Research on depression treatment usually reports averages, and the average patient does moderately well with either cognitive behavioural therapy or antidepressant medication. Averages hide the people at each end. Vittengl and colleagues went looking for them.
The researchers pooled individual patient data from 16 randomised trials comparing the two treatments, covering 1,700 adults with depression. Rather than asking how the typical person fared, they counted unusual outcomes at both extremes.
Poor outcomes were uncommon. Between 5% and 7% of patients got worse at all, and only about 1% got reliably worse, meaning a change large enough to be more than measurement noise. Between 4% and 5% finished treatment still severely depressed. At the other end, 6% to 10% improved dramatically, and 4% to 5% finished with essentially no symptoms.
How unwell someone was at the start helped forecast which end they might land at, but it did not indicate which of the two treatments would suit them better. Neither approach protected against a poor result.
Depression treatment works for most people and not for everyone, and knowing that in advance makes it easier to speak up early if progress stalls. If you are weighing up treatment, a psychologist can help you think it through.
What this study set out to see
Trial reports are dominated by group averages, and the average tells you what happened to nobody in particular. A treatment that helps most people moderately can still leave a minority worse off, and averages will not show it. This study asked how often depressed patients end up at either extreme, what predicts it, and whether the choice between cognitive behavioural therapy and medication changes the odds.
The distinction between predictors and moderators matters for what follows. A predictor forecasts outcome regardless of which treatment someone receives. A moderator tells you that one treatment suits a particular kind of patient better than the other. Only a moderator can guide the choice of treatment, which is what a clinician actually needs.
How it was done
The authors drew individual pre-treatment and post-treatment scores from 16 randomised clinical trials comparing cognitive behavioural therapy with pharmacotherapy for unipolar depression, covering 1,700 patients. Symptoms were measured on the Hamilton Depression Rating Scale, completed by a clinician, and the Beck Depression Inventory, completed by the patient.
Five outcomes were defined in advance. Any deterioration meant a symptom increase of at least 1 point. Reliable deterioration required a much larger increase, at least 8 points on the Hamilton scale or 9 on the Beck, a threshold set so that the change exceeds what measurement error alone could produce. Extreme nonresponse meant finishing treatment with a Hamilton score of 21 or above or a Beck score of 31 or above. At the favourable end, superior improvement meant a decrease of at least 95%, and superior response meant finishing with a score of 0. Multilevel models were used, an approach suited to data where patients are grouped within separate trials.
What it found
Extreme outcomes in either direction were infrequent. About 5% to 7% of patients showed any deterioration, and only 1% showed reliable deterioration. Extreme nonresponse occurred in 4% to 5%. Superior improvement occurred in 6% to 10%, and superior response in 4% to 5%.
Two differences between treatments emerged. Superior improvement on the clinician-rated scale only, and attrition, meaning dropping out before the end, were both more frequent in pharmacotherapy than in cognitive behavioural therapy, with an odds ratio of 1.67 for each. An odds ratio above 1 indicates the outcome was more likely in that group.
Pre-treatment symptom level predicted where someone landed, though not in one direction. Patients who deteriorated or achieved a superior response had lower symptom levels before treatment, while those with extreme nonresponse or superior improvement had higher levels. Part of that pattern reflects arithmetic as much as clinical fact: someone starting with mild symptoms has less room to improve and more room to worsen on the scale.
The authors are direct about the practical implication. Pre-treatment symptom levels help forecast unusual outcomes but do not guide selection between the two treatments.
What follows
Two things are worth carrying away. Getting worse during treatment is uncommon and not unheard of, and monitoring symptoms during a course of treatment is how it gets noticed while there is still time to change course. And the hope that a simple baseline measure could tell us who needs therapy rather than medication is not supported here.
These findings come from patients in randomised trials, who are typically screened more tightly than people presenting to an ordinary clinic, so the frequencies may not transfer exactly. Medication decisions rest with a doctor rather than a psychologist, and the two treatments are often combined in practice.
If treatment for depression has not helped as you hoped, or you are trying to decide where to start, a psychologist can help you look at the options.