Drop-out rates in placebo-controlled trials of antidepressant drugs: A systematic review and meta-analysis based on clinical study reports.

Objective: To study the drop-out rates in trials of selective serotonin and serotonin-norepinephrine reuptake inhibitors (SSRIs and SNRIs).

Methods: This study is a systematic review and meta-analysis of trials. The main outcome measure: Overall drop-out rate. Secondary outcomes were drop-outs due to adverse events and lack of effect. We obtained clinical study reports (CSRs) of five antidepressant drugs from the European Medicines Agency and the UK's Medicines and Healthcare products Regulatory Agency. The eligibility criteria for selecting studies: double-blind randomised, placebo-controlled trials for any indication.

Data Extraction And Analysis: The primary outcome was extracted by two researchers independently and meta-analysed using the Mantel-Haenszel method (fixed effect model). The secondary outcomes were extracted by one researcher and checked by another. Sensitivity analyses were performed using Peto's odds ratio and beta binomial methods, due to presence of null events, and by excluding unreliable trials.

Results: We included 71 CSRs (67,319 pages) with information on 73 trials (11,057 patients on SSRI or SNRI drugs, and 7,369 on placebo). There were minor discrepancies within the CSRs when a modified intention to treat principle was used and patients lost to follow up early in the trial were not accounted for. Significantly more patients dropped out on active drug than on placebo, risk ratio 1.08 (95% CI 1.03 to 1.13), with no difference between adults and children/ adolescents, RR = 1.08 (1.03 to 1.13) and 1.07 (0.95 to 1.21), respectively. When three trials with a prior single-blind phase on active drug were removed, the difference was a risk ratio of 1.12 (1.07 to 1.18), whereas the result was the same after removal of three trials with fraudulent data or other issues with data validity, risk ratio 1.08 (1.03 to 1.13). There were more drop-outs due to adverse events on active drug than on placebo, risk ratio 2.63 (2.33 to 2.96). There were fewer drop-outs due to lack of effect, risk ratio 0.47 (0.43 to 0.53). However, this result is biased; when more people drop out due to adverse effects, fewer can drop out because of lack of effect.

Conclusions: By using CSRs, we were able to demonstrate for the first time that more patients dropped out on active drug than on placebo. As it can be argued that the drop-out rate reflects the patients' overall assessment of the balance between benefits and harms, our review adds to the growing concern that SSRIs and SNRIs might not have the desired effect. Our review also highlights the importance of using CSRs for undertaking reviews of drugs.