This evidence review separates the outcome being measured from the claim being made. The distinction matters because an improvement on a questionnaire, an observed change in behavior and a reduction in mortality require different evidence.
What the two reviews actually measured
Oh and colleagues’ 2024 meta-analysis included nine randomized trials and 1,779 participants. Its suicidal-ideation analysis reported Hedges’ g of 0.478, while the depression estimate was 0.316. These are standardized differences between groups, not percentages of patients cured or deaths prevented.
The ideation results were highly heterogeneous, with I² around 88%. The review assessed immediate post-intervention effects and acknowledged gaps concerning suicide plans, attempts and longer-term outcomes. It therefore supports a qualified symptom finding, not a general claim that digital psychotherapy prevents suicide deaths.
Büscher and colleagues’ individual-participant-data meta-analysis obtained data from nine of ten eligible trials, covering 2,037 participants. Internet-based cognitive behavioral therapy was associated with reduced suicidal ideation. Reliable improvement occurred in 40.5% of the intervention group and 27.3% of controls; reliable deterioration occurred in 2.8% and 5.1%, respectively.
Those percentages describe changes under the study definitions. They are not recovery rates or evidence that a particular person is no longer at risk. The authors identified missing outcome data, risk of bias and limited evidence about suicidal behavior as important constraints. An absence of a detected difference between participant subgroups would also not prove equal suitability for everyone.
Why an RCT does not establish safety for every user
A randomized trial compares an intervention with a specified control in an enrolled population. Its result cannot automatically be transferred to a different app, an updated algorithm, an unsupported self-help version or someone excluded from the trial. Evidence for structured CBT also cannot be borrowed by an unrelated generative chatbot merely because both operate on a phone.
Before treating a product’s research claim as meaningful, identify the exact intervention, eligibility criteria, comparison group, follow-up period and how harms and missing outcomes were handled. “Tested in an RCT” is the beginning of that check, not its conclusion.
Support is part of the service being evaluated
Delivery arrangements matter. In its assessment of digitally enabled therapies for adult depression, NICE’s committee discussion distinguishes supported therapy from simply providing access to software. It notes that people at high risk or with difficulty engaging digitally may be better suited to approaches with more therapist involvement.
This is why a product description should say who reviews progress, what happens when symptoms worsen and whether messages are actually monitored. An automated risk notification is not evidence that a qualified person will respond promptly. A scheduled program should not be presented as emergency support.
A narrower and more useful conclusion
The evidence supports investigating specific digital therapies within defined care arrangements. It does not support ranking all apps as effective, treating reduced ideation as reduced mortality, or recommending that readers replace existing treatment on the strength of these pooled results.
For the separate question of what distinguishes a regulated treatment from a general wellness product, see our comparison of digital therapeutics and medicines. That product-level distinction must come before any claim about substitution.
