Why Promising Findings Sometimes Fail
Why exciting early results often do not hold up: small studies, chance, differences between models and people, and the high failure rate of drug candidates in human trials.
Content last updated . This page has not been medically reviewed.
Early results are not final results
A striking early finding is a starting point. Many do not survive larger or better-designed testing. That is not always fraud or carelessness; it is how research filters ideas. This page explains the common reasons so that you can read emerging-science news with the right amount of caution.What the numbers show
- Highly cited studies get overturned. In a JAMA analysis of 49 original clinical studies published in major journals from 1990 to 2003 and cited more than 1,000 times, 45 claimed an intervention worked. Of those, 7 (16%) were later contradicted, 7 (16%) had found stronger effects than later studies, 20 (44%) were replicated and 11 (24%) were largely unchallenged. [1]
- Study design matters. Five of 6 highly cited nonrandomized studies were contradicted or had stronger effects than later work, compared with 9 of 39 randomized trials. Among randomized trials, the contradicted or exaggerated ones were smaller. [1]
- Most drug candidates that reach human trials fail. A 2022 review states that about nine of ten candidates entering clinical studies fail, and reports an overall success rate of 10 to 15 percent. For failures in 2010 to 2017 it lists lack of clinical efficacy (40 to 50 percent), unmanageable toxicity (30 percent), poor drug-like properties (10 to 15 percent) and commercial or planning reasons (about 10 percent). [2]
These are studies of drug development and highly cited research. They do not give a failure rate for any specific wellness product, and the authors' categories are their own groupings.
Why a good early result may not repeat
| Reason | In plain terms |
|---|---|
| Small samples | Chance can make a modest effect look large. Smaller trials were more often the ones later contradicted. [1] |
| Weaker designs | Without randomization and comparison groups, other explanations are harder to rule out. [1] |
| Wrong model | Cell or animal results may not apply to people. See cell studies vs animal studies vs human trials. |
| Stand-in outcomes | A marker moved but patients did not improve. See surrogate vs patient-important outcomes. |
| No real benefit or unacceptable harm | In human trials, lack of efficacy and toxicity were the leading reasons for failure in the 2022 review. [2] |
| Initial effects too strong | In the JAMA analysis, 16% of highly cited claims had found stronger effects than later studies. [1] |
What this means for readers
- A single study, especially a small, early or non-randomized one, is a lead. Look for repeat testing and for reviews that pool several studies: systematic reviews and meta-analyses explained.
- Failure at a later stage is normal and informative. It is not a reason to assume that all new research is worthless, only that "promising" is a prediction, not a result.
- Be careful with the opposite claim too: a failed trial does not prove an idea can never work; it shows it did not work in that design, population and dose.
Where to go next
Reading early-stage research covers animal studies, small trials and preprints. The emerging science evidence landscape shows how types of evidence fit together.Frequently asked questions
Does a failed trial mean the treatment is dangerous?
No. Trials fail mostly for lack of efficacy, but toxicity is also a common reason. The reasons differ by case, so check what the trial reported. [2]
Are randomized trials always right?
No. Even highly cited randomized trials were sometimes challenged and refuted over time, especially small ones, though they were contradicted less often than nonrandomized studies. [1]
What share of drugs that enter human trials succeed?
One 2022 review reports an overall success rate of 10 to 15 percent for clinical drug development. This is for drugs, not supplements or devices. [2]
Why do news stories rarely follow up on failures?
That is not something the sources here measured. See research hype and press releases for what was studied about early reporting.