Therapy, Medication, or Both for Depression: Reading the Comparative Evidence

For most adults with major depression, the strongest comparisons find that psychotherapy on its own and antidepressant medication on its own perform about equally well, and that combining the two tends to do a little better than either alone. That is the headline, but the more useful skill is knowing how to read the numbers behind it, because those numbers are what separate a genuine finding from a confident-sounding overstatement.

Key points#

• Head-to-head trials show no reliable average gap between psychotherapy and antidepressants for adult major depression.
• Combined treatment tends to outperform either single approach, most clearly in more severe, chronic, or recurrent illness.
• The advantage is small-to-moderate: an effect size around 0.43 and a number needed to treat near four in the pooled data.
• Averages describe groups. Preference, side effects, cost, and access all belong in an individual decision.

Start with the measuring stick, not the verdict#

Before asking which treatment wins, it helps to understand how researchers even express "winning." Two tools do most of the work.

The first is the effect size. Rather than reporting a single dramatic result, trials summarize the gap between two groups in units of variability, most often a standardized mean difference called Hedges' g. As a rough guide, 0.2 counts as small, 0.5 as moderate, and 0.8 as large. When a 2014 meta-analysis in World Psychiatry pooled trials that added psychotherapy on top of medication, it landed on a g near 0.43, a small-to-moderate advantage for the combination over medication by itself.

The second is number needed to treat, which turns that abstraction into something you can picture. It answers a plain question: how many people would you treat with the better option for one additional person to benefit who otherwise would not have? The same 2014 analysis put that figure at roughly four. In other words, for about every four patients offered combined care instead of medication alone, one extra person reached a good outcome. Smaller numbers mean a bigger practical effect. Relative measures point the same way; the later network analysis found combined treatment raised the odds of response by about a quarter over either single approach.

Keep both tools in hand and the rest of the evidence becomes much easier to interpret.

How the head-to-head evidence is built#

The cleanest comparison is a randomized trial that assigns patients to one treatment or another and tracks who improves. A 2013 World Psychiatry meta-analysis pooled the direct head-to-head trials of therapy versus medication and found the two broadly comparable, with small differences that hinged on the particular therapy and the particular drug.

The trouble is that few trials pit every option against every other option. Network meta-analysis was designed to stretch further: it stitches together direct comparisons (A versus B) with indirect ones inferred through a shared comparator (A versus C and B versus C) to rank treatments that were rarely, if ever, tested against each other. That reach rests on an assumption called transitivity, meaning the linked trials have to be similar enough in their patients, severity, and outcome definitions for the indirect arithmetic to be fair. When that assumption is strained, the rankings can look more settled than the underlying data justify. It also matters that most analyses define response as a 50 percent drop in symptom scores and remission as falling below a clinical cutoff, so the definition of success shapes what "better" means.

Where combination care pulls ahead#

The largest study on this question, a 2020 network meta-analysis in World Psychiatry drawing on more than one hundred trials and roughly twelve thousand patients, found combined treatment more effective than either psychotherapy or pharmacotherapy alone for achieving response, with a similar pattern for remission. It also weighed acceptability by counting dropouts, and here combined treatment and psychotherapy were better tolerated than medication alone.

The combination tends to show its clearest edge in moderate-to-severe, chronic, or recurrent depression, where added structure and two working mechanisms may reinforce one another. Practice guidance mirrors this. The American Psychological Association's 2019 depression guideline suggests offering either psychotherapy or a second-generation antidepressant as a reasonable starting point, and supports pairing cognitive-behavioral or interpersonal therapy with medication when a combined route is chosen.

Why "one treatment wins" overstates the case#

Several features of this literature should temper any urge to crown a single champion.

Psychotherapy cannot be blinded the way a pill can be tested against a matched placebo, and the comparison group does a lot of hidden work: measuring therapy against a waitlist tends to flatter it relative to an active control. Researcher allegiance, the tendency of investigators to favor the treatment they built or prefer, has been shown to tilt results. Publication bias, small samples, and uneven risk of bias across studies all widen the true uncertainty around any neat point estimate.

Then there is the gap between the average patient and the person actually sitting in a clinic. An effect size near 0.43 and a number needed to treat near four describe a meaningful but modest edge, not a promise, and they say nothing about which particular individual will respond to which approach. Side effects, cost, access, past experience, and personal preference all belong in the decision, and the acceptability data suggest many people stay in treatment longer when a psychological component is part of it.

Read this way, the comparative evidence is reassuring rather than discouraging. It says that people facing depression usually have more than one route with real support behind it, that combining approaches is a sensible and often stronger choice for heavier illness, and that claims of one universal best treatment tend to promise more than the data can deliver.