Impact of chosen cutoff on response rate differences between selective serotonin reuptake inhibitors and placebo

Response defined as a 50% reduction in the sum score of the Hamilton Depression Rating Scale (HDRS-17-sum) is often used to assess the efficacy of antidepressants. Critics have, however, argued that dichotomising ratings with a cutoff close to the median may lead to scores clustering on either side, the result being inflation of miniscule drug-placebo differences. Using pooled patient-level data sets from trials of three selective serotonin reuptake inhibitors (SSRIs) (citalopram, paroxetine and sertraline) (n = 7909), and from similar trials of duloxetine (n = 3478), we thus assessed the impact of different cutoffs on response rates. Response criteria were based on (i) HDRS-17-sum, (ii) the sum score of the HDRS-6 subscale (HDRS-6-sum) and (iii) the depressed mood item. The separation between SSRI and placebo with respect to response rates increased when HDRS-17-sum was replaced by HDRS-6-sum or depressed mood as effect parameter and was markedly dependent on SSRI dose. With the exception of extreme cutoff values, differences in response rates were largely similar regardless of where the cutoff was placed, and also not markedly changed by the exclusion of subjects close to the selected cutoff (e.g., ±10%). The observation of similar response rate differences between active drugs and placebo for different cutoffs was corroborated by the analysis of duloxetine data. In conclusion, the suggestion that using a cutoff close to the median when defining response has markedly overestimated the separation between antidepressants and placebo may be discarded.