We conducted the Healthy Campus Trial, a fully factorial trial of 5 iCBT components: SM, BA, CR, AT, and PS. This study was designed as a fully factorial randomized controlled trial. The original protocol [13] explains the finer details of the Healthy Campus Trial.

Undergraduate or graduate students from 5 universities in Japan were recruited between September 2018 and May 2021. Students aged between 18 and 39 years, possessing a smartphone, were introduced to online screening with the 9-item Patient Health Questionnaire (PHQ-9). To examine the effect of smartphone CBT on those with and without mild mental symptoms, based on the screening PHQ-9 scores, we included a random tenth of students scoring 4 or less on the PHQ-9 and all of the subthreshold depressive students with PHQ-9 scores between 5 and 9, or between 10 and 14 with suicidal ideation for less than half of the days. We were able to include only 1 in 10 of the students with no to very mild symptoms because including them all would have made the total sample size 10 times greater, and our budget and manpower could not afford this. Since the focus of the acute phase intervention [12,13] was on those with subthreshold depression, we prioritized their inclusion, while for the long-term outcome, we also wished to examine the effect of the smartphone CBT on those with or without mild symptoms. We excluded students under treatment by any mental health professionals. No student was diagnosed with an MDE at baseline.

All participants received a psychoeducation module. With psychoeducation, participants learned the importance of monitoring their own mental state through the trial duration using the smartphone app “Resilience Training,” which was developed for this study. After psychoeducation and baseline PHQ-9 assessment, iCBT combinations were randomly allocated for each participant among 32 combinations (with or without each of the 5 iCBT components, hence 2⁵=32 combinations, see Multimedia Appendix 1). Participants could be either assigned to or not assigned to each one individually. As a result, some participants might end up with zero iCBT components after the psychoeducation, while others could receive all 5. We used block randomization stratified by participant university and baseline PHQ-9 score (4 or less versus 5 or more). Each component was supposed to take 1 week. All participants, regardless of the number of the allocated components, were expected to finish the app in 8 weeks. Participants work on each component for 7‐10 days. They were able to review any assigned components as often as they liked during the 8 weeks and fill in as many additional worksheets as they wished. The participants had to complete (read all the lessons and complete at least one worksheet) each component before they could advance to the next component. Participants received digital gift cards worth US $9 for completing self-checks in weeks 4 and 24 and US $18 in weeks 8 and 52. In addition, research staff sent regular emails to encourage all participants to move forward with the program and reminder emails to fill in questionnaires if required. These emails did not include any specific guidance on the content of iCBT.

The contents of the five iCBT components were as follows:

All CBT component lessons were delivered through the “Resilience Training” app.

The primary outcome was time to the first MDE by 52 weeks post randomization. The occurrence of an MDE was evaluated by the computerized version of the WHO Composite International Diagnostic Interview (CIDI) version 3.0 depression section, which has demonstrated good reliability [14,15]. The secondary outcomes included depressive symptoms, anxiety symptoms, and 5 CBT skills, as measured by the PHQ-9 [16,17], 7-item General Anxiety Disorder (GAD-7) [18,19], and CBT Skills Scale [20], respectively. The reliability of each questionnaire is well supported by previous evidence, including our study in which the CBT Skills Scale was developed and validated [20]. The timing for each questionnaire is detailed in the original protocol [13]. For the current analysis, the PHQ-9 was measured at baseline and at weeks 1 through 8, followed by every 4 weeks thereafter, up to week 52. The GAD-7 and the CBT Skills Scale were measured at baseline, and at the fourth, eighth, 24th, and 52nd weeks of this study. The CIDI results were obtained at week 52. To ensure participant safety, those with high levels of depression and suicidal ideation measured by the PHQ-9 were monitored and advised via email to seek professional care. All questionnaires were administered using the Japanese version.

We used SAS (version 9.4, SAS Institute) and R (version 4.2.1, R Foundation for Statistical Computing) for all analyses. We performed Cox regression analyses to investigate whether the presence or absence of each CBT component was related to the occurrence of an MDE. The model involves 5 iCBT components, baseline PHQ-9 scores, university, age, and sex as explanatory variables. Allocation of each iCBT component was defined based on an intention to treat (ITT). We plotted Kaplan-Meier survival curves for a graphical representation of MDE-free survival for each iCBT component. Interval censorings were addressed with Anderson-Bergman adjustment methods [21,22]. In addition, changes in PHQ-9, GAD-7, and CBT Skills Scale were repeatedly measured up to 52 weeks and analyzed with mixed models for repeated measures [23]. The correlation matrix of repeated outcomes was assumed to be unstructured. For explanatory variables, 5 iCBT components, week, iCBT component by week interaction, university, age, and baseline scores were modeled. The between-group effect size was estimated by dividing the estimated mean difference by the observed SD of week 52 scores, which is the most widely used definition of standardized difference in clinical trials of psychiatry [24]. In addition, the within-group effect size was estimated by dividing the estimated mean change by the baseline SD.

This study involves human participants and has been approved by the Ethics Committee of the Kyoto University School of Medicine (C1357). All data obtained from participants were deidentified, and participants provided informed consent before participating in this study.

Figure 1 presents the CONSORT diagram. Of the 5063 students assessed for eligibility, 1627 students provided informed consent and were randomly allocated to 1 of the 32 iCBT combinations (Multimedia Appendix 1). Finally, 1626 students were included in this study, after 1 student withdrew consent and refused the use of their data for analysis. Table 1 tabulates the demographic characteristics of the participants. The mean age was 21.5 (SD 3.0) years, and 933 (57.38%) were women. The mean PHQ-9 and GAD-7 scores were 6.4 (SD 3.4) and 5.5 (SD 3.4), respectively. Participants were allocated to each module (SM: n=808; BA: n=817; CR: n=811; AT: n=814; PS: n=811). The characteristics of the participants per component are described in Multimedia Appendix 2, and the number of participants allocated to each of the 32 combinations is described in Multimedia Appendix 1.

Of the 1626 study participants, 1301 (80.01%) responded to the 52-week follow-up survey. Among the responders, 133 (10.22%) reported incident MDE by week 52. The number of completions and their percentages for each assigned module were recorded (SM: 743/808, 92.6%; BA: 703/817, 86%; CR: 688/811, 84.8%; AT: 680/814, 83.8%; PS: 666/811, 82.1%). Table 2 summarizes the results of the multivariable Cox regression analysis of each CBT component on MDE occurrence. The hazard ratios (95% CI) of presence over absence were 1.26 (0.88‐1.79) for SM, 1.09 (0.77‐1.54) for BA, 1.18 (0.83‐1.68) for CR, 0.85 (0.60‐1.20) for AT, and 1.18 (0.82‐1.68) for PS. Multimedia Appendix 3 shows the full results of the analysis with coefficients for the included covariates. The results of the univariable Cox regression analysis (Multimedia Appendix 4) showed similar results to those of the multivariable analyses. Figure 2 presents the Kaplan-Meier survival curves for the occurrence of an MDE for each of the 5 iCBT components. We did not find a significant difference in the occurrence of an MDE in each iCBT component.

Table 3 shows the estimated least squares mean change scores of the PHQ-9 for participants allocated to the presence or absence of each component using the mixed models for repeated measures. Compared to the baseline, the estimated least squares mean change scores ranged between −1.77 and −1.97 at the 52-week follow-up for both the presence and absence groups across all iCBT components. Regardless of the allocated iCBT component, average depressive symptoms were reduced from baseline. However, no significant differences in change scores were found in the comparison between the presence group and the absence group across all iCBT components. Analysis of the GAD-7 showed similar results for anxiety symptoms, as shown in Table 4. Table 5 shows an analysis of the relevant CBT skills at week 52 for each CBT component. There was no difference in changes in CBT skills between the presence or absence of each iCBT component.

We conducted a fully factorial randomized trial of 5 iCBT skills to reduce the incidence of MDEs among university students. We did not observe significant preventive effects attributable to individual iCBT components in depression prevention. Secondary analyses showed no significant difference in change scores for PHQ-9, GAD-7, and CBT skills, regardless of the presence or absence of each iCBT component either. Participants exhibited improvements in depression and anxiety symptoms from the baseline to the 52 weeks, irrespective of the specific iCBT component allocation.

The significantly better PHQ-9 and GAD-7 scores compared to baseline at 52 weeks, irrespective of the iCBT component, replicate the results from our previous 8-week short-term analysis [12]. Multiple extra-intervention factors may have contributed to the general improvement effects. First, the frequent psychological assessments administered to all participants may have introduced a significant nonspecific therapeutic effect. This hypothesis is supported by findings from a prior meta-analysis of antidepressant efficacy trials, which demonstrated that follow-up assessments significantly influence placebo responses [25]. In this study, frequent follow-up visits significantly impacted depressive scale scores, where the extra follow-up visits accounted for approximately 34%‐44% of the overall placebo response observed over the course of the trials. These results emphasize that the therapeutic effect of these assessments is not only statistically significant but also cumulative and proportional to the number of interactions, thereby suggesting a potent nonspecific therapeutic effect stemming from the assessment frequency itself [25].

In our study, a rigorous schedule of assessments was implemented, occurring as many as 19 times throughout the duration of this study (specifically at weeks 1-8, 12, 16, 20, 24, 28, 32, 36, 40, 44, 48, and 52). This high frequency of assessments might have played a significant role in mediating a therapeutic effect, potentially confounding the distinct impacts attributable to each individual iCBT component. It is important to note, however, that all participants in this study were subjected to the same number of assessments, making it challenging to isolate the effects of the assessments from those of the iCBT interventions themselves. Consequently, while it is tempting to speculate about the influence of frequent assessments, such speculation cannot be conclusively confirmed given the uniform assessment protocol applied across all participants.

Given these considerations, it is recommended that future research should explore the implications of varying the frequency of assessments by including experimental groups subjected to differing assessment schedules. This approach would allow for a more refined analysis of how frequent interactions might influence outcomes, thereby enabling a more accurate assessment of the true efficacy of the iCBT components. By distinguishing the effects of the intervention from those of the assessment frequency, future studies could yield more reliable and generalizable findings.

In addition to frequent assessments, application help notifications coupled with encouragement emails may have produced extra-intervention improvement effects. Regardless of allocated iCBT components, participants were instructed to keep the app installed until the end of the 52-week follow-up. This instruction also conveyed crisis management information, such as where to go for help in case of serious problems. Providing emergency contact information was reported to reduce rates of repeated self-harm [26]. Thus, it is possible that the crisis management information in our study could have contributed positively to mental health conditions. Future studies should consider incorporating groups with different frequencies of self-assessments to derive more reliable results.

Together with the frequent follow-up assessments and application help notifications, university grade level could have affected the results. A longitudinal study reported that depressive and anxiety symptoms were better in the fourth year of university than in the first 3 years [27]. This trend toward improvement may mask the effect of our iCBT intervention. Similarly, baseline depression level may have affected our results. As depression symptoms become milder, placebo effects become more prominent relative to the intervention. This assertion is supported by a meta-analysis examining the effects of antidepressants compared to placebo [28]. If we targeted university students with more severe depression symptoms, we may have a greater ability to detect intervention effects. On the other hand, another study assessing the effects of an iCBT program with healthy participants showed a significant improvement in subthreshold depressive symptoms [10]. Therefore, our failure to identify the effects of the intervention may not be attributed solely to the depressive severity of the participants.

The latter half of our study unfolded during the COVID-19 pandemic, accompanied by extensive social restrictions. This period has been scrutinized across various studies, notably in a systematic review of epidemiological research which reported a temporary surge in depression and anxiety within the first 3 months of the pandemic [29]. However, our own secondary analysis, using the same data set [30], did not find a statistically significant link between depressive symptoms, as measured by the PHQ-9, and the implementation of 4 distinct states of emergency in Japan aimed at mitigating the spread of the virus. Although our results suggest no direct correlation between these emergency measures and depressive symptoms, the overarching impact of the pandemic—its indirect influence on mental health through other unmeasured factors—remains a plausible explanation for variations in our findings. Consequently, we acknowledge the complexity of the pandemic’s effects, which may interact with psychological outcomes in nuanced and varied ways that are not immediately apparent in our data.

The validated CBT Skills Scale [20] was used to measure the main 5 CBT skills. The results at 52 weeks showed no significant difference in change scores with or without each of the iCBT components. This finding was unexpected as our previously reported short-term analysis revealed significant improvement of CR and AT skills [12]. The 52-week results lack of significant differences may be due to the attenuation of skills learned during the first 8-week intervention period. However, some studies reported that the effect of CBT lasts longer than 46 months as an adjunct to pharmacotherapy for treatment-resistant depression [31]. These studies focused only on symptoms, not on CBT skills. Our report is the first to assess the persistence of CBT skills using our original CBT Skills Scale. Further research is needed to determine how long CBT skills are usually maintained. It is possible that maintenance CBT may be effective in maintaining CBT skills, despite the reported fact that maintenance CBT on depression prevention is superior to psychoeducation only among patients with a high risk of depression relapse [32].

There are several limitations in our study. First, our participant pool was limited to 2 prefectures in Japan (ie, Kyoto and Aichi). Nonetheless, this well-powered multisite trial experienced minimal attrition and included 3 types of universities, national, public, and private, which attracted students from all regions of Japan. Additionally, we ensured adequate allocation concealment during randomization, resulting in a well-distributed sample. Nevertheless, caution is advised when generalizing these results. Second, our study did not account for potentially significant protective factors, such as good academic performance and financial stability. Although our research used a fully factorial design intended to evenly distribute unknown confounders across study groups, we cannot definitively exclude the influence of unmeasured variables. Third, our measures of MDEs, mental state, and CBT skills relied on self-administered questionnaires that, despite their established reliability, are susceptible to biases such as social-desirability. This susceptibility potentially obscures the true symptoms of depression, and the accuracy of the PHQ-9 may also be compromised by the repeated measurement effect. To mitigate these limitations and enhance assessment accuracy, face-to-face evaluations conducted by trained professionals would be more ideal. Finally, the validated CBT Skills Scale used in our study may not have been sensitive enough to capture subtle differences in skill acquisition and retention over the extensive follow-up period. While the scale is robust for assessing general CBT competency, it may not adequately reflect incremental changes or the long-term retention of specific skills. Future research should use such tools to ensure a more precise and sensitive evaluation of the long-term skill acquisition and retention of CBT skills.

This study did not show the specific impacts of individual components of iCBT on preventing MDEs. However, participants’ scores on the PHQ-9 and GAD-7 generally improved over the year during their involvement in this study. In addition to the statistical phenomenon of regression to the mean and the natural course of the conditions, the psychoeducation given to all participants, the frequent psychological assessments, and the experience of being involved in this study itself may have contributed to some alleviation of depression and anxiety symptoms among university students. Future iCBT optimization trials should consider psychological assessment frequency, the impact of common intervention elements, the natural history of the mental state of the target group, and the baseline depression level.