A major analysis of internet-based depression treatments has identified who is most likely to stop participating before completing the program, offering new clues about why digital mental-health care often loses users along the way. The study found that younger adults, men, people with lower levels of education and employed participants tended to complete a smaller proportion of treatment modules than other users. The results suggest that improving online therapy may depend not only on making interventions clinically effective, but also on designing them around the real-world needs of people most at risk of disengaging.
Published in Nature Mental Health, the research is the largest individual participant data meta-analysis to examine predictors of adherence to internet-based interventions for depression. Rather than relying only on summary statistics reported by individual studies, the researchers combined participant-level data from 71 randomized controlled trials. The analysis covered 85 treatment arms and 8,082 adults with elevated depressive symptoms, creating a detailed dataset capable of revealing patterns that may be difficult to detect in smaller studies.
The investigators searched PubMed, Embase and PsycINFO for eligible trials through 6 February 2024. The interventions included digitally delivered psychological programs, ranging from guided treatments supported by a therapist or coach to self-guided programs completed independently. Most internet-based psychological interventions are structured into a series of modules, each addressing skills such as behavioral activation, cognitive restructuring, problem-solving or relapse prevention. In the study, adherence was defined as the proportion of modules participants completed after entering the intervention.
To analyze this proportion, the researchers used a one-stage multilevel beta regression with a logit link. This statistical approach is designed for outcomes that fall between zero and one, such as the fraction of completed modules. The multilevel structure accounted for the fact that participants were clustered within treatment arms and trials, while the logit link allowed the model to represent changes in adherence across its full proportional range. By analyzing all trials in one model, the researchers could estimate the influence of individual characteristics while taking differences between studies into account.
Age emerged as one of the significant predictors. The positive coefficient for age, β = 0.005 with a standard error of 0.002, indicated that adherence increased modestly as participant age increased. The association reached statistical significance, with P = 0.028. Although the effect for each additional year was small, such differences can become meaningful across a broad age range. Younger adults may face more competing demands, use digital platforms in less structured ways or perceive an online program as less relevant if it does not match their expectations for immediate, flexible support.
Gender was also associated with participation. Male participants completed fewer modules on average than female participants, reflected by a coefficient of β = −0.163, a standard error of 0.053 and P = 0.002. The researchers did not interpret this finding as evidence that gender itself causes disengagement. Instead, it may reflect differences in help-seeking behavior, attitudes toward psychological treatment, symptom presentation, digital communication preferences or practical barriers. The result points to the need for more research into how online interventions can be made appealing and accessible to people who are traditionally less likely to engage with mental-health services.
Education and employment status showed additional relationships with adherence. Participants with lower educational attainment had lower completion rates, with β = −0.133, standard error 0.05 and P = 0.008. Employment was also associated with reduced adherence, β = −0.113, standard error 0.054 and P = 0.037. These patterns may reflect the cognitive and practical demands of completing structured digital therapy. People balancing work, family responsibilities or irregular schedules may struggle to reserve time for modules, while written therapeutic materials may be less accessible when they require substantial reading, concentration or familiarity with psychological terminology.
The researchers found no significant interactions between the individual predictors and intervention format. In other words, the relationship between characteristics such as age, gender, education or employment and adherence did not clearly differ between guided and self-guided programs. This finding challenges the assumption that simply adding human support will eliminate demographic gaps in engagement. Guidance can be valuable, but the most effective strategy may involve combining support with shorter modules, reminders, adaptive pacing, clearer language, mobile-friendly design and rapid responses when users begin to disengage.
The analysis also connected adherence with clinical outcomes. Participants who completed a greater proportion of modules tended to have lower depression severity after treatment, even after the researchers adjusted for baseline depression severity. The association was substantial, with β = −0.30, standard error 0.04 and P < 0.001. This does not prove that completing more modules directly caused symptoms to improve; people who begin feeling better may be more motivated to continue, and other factors may influence both adherence and recovery. Nevertheless, the finding reinforces the clinical importance of engagement. A digital intervention cannot deliver its full therapeutic content if users disappear after the first few sessions.
The study’s authors argue that identifying people at risk of low adherence could help developers and clinicians provide targeted support before disengagement occurs. Digital programs could use early participation patterns, such as missed modules or delayed log-ins, to trigger personalized reminders or optional human contact. However, risk prediction must be used carefully and should not label people as unlikely to benefit. The broader message is that adherence is not merely a user problem: it is a design and implementation challenge. As internet-based depression care expands, the programs most likely to make a lasting impact may be those that combine evidence-based therapy with a precise understanding of how different people actually live, work and seek help online.
Subject of Research: Predictors of adherence to internet-based psychological interventions for depression in adults
Article Title: An individual participant data meta-analysis of predictors of adherence to internet-based interventions for depression
Article References: Tong, L., Panagiotopoulou, O. M., Miguel, C. et al. “An individual participant data meta-analysis of predictors of adherence to internet-based interventions for depression.” Nature Mental Health (2026). https://doi.org/10.1038/s44220-026-00707-4
Image Credits: AI Generated
DOI: https://doi.org/10.1038/s44220-026-00707-4
Keywords: Depression, internet-based psychological interventions, treatment adherence, digital mental health, online therapy, individual participant data meta-analysis, depression symptoms, behavioral health, treatment engagement
