treated as the independent variable. Mindfulness and condition-specific reperceiving mediator at postintervention assessment were treated as the mediators. Primary and secondary outcomes at 1-month and 3-month follow-up assessments were treated as the dependent variables. Baseline scores of all variables included in the model were controlled.

The demographic data and baseline characteristics of the participants are summarized in Table 1. Overall, the dropout rate was consistent between the intervention and wait-list control conditions (insomnia: χ²₁ (N=333)=0.2; P=.62; pain: χ²₁ (N=235)=1.2; P=.27; dysregulated eating: χ²₁ (N=351)=0.1; P=.80). Chi-square tests and independent t tests revealed no significant association between the demographics variable and the dropout rate. No significant baseline difference between completers and dropout participants on any outcome measures was found.

Correlations among the variables are presented in Table 2. Repeated measures ANOVA with the imputed data revealed significant intervention by time interaction effect on insomnia severity (F_3,790.1259=7.434; P<.001), somatic presleep arousal (F_3,887.221=4.504; P=.004), cognitive presleep arousal (F_3,795.662=5.286; P=.001), dysfunctional beliefs and attitude about sleep (F_3,775.253=5.784; P<.001), mindfulness (F_3,749.436=3.590; P=.01), depression (F_3,838.002=3.938; P=.008), anxiety (F_3,845.675=4.554; P=.004), and mental well-being (F_3,826.162=3.482; P=.02), whereas the intervention by time interaction effect on functional adjustment was found to be nonsignificant (F_3,681.529=0.964; P=.41). Further analysis revealed that the intervention condition showed a better outcome on most of the outcome measures at postintervention, 1-month, and 3-month follow-up, compared with its respective wait-list control condition (see Table 3). Effect sizes are presented in Table 4.

Correlations among variables are presented in Table 6. Repeated measures ANOVA results showed significant intervention condition by time interaction effect on pain acceptance (F_3,535.514=4.186; P=.006), and pain catastrophizing (F_3,558.550=3.179; P=.02), mindfulness (F_3,597.884=3.003; P=.03), depression (F_3,592.642=2.781; P=.04), and mental well-being (F_3,593.052=3.762; P=.01). Meanwhile, interaction effects were found to be nonsignificant on subjective intensity of pain (F_3,591.397=0.464; P=.71), functional adjustment (F_3,541.571=1.720; P=.16), and anxiety (F_3,569.171=2.230; P=.08). Further analysis showed that the significant interaction effects were largely maintained across postintervention, 1-month, and 3-month follow-ups. Participants in the intervention condition reported better outcomes on most of the measures (See Table 7). Effect sizes are presented in Table 8.

Correlations among variables are presented in Table 10. Results of repeated measures ANOVA showed significant intervention condition by time interaction effect on uncontrolled eating (F_3,912.364=3.041; P=.03), emotional eating (F_3,944.138=4.294; P=.005), mindfulness (F_3,854.310=2.670; P=.05), and mental well-being (F_3,882.777=4.457; P=.004). Meanwhile, analyses on other measures showed nonsignificant results. For the outcome with significant results, further analyses showed that significant group difference was maintained and the intervention condition reported better outcomes at postintervention, 1-month, and 3-month follow-up (See Table 11). Effect sizes are presented in Table 12.

This study hypothesized that text messaging–based MBI is effective in reducing distress related to insomnia, pain, and dysfunctional eating. It was encouraging that effectiveness was demonstrated in all 3 randomized controlled trials. Most outcomes had improved significantly at 1-month follow-up compared to wait-list control condition and some effects were able to sustain at 3-month follow-up, including both primary outcomes (eg, insomnia, pain, dysregulated eating indicators) and secondary outcomes (eg, depressive, anxiety symptoms). Our findings showed that text messaging–based interventions not only serve as an adjunct or supplementary tool for traditional treatment programs but also can serve as the core treatment modality. Not only the effectiveness is promising, the retention rates are also high across 3 trials at postmeasurement for intervention groups, ranging from 72.9% (129/177) up to 83.2% (139/167), which demonstrated that text messaging–based intervention is a feasible solution that can maintain a high retention rate. This study contributed to the knowledge of how we can utilize the text messaging application to a fuller extent for intervention purposes so that people with a range of health-related issues can consider evidence-based options delivered via mobile messaging technology.

The results of the randomized controlled trials demonstrated the effectiveness of the text messaging–based MBI on insomnia. Primary outcomes, namely, insomnia intensity, dysfunctional beliefs and attitudes about sleep, and presleep arousal were found to be significantly reduced compared to those in the wait-list control condition, and the effect was sustained at 3-month follow-up. Significant improvement of secondary outcomes, including depressive symptoms, anxiety symptoms, and mental well-being, were also found with sustained effect at 3-month follow-up. The results were consistent with those reported in previous studies on MBIs for insomnia that MBIs are effective in improving insomnia, depression, and anxiety and increasing mindfulness [12,13].

Dysfunctional beliefs and attitudes about sleep mediated the effect of the intervention on all primary outcomes and most secondary outcomes at both 1-month and 3-month follow-up. Mindfulness was found to mediate the effect of the intervention on presleep arousal (cognitive) at 1-month follow-up with additional mediating effects on presleep arousal (somatic) at 3-month follow-up. The mediating effects of dysfunctional beliefs and attitudes about sleep on the outcomes aligned with the mechanism proposed that it was suggested that mindfulness can improve sleep by changing the pattern of worry and rumination [24]. In this study, the mediation model showed that improved sleep might be made possible through reduced dysfunctional beliefs about sleep. However, it was unclear why mindfulness has a mediation effect only on the primary outcomes but not on the secondary outcomes. This pattern might be explained by the specificity of content that was targeted at insomnia.

The results of repeated measures ANOVA showed that MBIs are effective in enhancing chronic pain acceptance and reducing pain catastrophizing at 3-month follow-up in the intervention group compared to those in the wait-list control group. Mindfulness, depressive symptoms, and mental well-being were also found to improve significantly at 3-month follow-up. However, no significant improvement was found in pain intensity, which can be explained by the non–symptom-focused approach of acceptance as it aims to improve pain adjustment independent of pain intensity [58]. Other results were largely consistent with previous research that MBI improved pain acceptance, pain catastrophizing, mindfulness, depression, and well-being [14-16]. However, anxiety was not reduced significantly, which might be explained by the floor effect that the baseline score was low. For the mediation model, preliminary evidence suggested that pain catastrophizing mediated the effect of MBI on pain intensity and functioning at both 1 month and 3 months, whereas mindfulness was only found to be mediating the effect of intervention on secondary outcomes but not primary outcomes. Our findings were consistent with a previous study conducted by Elvery et al [25] that pain catastrophizing emerged as the most robust process and it was the most predictive of pain intensity compared to pain acceptance and mindfulness. However, other studies showed a different process and mechanism wherein pain catastrophizing did not mediate the outcome but pain acceptance did, which might imply that the psychological process may be more important than cognitive process [80]. Further research is needed to verify the mechanism of how MBI leads to change in various outcomes.

MBI was found to be effective in improving uncontrolled eating, emotional eating, and well-being at 3-month follow-up compared to the wait-list control condition. Nonsignificant results were found in other primary and secondary outcomes. However, for data without 3-month follow-up, a significant interaction effect was found for mindfulness. Owing to the complexity of eating issues, the materials used in this condition may not be focused enough to induce significant and sustained change. Preliminary moderation analysis showed that participants with higher uncontrolled eating scores showed greater reduction in uncontrolled eating at postintervention in this study. This finding implies that the intervention may have greater benefits for people who have more severe uncontrolled eating problems. The results of the mediation analysis showed that the power of food mediated the effect of intervention on both uncontrolled and emotional eating at both 1-month and 3-month follow-ups. Mindfulness was also found to mediate the effect on depressive symptoms at both 1-month and 3-month follow-ups. The pattern of mediation is similar to that in the pain condition that mindfulness only mediates secondary outcomes but not primary outcomes. Given little research has been conducted on the relationship between MBI and eating concerns, further research is needed to confirm the model.

This study took a significant step in demonstrating the effectiveness of using text messaging–based MBIs to alleviate distress related to insomnia, pain, and dysregulated eating, which were affecting at least a quarter of the general population. Such interventions are potentially scalable to the population level and can be widely disseminated with relatively low costs and human resources. This novel way of delivering MBIs showed a high retention rate ranging from 72.9% (129/177) to 83.2% (139/167) for intervention groups at postintervention, which is higher than the median retention rate of internet-based interventions [29]. Given the potency of the text messaging MBIs in alleviating specific health concerns and general psychological distress, text messaging–based intervention deserves more attention in the future in delivering other forms of psychological interventions. Not only as adjunct to traditional face-to-face psychological interventions, text-messaging interventions can potentially become a core intervention modality that has high user engagement and positive treatment outcomes. This study also shed light on the mediation models of MBIs, which have not been thoroughly investigated. Across the 3 randomized controlled trials, mindfulness was found to mediate the effect of intervention on both primary and secondary outcomes. Condition-specific mediators were also found, including dysfunctional beliefs and attitudes about sleep for insomnia, pain catastrophizing for pain, and power of food for dysregulated eating. These findings supported that the cultivation of mindfulness and the ability to reperceive the present experience can alleviate health-related concerns and distress. Future designs for MBIs can consider focusing on these mechanisms of change. Nonetheless, replication is needed to confirm the mediation processes and how different mechanisms are related to different outcomes.

This study had several limitations that warrant attention. First, wait-list control was used instead of active control that can account for demand characteristics. As this study focused on investigating the effects of text messaging–based interventions among urban dwellers, comparison with wait-list controls is closer to the real-life situation where people generally do not seek help for their health-related conditions. Even though the placebo effect cannot be ruled out, as the active control group was not included [81], these findings demonstrated effectiveness in a wide range of psychological outcomes for 3 common health concerns.

Second, this study solely adopted self-report measures that may lack objectivity on the severity of health concerns such as sleep habits, pain severity, and eating patterns. Given the interventions were delivered over WhatsApp, the use of self-report measures that are completed online is consistent with accessing the intervention materials over mobile text messaging. Nevertheless, future studies may consider including behavioral or physiological measures to corroborate with self-report findings.

Third, the preponderance of the participants were women. The results of this study may not be generalizable to other genders. The skewness is consistent with gender difference in help-seeking, where men are less likely to seek help than women [82] and women are more interested in practicing mindfulness [83]. A systematic review also found that among 117 studies of randomized controlled trials of mindfulness-based cognitive therapy or mindfulness-based stress reduction with 9820 participants, only 29% of the total participants were men [84]. Future studies on smartphone-based MBIs should sample more men to examine whether these interventions may be more acceptable to men and conducive to their well-being.

Lastly, despite the high retention rates at postmeasurement, the return rates of questionnaires were low at 3-month follow-up, ranging from 61% (72/118) to 67.1% (112/167) for intervention groups and 68.7% (114/166) to 70.1% (82/117) for wait-list control groups. In future studies, strategies are needed to boost the return rates of questionnaires so that the sustained effectiveness can be captured more accurately. One of the possible solutions may be increasing the incentive for completing questionnaires at long follow-up periods.

To conclude, this study showed that text messaging–delivered MBIs are effective in improving issues related to sleep, pain, and eating. Text messaging has the potential to be a core intervention modality to cater to the needs of people with a fast-paced lifestyle or increase accessibility to MBIs. The demonstrated mechanisms of change illuminate directions for future design of materials and focus on MBIs. Given the increasing health needs of the general population and low availability of evidence-based face-to-face interventions, text messaging–based interventions provide a viable alternative to expand on the availability of effective interventions to the public.