Clinical response predictors in a double-blind, placebo-controlled trial of fluoxetine for geriatric major depression. Fluoxetine Collaborative Study Group.

BACKGROUND

We attempted to determine baseline characteristics predicting response in a 6-week, double-blind, geriatric depression trial, which showed a significantly higher remission rate for fluoxetine (20 mg daily) than for placebo (31.6% vs. 18.6%, p < .001).

METHODS

Outpatients (N = 671) were 60 years or older (mean +/- SD = 67.7 +/- 5.7), met Diagnostic and Statistical Manual of Mental Disorders (3rd ed., rev., American Psychiatric Association, 1987) criteria for unipolar major depression, had baseline scores on the 17-item Hamilton Depression Rating Scale (HAMD17) of 16 or more, and were randomized after a 1-week placebo lead-in. Potential baseline predictors of percentage change in last-visit-carried-forward HAMD21 total scores were entered into a stepwise regression model. The sample was randomly divided into two groups (development and validation data sets) so that potential predictors could be confirmed in a second analysis.

RESULTS

Of the 266 variables considered for their prognostic ability, 13 were found to be significant predictors using the development data set, including (a) presence of somatic complaints, absence of agitation, and presence of previous accidental injury for fluoxetine response; and (b) reported feelings of emptiness, absence of somatic complaints, and absence of early insomnia for placebo response. The second analysis using the validation data set failed to confirm statistical significance of predictors identified in the development data set.

CONCLUSIONS

Although potentially useful baseline characteristics were initially identified as response predictors, conservative statistical methods failed to confirm any significant predictors of differential responses between fluoxetine and placebo in this double-blind, placebo-controlled trial. These results suggest that response predictor analyses require confirmation before conclusions can be generalized.