Publication Bias in Clinical Research: The Problem of Unreported Results
Imagine a company runs a clinical trial on a new drug. Thousands of patients participate. Millions of dollars are spent. The trial finishes, and the results are never published. Did the trial really happen? To the scientific community, effectively no. Unreported results cannot inform clinical decisions, guide future research, or add to medical knowledge. And the uncomfortable truth, as the Clarity Clinical Solutions video on publication bias explains, is that roughly half of all completed clinical trials are never published at all.
The three routes results take
Trial results reach the public through three channels. The first is peer-reviewed journals: the New England Journal of Medicine, The Lancet, JAMA, and others. This is the traditional route, but publication can take one to three years from trial completion, and journals favor novel, positive findings. The second route is results registries. ClinicalTrials.gov, run by the U.S. National Library of Medicine, is a government-mandated platform where sponsors must register trials and submit summary results (ClinicalTrials.gov: FDAAA 801 requirements). The third route is conference presentations and preprints: fast, but without the rigor of full peer review.
What publication bias looks like
Publication bias is the systematic tendency for positive results to be published more often, faster, and in higher-profile journals than negative or null results. The video puts the numbers plainly. Positive results are about three times more likely to be published than negative ones. They are published faster, typically within two to four years compared with five to seven years for negative results. They land in higher-impact journals, get more media coverage, and are cited more often. Negative results get consigned to what researchers call the file drawer.
The scale of the problem is striking. Around 50% of completed trials remain unpublished. A landmark BMJ study found that only 41% of trials registered on ClinicalTrials.gov had published results within four years of completion, and even among large, well-funded phase 3 trials, 29% were still unpublished years after finishing. Those figures come from the video; the underlying pattern is well documented. The estimated cost of this waste: about $85 billion a year in biomedical research that is incomplete, unreported, or redundant.
Why results go missing
Three forces drive the problem. First, sponsor disinterest. When a drug fails, the sponsor has no commercial reason to publish the failure, and may prefer to keep the data confidential so competitors cannot learn from it. Second, journal bias. Even researchers who try to publish null findings face rejection rates far above those for positive work. Third, resource constraints. Publishing takes time, money, and expertise, and open-access fees can run thousands of dollars per paper.
The law that tried to fix it
The most important response came with the FDA Amendments Act of 2007. It required most trials to register on ClinicalTrials.gov before enrolling patients and to submit summary results within 12 months of completion. For the first time, the public could see which trials were planned and what they found. The law's weakness is enforcement: penalties for non-compliance are almost never applied. As of 2023, the NIH had issued zero fines for missing results despite thousands of violations. Registration compliance is high, above 90%, but results reporting compliance hovers around 60 to 70%. The NIH's current policy spells out the reporting expectations for NIH-funded trials (NIH policy on clinical trial results reporting).
Outcome switching and other games
Even when results are published, the endpoint problem remains. Outcome switching happens when the outcome registered as primary before the trial began differs from the outcome reported as primary in the published paper. Studies consistently show that 40 to 60% of published trials have such discrepancies. The tactics are familiar: swapping primary and secondary outcomes, dropping unfavorable endpoints, adding new endpoints after the data is in, and changing how or when outcomes are measured. Most of these changes are never disclosed.
Selective reporting goes further. Cherry-picking means measuring twenty outcomes and reporting only the one or two that show significance. Subgroup fishing means slicing the data by age, gender, or disease severity until a significant result emerges by chance. P-hacking means running tests, stopping early, or switching methods until the p-value drops below 0.05. All of these inflate the apparent effectiveness of treatments.
The push for transparency
A global transparency movement has grown in response. The AllTrials campaign, launched in 2013, calls for all past and present trials to be registered and their results reported (AllTrials campaign). The World Health Organization has declared that results from all clinical trials must be published within 12 months of completion and runs an international registry platform to make reporting the norm (WHO International Clinical Trials Registry Platform).
Registries have one structural advantage over journals: structured data entry forces sponsors to report against their pre-specified endpoints, and there is no editorial selectivity. Every study gets a results page, due within 12 months, not years.
Registered reports are the most elegant fix on the publishing side. Journals peer-review and accept a study protocol before results are known, then commit to publishing regardless of outcome (Center for Open Science: registered reports). More than 300 journals have adopted the format. Preprint servers allow rapid sharing of all results without journal gatekeeping, and funders increasingly require raw data to be deposited in public repositories.
Why it matters to patients
Publication bias is not an academic nicety. Meta-analyses that include only published studies overestimate treatment effects by 15 to 30% on average. The video cites two famous corrections. In 2008, researchers used the Freedom of Information Act to obtain unpublished FDA data on antidepressants. The published literature showed these drugs as highly effective, but once the unpublished trials were included, the benefit for mild to moderate depression largely disappeared, and clinical guidelines were revised. And for Tamiflu, years of published claims that the drug reduced hospitalizations collapsed under scrutiny: full clinical study reports showed it shortened symptoms by about half a day, after governments had stockpiled billions of dollars' worth.
The bottom line
Every clinical trial result, positive, negative, or null, is a piece of evidence. When only the positive results are visible, the picture of what works becomes distorted and overoptimistic. Publication bias wastes money, skews the evidence base, and can lead patients to receive treatments that do not actually work. The movement toward registries, data sharing, and accountable publishing is one of the most important changes happening in medicine today. Science works best when the full picture is visible.
This article is based on the Clarity Clinical Solutions video "Publication Bias Results Reporting." Watch it here: Publication Bias Results Reporting
References
- Clarity Clinical Solutions video — the framework for this article: publication rates, outcome switching, selective reporting, the antidepressant and Tamiflu cases. https://www.youtube.com/watch?v=kQixtQiwue4
- ClinicalTrials.gov — FDAAA 801 registration and results submission requirements. https://clinicaltrials.gov/fdaaa
- NIH — Policy on clinical trial results reporting. https://grants.nih.gov/policy/clinical-trials/reporting/index.htm
- AllTrials — campaign for clinical trial registration and reporting. https://www.alltrials.net/
- Center for Open Science — Registered Reports publishing format. https://www.cos.io/initiatives/registered-reports
- WHO — International Clinical Trials Registry Platform. https://www.who.int/clinical-trials-registry-platform