Effectiveness of app-based cognitive behavioral therapy for insomnia on preventing major depressive disorder in youth with insomnia and subclinical depression: A randomized clinical trial

This was a randomized, assessor-blind, parallel group-controlled trial conducted in Chinese youth (aged 15−25 years) with insomnia disorder and subclinical depression. Potential participants were recruited from universities, high schools, and communities in mainland China and Hong Kong from September 9, 2019 to November 25, 2022. They were invited to complete online screening assessments to confirm the presence of moderate-to-severe insomnia symptoms (Insomnia Severity Index [ISI] [17] score ≥15) and subclinical depressive symptoms (Patient Health Questionnaire-9 [PHQ-9] [18] score greater than 4 but less than 20). Participants who met the screening criteria were invited for a telephone diagnostic interview to ascertain the following inclusion criteria: (1) presence of insomnia disorder based on International Classification of Diseases, 10th Revision diagnostic criteria, defined as insomnia symptoms ≥3 times per week, accompanied by significant distress or functional impairments for at least 1 month, which cannot be adequately explained by another sleep–wake disorder, mental disorder, or medical condition [19] and (2) absence of a current diagnosis of MDD or a prior episode of MDD within past 2 months by the Mini-International Neuropsychiatric Interview (MINI) [8]. Additional details about inclusion and exclusion criteria can be found in the research protocol (S1 Protocol).

All eligible participants were randomly assigned to 6-week app-based CBT-I or 6-week app-based health education (HE) at 1:1 ratio, irrespective of any pharmacological treatments for insomnia that the participant has been receiving (Fig 1). Assessments, including telephone clinical interviews and online self-administered questionnaires, were conducted at baseline, post-intervention, 6- and 12-month follow-ups after intervention. Two additional assessments were conducted upon the completion of sessions 2 (post-session 2) and 4 (post-session 4) to evaluate sleep and mood symptoms. If participants experienced a depressive episode before the 12-month follow-up, they were considered as completed cases that met the endpoint of the study. The study was reported in accordance with the CONSORT Guidelines (S1 Checklist).

Written informed consent was obtained from participants aged 18 and above. For participants under 18, assent from them and consent from their parent(s)/caregiver(s) were obtained. Ethical approval for the study was granted by Joint Chinese University of Hong Kong-New Territories East Cluster Clinical Research Ethics Committee (CREC Ref. No.: 2019.044) and Medical Ethics Committee of Peking University Sixth Hospital (Issue No. 21 [2019]). The trial is registered with Clinical Trials.Gov (NCT04069247↗).

Randomization was stratified by sex and insomnia severity (15 ≤ ISI ≤ 21 versus ISI > 21) at baseline. A research assistant who was not involved in the study generated the random number sequence for this trial using RAND and RANK functions in Excel with a block size of four. Study participants were not informed of their group assignment, although it was likely that most of the participants were aware of their allocation based on the content of the program. Trained investigators who conducted telephone clinical interviews were masked to group allocation.

The primary outcome was time to onset of MDD determined by telephone clinical interviews using the Chinese version of MINI at post-intervention, 6- and 12-month follow-ups. The MINI has been validated in Chinese psychiatric patients with adequate psychometric properties (sensitivity for diagnosing depression = 92.2%; specificity = 86.0%) [20]. During each interview, the evaluation covered the period from the previous assessment to establish the onset of a depressive episode. In cases at which the participants were unable to recall the exact date, they were asked to indicate the closest week or month, and the midpoint of that week or month was used as the reference point. To assess interrater reliability, interviews were recorded and independently reviewed by a second investigator. The κ coefficient for interrater agreement regarding the diagnosis of MDD was 0.82 based on randomly selected data from 12% (85/708) of the participants, indicating excellent agreement.

The secondary outcomes included depressive symptoms and the measures of insomnia at both symptom and disorder levels. Remission of insomnia disorder was determined by the following criteria [21]: (1) no functional impairments or distress with sleep, (2) insomnia symptoms less than three times per week for at least 1 month, and (3) sleep-promoting medication use less than once per week in the past month. Although the use of sleep-promoting medication is not a standard criterion for diagnosing insomnia disorder, it may mask the underlying symptoms and was thus taken into account in the current study [21]. Changes in depressive and insomnia symptoms were assessed with PHQ-9 and ISI, respectively.

Other prespecified outcomes included anxiety symptoms (Generalized Anxiety Disorder 7-item [GAD-7]) [22], suicidal ideation (Scale for Suicide Ideation-Current) [23], Fatigue (Multi-dimensional Fatigue Inventory) [24], circadian preference (reduced Morningness-Eveningness Questionnaire) [25], sleep-related beliefs and attitudes (Brief version of Dysfunctional Beliefs and Attitudes about Sleep) [26], and sleep parameters including time in bed (TIB), TST, sleep onset latency (SOL), wake after sleep onset (WASO), sleep efficiency (SE) measured by 7-Day Sleep Diary.

Adverse events related to the interventions that occurred from randomization until the 12-month follow-up were recorded during telephone interviews by asking participants if they experienced any adverse events attributed to the interventions from the previous assessment to the current interview, and to specify if they answered yes.

In the study protocol, multiple outcomes were designated as primary outcomes based on the study hypothesis, including depression and insomnia at both symptom and disorder levels. The presence of multiple primary outcomes and the lack of a defined hierarchy may pose some concern to the interpretation of the major findings of the study. Therefore, we reorganized the hierarchy of outcomes when reporting the study results to ensure that the trial was chiefly judged by the outcome for which it was fully powered, namely the incidence of depressive disorder.

The sample size calculation was based on the incidence of depressive disorder. As the study sample included a mixture of individuals with and without prior history of MDD, a 50% increase in the 1-year incidence of depression was expected compared to that (11%) reported in the previous meta-analysis on patients with insomnia [6]. According to previous research on self-help intervention for prevention of MDD [27] and data on CBT intervention for depression in youth population [28–30], we expected a 50% reduction in the incidence rate for the intervention group. Based on a 1-year incidence rate of 16.5% (11% × 1.5), power analysis indicated that 282 participants per condition were needed to detect a hazard ratio (HR) of 0.5 under 1:1 randomization at α < 0.05 (two-tailed) with 80% statistical power. To account for 40% potential attrition rate, a total of 940 participants would be initially needed. In the amendment of study protocol, the recruitment was reduced to a sample size of 708 (354 per condition) based on an estimated attrition rate of approximately 20%, as the attrition rate was much lower than initially expected.

The statistical analysis plan was finalized before the completion of data collection (S1 Protocol), except for subgroup analyses of the primary outcome and exploratory analysis controlling for potential confounders affecting the primary outcome, including prior history of MDD, educational level, family income, comorbid medical illnesses, and use of sleep-promoting medications [31], to test the robustness of the results. Analyses were conducted based on the intent-to-treat approach. Kaplan-Meier survival analysis and Cox proportional hazard regression model were employed to determine differences in time to onset of MDD between the intervention and control groups over 12-month follow-up. Remission rates of insomnia were analyzed using a weighted generalized estimating equations model. Weights were computed as the product of a missing model weight to attenuate the effect of attrition conditional on strata variables and group allocation [32]. The intervention effects on continuous variables were analyzed using linear mixed-effects model analysis with residual maximum likelihood estimation. In the sensitivity analyses, missing scores due to participants reaching the study endpoint (experiencing a depressive episode) before the 12-month follow-up were handled using last observation carried forward (LOCF) method, as LOCF likely to be conservative when the target condition is expected to improve spontaneously over time (e.g., insomnia and depression) [33]. Mediation analyses using maximum likelihood estimation within a structural equation modeling framework were used to investigate the mediating effects of insomnia on depressive symptoms (the mediating effects were tested when the efficacy analysis showed significant differences between the two groups in depressive symptoms). Strata variables at baseline were included as covariates in all analyses. Data analyses were performed using the Stata 17.0 statistical software with standard two-tailed α of 0.05.

A total of 708 youth (407 females [57%]; mean [SD] age, 22.1 [1.9] years) with insomnia disorder (mean duration, 3.0 years) were allocated to either app-based CBT-I (n = 354) or app-based HE group (n = 354) (Table 1). Majority of the participants (90%; 639/708) were undergraduate students or university educated. Nearly half of the participants (48%; 339/708) had a history of MDD, and 17% (123/708) of them had suicidal ideation in the past month as confirmed by MINI interview, while only few of them had taken antidepressants before (5%; 38/708). In addition, 140 (20%) participants took sleep-promoting medications at baseline. The percentage of participants using sleep-promoting medications decreased in both groups at post-intervention (app-based CBT-I versus app-based HE, 12% versus 15%) and follow-up assessments (12-month, 10% versus 14%) with no significant group differences (Fig A in S1 Appendix).

Of the included participants (n = 708), 634 (90%) contributed data to the post-intervention assessment, 565 (80%) completed the 6-month assessments, and 500 (71%) completed the assessments at 12 months (Fig 1). Attrition rates from assessments at post-intervention and follow-ups did not differ significantly between the two groups. In the app-based CBT-I group, 333 participants (94%) logged on at least one session and 296 (84%) completed all six sessions (Table A in S1 Appendix). As for the app-based HE group, 332 (94%) participants logged on at least one session and 301 (85%) completed six sessions. No adverse events related to the interventions were reported.

Thirty-seven participants (incidence proportion, 10%; 37/354) in the intervention group and 62 participants (incidence proportion, 18%; 62/354) in the control group developed new-onset MDD during the 12-month follow-up. Fig 2 shows the survival curves for the app-based CBT-I and control groups. The estimates of the cumulative 1-year incidence rate (person-time rate) of MDD were 12% (95% confidence interval [CI] 9% to 17%) for the app-based CBT-I group and 21% (17% to 27%) for the HE group. The number needed to treat to prevent MDD at 1 year was 10.9 (6.8–26.6). Cox regression revealed an HR of 0.58 (0.38–0.87; p = 0.008) for MDD favoring app-based CBT-I intervention.

After additionally adjusting for prior history of MDD, educational level, family income, comorbid medical illnesses, and use of sleep-promoting medications in exploratory analysis, the risk of developing MDD remained significantly lower for the app-based CBT-I intervention group compared with the controls (adjusted HR 0.60 [0.40–0.90]; p = 0.01) (Table B in S1 Appendix). Further analyses demonstrated that the preventive effect of app-based CBT-I on MDD across various subgroups was generally consistent with the effect observed in the overall sample (Fig B in S1 Appendix), and there was no interaction between the intervention effect and any of the variables that defined the subgroups (ps > 0.05 for all interactions). In the sensitivity analyses conducted among participants with persistent insomnia (insomnia duration ≥ 3 months, n = 616), all results remained consistent (Table C and Fig C in S1 Appendix).

The app-based CBT-I intervention produced greater improvements than the app-based HE for most of the outcomes at post-intervention (Tables F–G and Fig F in), including SOL, TIB, SE, anxiety symptoms, fatigue, dysfunctional sleep beliefs, and a greater shift toward morningness, which were maintained at least up to the 6-month follow-up, except for anxiety symptoms. There were no significant differences between the two groups in WASO, TST, or suicidal ideation. S1 Appendix