Exploring dropout in internet-delivered cognitive behavioral therapy for insomnia: A secondary analysis of prevalence, self-reported reasons, and baseline and intervention data as predictors.

Introduction: Internet-delivered cognitive behavioral therapy for insomnia (iCBT-I) is an effective treatment. However, dropout is a common challenge in digital therapeutics. This study examines dropout in iCBT-I by analyzing reported reasons for dropout and investigating whether baseline variables and intervention usage data can predict dropout. Methods: This is an exploratory secondary analysis of a clinical trial investigating a stepped care model for insomnia featuring an eight-module iCBT-I. Reasons for dropout from the iCBT-I were assessed via self-developed items in follow-up surveys, and a dropout survey was sent to all patients who had not completed at least seven modules of the iCBT-I within 12 weeks. The proportion of respondents who agreed with the respective items was calculated. Additionally, bivariate models were specified to explore whether baseline variables and intervention usage data can predict dropout. Results: The patients included in this sub-study had a mean age of 49.3 (SD=13.0), with 73.4 % identifying as female. At pre-treatment, their mean insomnia severity was 18.6 (SD=3.9). Among the 233 patients, 103 (44.2 %) were categorized as dropouts. The most frequently reported reasons for dropout were distractions from daily life, the perception of the content not being useful, and difficulties resuming after a break. None of the examined baseline variables significantly predicted dropout, whereas the time needed to complete the first module (OR=1.16; 95 %CI=1.08-1.27) and the number of sleep diary entries in the first week (OR=0.73; 95 %CI=0.65-0.80) significantly predicted dropout. Discussion: This study highlights dropout as a relevant challenge in iCBT-I, affecting over 40 % of patients. Self-reported reasons indicate the importance of compatibility with distractions from daily life and perceived effectiveness. The prediction models suggest that dropout risk profiles can be developed based on first-week treatment data. Future research should focus on validating such models to improve effectiveness and user retention.