Reading focus
Syllabus: How Do We Know When Psychotherapy Helps—or Harms?
Compare deterioration risks, measurement rules and follow-up uncertainty; treat the subgroup analyses cautiously.
Why you’re reading it. Apply reliable-change reasoning to a specific adverse outcome without mistaking it for a complete safety assessment.
A question to consider. How would you revise the claim “this treatment does not cause harm” if its evidence measured only short-term worsening on depression questionnaires?
Why this reading
Average improvement can coexist with worsening for some participants. This analysis uses individual data from randomized trials to compare reliable symptom deterioration in guided online self-help and control groups. It makes a specific adverse outcome measurable without capturing every possible harm.
Before you read
Individual participant data allow a common change rule to be applied across studies. A risk ratio compares the frequency of an outcome between groups; absolute risks describe how often it occurs. Imputation estimates missing values under assumptions rather than observing them.
Read for
- Trace the definition of deterioration, missing-data assumptions and alternative change criterion.
- Compare the overall post-treatment result with the smaller follow-up samples and exploratory subgroup analyses.
Consider afterwards
- How could a treatment reduce symptom deterioration relative to a control while still producing unwanted effects for some people?
- What extra measurements and follow-up would be needed to investigate harms beyond worsening depression scores?
Editorial context
The analysis includes 18 trials of guided self-help for adult depression; it does not assess all online therapies or all harms. Follow-up evidence is limited, and the education finding is exploratory, not a validated rule for excluding people from care. Some confidence intervals in the printed Table 3 are internally inconsistent, including a negative limit for a risk ratio; do not infer corrected values. The overall post-treatment estimate is consistent across the abstract and table.