Antidepressants Work But Not How You Think

The Most Important Drug Study You Probably Misread

In 2018, a team of researchers led by Andrea Cipriani at the University of Oxford published what remains the largest analysis ever conducted on antidepressant drugs. They combed through 28,552 citations, winnowed them down to 522 clinical trials, and pooled data from 116,477 patients. The study appeared in The Lancet, one of the most prestigious medical journals in the world. It has been cited over 3,400 times.

Here is the finding that made headlines: all 21 antidepressants tested were more effective than placebo. The odds ratios ranged from 2.13 for amitriptyline down to 1.37 for reboxetine (Cipriani et al., 2018). News outlets ran with this. The message was simple: antidepressants work.

But here is the thing nobody told you. That number, 2.13, sounds impressive. It means you are twice as likely to respond to amitriptyline than to a sugar pill. But the real story is not that antidepressants beat placebo. The real story is how small the differences are between the drugs themselves, and what that tells us about what these pills are actually doing.

The Numbers That Never Made the Headlines

The Cipriani meta-analysis measured two things: efficacy (how well a drug reduced depression symptoms) and acceptability (how many patients dropped out of the trial for any reason). For efficacy, they calculated odds ratios comparing each drug to placebo. The best performer, amitriptyline, had an odds ratio of 2.13. The worst, reboxetine, was 1.37.

But here is the part that changes the conversation. When the authors compared drugs against each other directly, the differences were tiny. The odds ratios between antidepressants ranged from just 1.15 to 1.55 (Cipriani et al., 2018). That means the most effective drug was only about 50 percent more likely to work than the least effective one. In practical terms, you could pick almost any antidepressant off the shelf and have roughly the same chance of getting better.

This is not what you hear in the advertising. This is not what your doctor tells you when they write a prescription for the newest, most expensive drug. The data say that after decades of research and billions of dollars, we have a class of drugs where the best and worst are separated by a whisper.

Why Head to Head Trials Tell a Different Story

The Cipriani team did something clever. They separated their analysis into two kinds of evidence: placebo controlled trials and head to head trials where drugs were compared directly. In the placebo controlled trials, all drugs looked roughly similar. But in the head to head trials, variability appeared.

Some drugs stood out. Agomelatine, amitriptyline, escitalopram, mirtazapine, paroxetine, venlafaxine, and vortioxetine were more effective than other antidepressants, with odds ratios ranging from 1.19 to 1.96 (Cipriani et al., 2018). Meanwhile, fluoxetine, fluvoxamine, reboxetine, and trazodone were the least effective, with odds ratios between 0.51 and 0.84.

But here is the catch. When the authors included all trials, including the placebo controlled ones, those differences shrank. The head to head trials are where the signal is strongest, but they are also the trials most likely to be biased. Drug companies fund most of these studies, and they have a financial interest in showing their drug is better than a competitor. The Cipriani analysis accounted for this, but the uncertainty remains.

The Acceptability Paradox

Here is where things get interesting. The study also measured acceptability, meaning how many people stayed on the drug. Only two drugs, agomelatine and fluoxetine, had fewer dropouts than placebo (Cipriani et al., 2018). For agomelatine, the odds ratio was 0.84. For fluoxetine, it was 0.88. That means people on these drugs were slightly less likely to quit the trial than people on sugar pills.

But clomipramine was worse than placebo, with an odds ratio of 1.30. More people dropped out of the clomipramine group than the placebo group. That is a drug that makes people feel worse, not better.

The paradox is this: the drugs that work best for efficacy are often the ones with the worst side effects. Amitriptyline, the most effective drug in the analysis, is an older tricyclic antidepressant with significant side effects including sedation, weight gain, and cardiac issues. It is rarely prescribed as a first line treatment anymore. The newer drugs, the SSRIs and SNRIs, are less effective but better tolerated.

This is the trade off that never makes it into the marketing. You can have a drug that works better, but you might not want to take it. Or you can take a drug that is easier to handle, but it might not work as well. There is no free lunch.

What This Study Actually Proves

The Cipriani analysis is a meta analysis, not a single experiment. Meta analyses are powerful because they combine data from many studies, increasing statistical power. But they also inherit the weaknesses of the studies they include. Of the 522 trials, only 96 (18 percent) were rated as low risk of bias. The majority, 380 trials (73 percent), were rated as moderate risk. And 46 trials (9 percent) were high risk (Cipriani et al., 2018).

The certainty of evidence ranged from moderate to very low. That is not a criticism of the authors. That is the reality of antidepressant research. Many trials are small, poorly designed, or funded by companies with a vested interest in positive results. The Cipriani team did their best to account for these problems, but the underlying data is messy.

What the study does prove is that antidepressants are not placebos. They have a real, measurable effect. But that effect is modest. The difference between the best drug and the worst drug is small. And the difference between any drug and placebo, while statistically significant, is not as large as most people assume.

The Question Nobody Is Asking

If all antidepressants work about the same, and the differences between them are tiny, then why do some people respond dramatically to one drug and not at all to another?

The Cipriani analysis cannot answer this question. It looked at group averages, not individual responses. And group averages hide a lot. In clinical practice, patients often try several drugs before finding one that works. Some people get better on the first try. Others cycle through five or six before giving up.

The data say that, on average, antidepressants work. But averages do not tell you what will happen to you. You might be the person for whom amitriptyline is a miracle. Or you might be the person who gets every side effect in the book and no benefit.

This is the gap between population level research and individual treatment. The Cipriani study is the best evidence we have for population level effects. But it cannot predict individual outcomes. That remains a mystery.

What This Actually Means

• ▸All antidepressants work, but none work dramatically well. The odds ratio for the best drug was 2.13. That means you are about twice as likely to respond to amitriptyline as to placebo. For the worst drug, reboxetine, the odds ratio was 1.37. That is a 37 percent improvement over placebo. Both are real effects, but neither is a cure. If you are considering antidepressants, set realistic expectations. They might help. They might not. The data say they are better than nothing, but not by a landslide.

• ▸The differences between drugs are small enough that your doctor should pick based on side effects, not efficacy. The odds ratios between antidepressants ranged from 1.15 to 1.55. That is a narrow window. In practice, the best drug for you is the one you can tolerate. If fluoxetine gives you fewer side effects than escitalopram, take fluoxetine. The efficacy difference is not large enough to matter.

• ▸Older drugs like amitriptyline are more effective but harder to tolerate. The most effective drug in the analysis was amitriptyline, a tricyclic antidepressant from the 1950s. It works. But it also causes sedation, weight gain, and cardiac problems. Newer drugs like escitalopram and sertraline are less effective but better tolerated. If you have tried several newer drugs without success, an older drug might be worth considering under close medical supervision.

• ▸Fluoxetine (Prozac) is the best tolerated drug, not the most effective. Fluoxetine had an acceptability odds ratio of 0.88, meaning fewer people dropped out compared to placebo. But it was among the least effective drugs in head to head trials. This is the trade off. Fluoxetine is easy to take. It just might not work as well.

• ▸The research is not as clean as you think. Only 18 percent of the trials were rated low risk of bias. The rest had moderate or high risk. The certainty of evidence was moderate to very low. This does not mean the results are wrong. It means they are provisional. Future research could change the picture. Treat these findings as the best available evidence, not the final word.

The Cipriani study is a landmark. It tells us that antidepressants work, but the story is more complicated than the headlines suggested. They work modestly. They work differently for different people. And the best drug for you might not be the one that works best on average. It might be the one you can actually take.