Why Most Preclinical Breakthroughs Never Reach Patients: Rigor and Reproducibility

Most preclinical breakthroughs never reach patients because the experiments behind them were not rigorous or reproducible enough to have been reliably true in the first place. When industry teams tried to reproduce landmark laboratory findings, they succeeded only a small fraction of the time. The usual culprits were not exotic: missing blinding, absent randomization, no sample-size justification, and results that were never independently replicated. A finding that cannot be reproduced was never a solid foundation to build a therapy on, and the failure often shows up years later, in a clinical trial that quietly reads out negative.

This is a different problem from the funding chasm I have written about elsewhere. The valley of death describes work that is correct but goes unfunded. Here I am describing something more uncomfortable: work that looked correct, attracted attention, and turned out to rest on shaky methods. This does more damage, because it wastes the money and patient participation of a trial before anyone learns the original signal was fragile.

What rigor and reproducibility actually mean

The National Institutes of Health defines scientific rigor as the strict application of the scientific method to ensure unbiased, well-controlled design, methodology, analysis, and reporting. Reproducibility is the plainer test that comes after: can independent scientists, working from the published description, obtain the same result? The NIH treats these as prerequisites for building the next study on top of the last one, which is exactly what translation requires. A drug program is a tower of experiments, and every floor assumes the one below it holds.

This matters for patients because a single unreliable floor can stay hidden for a long time. An eye-catching effect in a mouse model becomes a target, the target becomes a program, and the fragility of the original observation is not discovered until a large, expensive trial fails to reproduce it in people.

The evidence that the foundation is often weak

The clearest early alarm came in 2012, when C. Glenn Begley and Lee Ellis reported in Nature that scientists at Amgen had reproduced the findings of only six of fifty-three landmark preclinical cancer papers over a decade, roughly one in ten. Their commentary, "Raise standards for preclinical cancer research," argued that the problem was systemic rather than a matter of a few bad papers.

More recent work quantifies why. Writing in the Journal of Clinical Investigation in 2023, Michael McGill and David Threadgill noted that only about eleven percent of published landmark oncology findings were later validated in clinical trials, with each failed trial carrying an estimated economic burden on the order of tens of millions of dollars. They trace the failures to identifiable causes: species-specific biology that does not carry over to humans, statistical choices that can turn one dataset into opposite conclusions, and preclinical models so genetically uniform that they cannot represent a diverse patient population. Their proposed addition to rigor and reproducibility is robustness, meaning a result should hold across different genetic backgrounds before anyone bets a clinical program on it.

Surveys of the literature show how often the basic safeguards are absent, and how slowly that has changed. A nationwide analysis by Kousholt and colleagues in PLOS One in 2022 compared animal studies from 2009 and 2018. Reporting of randomization rose from about a quarter of papers to roughly two in five, and reporting of a sample-size calculation climbed from around three percent to fourteen percent. The direction is upward, but the levels remain low: even at the later point, most studies still did not report a sample-size calculation. These are not subtle refinements. They are the difference between a measured effect and wishful measurement.

Why weak methods produce false positives that survive

Each missing safeguard tilts a study toward finding something whether or not something is there. Without randomization, sicker or healthier animals can drift into one group. Without blinding, an investigator scoring an ambiguous outcome can, without any dishonesty, nudge it toward the hoped-for direction. Without a prespecified sample size, a study can be stopped when the numbers look good. Without reporting negative results, the literature fills with the lucky positives and hides the failures, so the published picture looks far more promising than reality.

The trap is that these fragile positives are often the most exciting ones. A large, clean, real effect tends to reproduce regardless of method. It is the borderline, first-of-its-kind result that most needs guardrails and most often lacks them, so the very findings that generate breakthrough headlines are the ones most likely to evaporate under scrutiny.

What the standards now ask for

Funders have responded by making rigor an explicit condition rather than an assumed virtue. The NIH now asks applicants to address the scientific premise of their work, rigorous experimental design, consideration of relevant biological variables such as sex, and authentication of key resources like cell lines and antibodies. Institute guidance spells out the specifics reviewers look for: sample-size justification and power calculations, documented randomization and blinding, criteria for excluding data, reporting of both negative and positive results, and independent validation of key preliminary findings.

None of this guarantees a molecule will work. It guarantees something more modest and important: a result strong enough to justify the next expensive, human step was measured in a way that can be trusted. Reporting standards for animal experiments push the same way, asking authors to state plainly what they did so others can repeat it.

How to read a breakthrough more carefully

For a reader outside the laboratory, a few questions separate durable findings from fragile ones. Was the work reproduced by an independent group, or does it rest on a single striking paper? Were the animals randomized and the outcomes scored blind? Did the study report how it chose its sample size, and did it include both sexes? Does the effect hold across more than one genetic background? A finding that clears these is not proven in humans, but it is worth carrying forward. A finding that does not may still be true, yet it is the kind that has repeatedly failed the translation it was announced to promise.

This is general education, not medical advice, and decisions about your own care belong with a qualified clinician who knows your history. The practical point is calibration: a preclinical breakthrough is a hypothesis about people, and its worth depends far more on how carefully it was tested than on how impressive it sounds.