Antidepressant trials succeeded 94% of the time in journals and 51% of the time in FDA files

Key takeaways

• When researchers compared FDA records on 74 antidepressant trials with the medical journals, 94% of trials looked positive in print, but only 51% were positive in the FDA’s files.
• In a check of 67 trials in five top medical journals, 58 reported outcomes that didn’t match the ones they had planned at the start.
• Across 52 studies of citation habits, papers with statistically significant results were cited 1.6 times as often as papers without them.
• A 2020 study in The Lancet found that only 40.9% of U.S. trials covered by a federal reporting law posted results on time.
• You can check whether a claim was cherry-picked in about ten minutes by finding an independent summary, the trial registration and the studies that disagreed.

Erick Turner spent three years reviewing drug studies at the Food and Drug Administration (FDA)². The agency reviews the trials companies run to win approval, including ones that never reach a journal¹². Doctors reading medical journals see only what gets published.

In 2008, Turner and his colleagues lined up the FDA’s records on 12 antidepressants against what had been published. The FDA had reviewed 74 trials. Nearly a third of them, covering 3,449 people, had never been published at all¹.

Read the journals, and 94% of the trials looked positive, meaning the drug beat a placebo, a dummy pill. In the FDA’s files, the real figure was 51%¹.

No published study had to be fake for that gap to open. Good news got printed. Much of the bad news stayed in a drawer.

That’s cherry-picking: choosing the evidence that supports a conclusion and leaving the rest out. It shapes drug research, supplement research and everything in between. This guide shows how it works and how to check whether “studies show” means what it seems to.

Three ways a pile of studies gets stacked

Cherry-picking rarely looks like fraud. It usually takes one of three forms.

• Publication bias. Studies with exciting results are more likely to be published than studies that find nothing. Researchers may not write up a flop, and journals may not want it.
• Outcome switching. A study plans to measure one thing, then reports something else that looked better. Researchers also call this selective outcome reporting.
• Selective citation. An article, review or ad mentions the studies that agree with it and skips the ones that don’t.

Each step can happen without anyone telling a lie. Stacked together, they can make a weak treatment look strong. The fix is the same for all three: look at everything that was done, not just what was shown to you.

The antidepressant files show how much hiding changes the picture

Here is what Turner’s team found when it sorted the 74 trials¹.

The FDA judged 38 trials positive. Of those, 37 were published.

The FDA judged the other 36 negative or questionable. Of those, 22 were never published, and 11 were published in a way that, in Turner’s team’s judgment, made them look positive. Only 3 were published as the failures they were¹.

The distortion was large. Compared with the FDA’s numbers, the journal articles made the drugs look 32% more effective overall, and 11% to 69% more effective for individual drugs¹.

The fair reading matters here, too. When all the data were pooled, each drug still beat the placebo³. The lesson is not that antidepressants don’t work. It’s that the published record made them look better than they were, and anyone weighing a treatment deserves the real size of the benefit.

If you take an antidepressant, don’t stop on your own because of this. Bring your questions to the clinician who prescribes it.

Natural-health research has the same blind spot

Many people who value natural health distrust drug-company research, and the antidepressant story shows why that instinct has a basis. But the same pattern can show up in research on traditional and natural therapies.

In 1998, a research team led by A.J. Vickers examined the published abstracts of 252 controlled trials of acupuncture. Every one from China, Japan, Hong Kong and Taiwan came out positive. In Russia and the former Soviet Union, 10 of 11 did⁴.

The team saw a similar pattern for other treatments. In England, three in four trials favored the new treatment over the comparison. The figures were 99% for China, 89% for Japan, 97% for Russia and the former Soviet Union, and 95% for Taiwan. Not one trial from China or Russia concluded that the treatment being tested had failed⁴.

The authors named publication bias as a possible explanation⁴. No field of medicine produces only winners. A record with no failures in it is a reason for a closer look, not for confidence.

This doesn’t show that acupuncture fails, and the data are nearly 30 years old. It shows that a pile of uniformly positive studies can’t settle the question by itself. Readers who care about natural remedies have the most to lose from a record like that, because it hides which remedies work.

When the goalposts move after the game

A well-run trial names its main goal before it starts. Picture a trial that plans to measure pain, finds no change, then features a better result on sleep. Nobody faked the sleep number. But it was not the question the trial set out to answer.

How often does this happen? A team led by researcher Ben Goldacre, called COMPare, checked 67 trials published in five of the world’s top medical journals over six weeks in late 2015. In 58 of them, the outcomes reported didn’t match the outcomes planned⁵.

On average, the trials correctly reported 76% of their planned main outcomes and 55% of their planned secondary ones. They also slipped in 365 new outcomes without saying so, about five per trial⁵.

The team wrote 58 letters asking the journals to correct the record. Only 23 were published. The New England Journal of Medicine and JAMA published none⁵⁶. Just 8 research teams publicly acknowledged a specific discrepancy, and one trial was corrected⁶.

Why does this matter so much? Measure enough things and one will look good by luck. The Federal Trade Commission (FTC), which polices health advertising, makes the same point in its guidance. In one of its examples, a study found one significant result among eight measures, and the agency says that single win may be chance and doesn’t prove a depression claim⁹.

The FTC also says analyses dreamed up after a study ends, which stray from the original plan, generally don’t count as proof⁹. They can suggest what to study next. That’s all.

Ads can pull a similar trick with a single study’s numbers. In 2009, Kellogg settled FTC charges over ads saying Frosted Mini-Wheats were “clinically shown to improve kids’ attentiveness by nearly 20%”¹⁰. According to the FTC, kids in Kellogg’s own study who ate the cereal averaged just under 11% better than kids who skipped breakfast¹⁰.

Only about half of them improved at all, and only about one in nine improved by 20% or more¹⁰. The study was real. The ad featured its most flattering slice. The settlement barred Kellogg from misrepresenting any tests or studies¹⁰.

Friendly citations make one side look bigger

Even when every study is published, writers choose which ones to mention. Researchers call the habit of favoring supportive papers citation bias.

A 2017 review pooled 52 studies of how scientists cite each other’s work. Papers with statistically significant results were cited 1.6 times as often as papers without them. Papers whose authors said their idea had been supported were cited 2.7 times as often⁷.

The gaps can be striking. A study of 1,523 trial reports spanning four decades, 1963 to 2004, showed that authors typically cited fewer than a quarter of the earlier trials on their question⁸. Another looked at 38 reviews that recommended physical measures against house dust mites. Of those, 27 leaned on supportive studies in their references, and just one leaned the other way⁸.

Copenhagen-based researcher Peter Gøtzsche argues that the worst cases of citation bias cross the line into scientific misconduct⁸. That’s his view, and not every scientist would go that far. But he cites a 2016 survey in which selective citation ranked as the most common kind of research misbehavior⁸.

Marketing takes citation bias further, and federal regulators have a rule for it. The FTC’s guidance says marketers shouldn’t focus only on research that supports an effect while waving away research that doesn’t⁹. One of its examples describes two short, six-week studies that found modest fat loss and an equally careful 12-week study that found none. A claim of substantial fat loss, the FTC says, isn’t backed up⁹.

Registration was supposed to fix this, and it helps, with gaps

The main fix is trial registration: posting a study’s plan, including its main outcome, in a public database before anyone enrolls. That record is how COMPare could tell what each trial had promised to measure.

The International Committee of Medical Journal Editors (ICMJE), whose rules many leading journals follow, says trials should be registered at or before the first participant enrolls. Its definition covers trials of diet changes and behavior programs, not just drugs and devices¹¹. Accepted registries include ClinicalTrials.gov and those recognized by the World Health Organization¹¹.

U.S. law also requires sponsors of many trials to post results on ClinicalTrials.gov, generally within a year after the trial’s main data are collected¹². Compliance has lagged. A 2020 study in The Lancet looked at 4,209 trials that were due to report. Only 40.9% posted results on time, and 63.8% had posted at any point¹².

This is a twist that cuts against easy assumptions. Industry sponsors were more likely to report on time than sponsors outside both industry and the U.S. government¹². Hiding results isn’t only a corporate habit.

Enforcement has been light. As of February 2026, the FDA listed eight formal notices of noncompliance since 2021, and its table showed no money penalties¹³. Registration is still the best tool you have, because you can use it yourself.

The cherry-pick detector

Run through these questions whenever someone says “studies show.” The more boxes you check, the more likely you’re seeing a hand-picked slice.

• Only one or two studies are named. Ask how many studies exist on the question in total.
• Nothing that disagrees is mentioned. Real research almost always includes some null or negative results.
• Every result is positive. A perfect record is a warning sign, not a selling point.
• The study is small, short or old, and newer, bigger ones exist. The FTC’s own fat-loss example turns on a longer study overriding shorter ones⁹.
• The headline result isn’t the main outcome. Look for words like “secondary,” “subgroup” or “exploratory.”
• Many things were measured and one improved. That pattern often reflects chance⁹.
• There’s no registration number. Ask for the trial’s ID from a registry such as ClinicalTrials.gov.
• The registration lists a different main outcome than the paper features. That’s a moved goalpost.
• The source sells the product. A seller’s research summary deserves a second, independent look.
• An independent summary disagrees. When an NIH page and an ad tell different stories, trust the one with nothing to sell.

Check the whole basket in ten minutes

You don’t need to read every study. You need to find people who already did, and then spot-check the claim.

1. Start with an independent summary (3 minutes). Look up the herb or treatment on the website of NIH’s National Center for Complementary and Integrative Health (NCCIH), on the NIH Office of Dietary Supplements fact sheets, or in a Cochrane review. Note the date each page was last updated.
2. Search the research yourself (3 minutes). On PubMed, the free federal database of medical studies, add a second search term to narrow results. NCCIH’s guide shows a search for echinacea dropping from 1,325 results to 175 when “colds” was added¹⁵. Then use the filters for randomized controlled trials and recent years¹⁵.
3. Find studies NIH paid for (1 minute). Typing “NCCIH [gr]” plus a topic in PubMed pulls up studies funded by the center¹⁵. Those were funded by NIH rather than by a company selling the product.
4. Check the registration (2 minutes). Search the product or trial name on ClinicalTrials.gov. Compare the main outcome listed there with what the ad or paper highlights.
5. Ask what’s missing (1 minute). If the summary and the claim disagree, the claim probably left something out.

This is how it works in practice. Picture a website that says ginkgo protects memory and cites one small, positive study. That study may be real.

NCCIH’s ginkgo page tells the rest of the story. In a federally funded trial called GEM, more than 3,000 adults age 75 or older took ginkgo or a placebo for a median of six years. Their dementia rates did not differ¹⁴. NCCIH says three large studies in older adults found that ginkgo didn’t lower dementia risk¹⁴.

There may be a modest effect on symptoms of existing dementia, especially at higher doses, but the results are inconsistent¹⁴. Ginkgo may also raise bleeding risk in people who take blood thinners such as warfarin, so check with a clinician first¹⁴. The small study didn’t lie. It just wasn’t the whole basket.

What the evidence says about cherry-picking

Claim	Rating	What’s behind it
Unpublished trials made antidepressants look more effective than they were	Strong evidence	FDA records on 74 trials compared with journals1
Studies with positive results get cited more often	Strong evidence	Review of 52 studies of citation habits7
Trials in top journals often don’t report the outcomes they planned	Strong evidence	58 of 67 trials checked by COMPare5
Registration and reporting laws have ended hidden results	Evidence says no	40.9% on-time reporting; few enforcement actions1213
Ginkgo prevents dementia in older adults	Evidence says no	GEM trial and other large studies14
Ginkgo eases symptoms of existing dementia	Promising but early	Modest, inconsistent results14

Questions readers ask

The bottom line

Cherry-picked evidence is persuasive because each piece can be real. The distortion comes from what’s missing: the trial that never got published, the outcome that changed, the study nobody cited.

The antidepressant files and the acupuncture record show that no field is immune. Your best defense is a habit, not a degree. Before you trust “studies show,” find an independent summary, check the registration and ask what the full basket of evidence looks like.

Related reading

• How systematic reviews work, and why cranberries failed one and passed the next
• Who paid for the study matters, even when the science is real
• Most health products aren’t scams, but many just haven’t been proven
• Only 1 in 20 treatments tested in animals wins approval for people
• Industry payments to your doctor are public, and you can check other experts too

Sources

1. Turner EH, et al. “Selective Publication of Antidepressant Trials and Its Influence on Apparent Efficacy.” New England Journal of Medicine, 2008 (abstract as reproduced by Healthy Skepticism). pubmed.ncbi.nlm.nih.gov
2. Psychiatric Times. “Study Faults Selective Publication of Antidepressant Trials” (interview with Erick Turner). 2008. psychiatrictimes.com
3. Medical News Today. Report on Turner et al., “Selective Publication of Antidepressant Trials.” January 2008. medicalnewstoday.com
4. Vickers A, Goyal N, Harland R, Rees R. “Do certain countries produce only positive results? A systematic review of controlled trials.” Controlled Clinical Trials, 1998 (abstract, Cochrane Colloquium). abstracts.cochrane.org
5. Goldacre B, et al. “COMPare: a prospective cohort study correcting and monitoring 58 misreported trials in real time.” Trials, 2019. link.springer.com
6. Goldacre B, et al. “COMPare: Qualitative analysis of researchers’ responses to critical correspondence on a cohort of 58 misreported trials.” Trials, 2019;20:124. link.springer.com
7. Duyx B, Urlings MJE, Swaen GMH, Bouter LM, Zeegers MP. “Scientific citations favor positive results: a systematic review and meta-analysis.” Journal of Clinical Epidemiology, 2017. research.vu.nl
8. Gøtzsche PC. “Citation bias: questionable research practice or scientific misconduct?” James Lind Library, 2021. jameslindlibrary.org
9. Federal Trade Commission. “Health Products Compliance Guidance.” FTC, 2022. ftc.gov
10. Federal Trade Commission. “Kellogg Settles FTC Charges That Ads for Frosted Mini-Wheats Were False.” Press release, April 20, 2009. ftc.gov
11. International Committee of Medical Journal Editors. “Clinical Trial Registration.” ICMJE, accessed 2026. icmje.org
12. DeVito NJ, Bacon S, Goldacre B. “Compliance with legal requirement to report clinical trial results on ClinicalTrials.gov: a cohort study.” The Lancet, 2020 (summary, Bennett Institute, University of Oxford). bennett.ox.ac.uk
13. U.S. Food and Drug Administration. “ClinicalTrials.gov – Notices of Noncompliance and Civil Money Penalty Actions.” FDA, content current as of February 5, 2026. fda.gov
14. National Center for Complementary and Integrative Health. “Ginkgo.” NIH, updated February 2025. nccih.nih.gov
15. National Center for Complementary and Integrative Health. “How To Find Information About Complementary Health Approaches on PubMed.” NIH, updated August 2021. nccih.nih.gov