A third-year fellow is sitting in a conference room with a whiteboard, sketching out a study design she has been building for two months: a modified intervention for a stubborn readmission problem on her hospital's cardiology floor. Her mentor is nodding along, mildly encouraging, when a visiting attending from a neighboring health system, in town for a joint grand rounds, glances at the board and says something that stops the room.
"We tried that. Four years ago. Recruitment collapsed at month six and we never wrote it up."
The fellow has spent two months on a literature search. She checked PubMed. She checked ClinicalTrials.gov, which shows that a trial was registered somewhere, occasionally, but tells her nothing about why it stopped or what the investigators learned. She found nothing, because there was nothing to find. The only record of this exact failure lived in one person's memory, and that person happened to be visiting today, for an unrelated reason, and happened to glance at a whiteboard.
If he had not been in the room, she would have spent the next eighteen months and a meaningful research budget rediscovering, the hard way, a recruitment problem someone else had already hit and quietly walked away from.
Nobody in medicine has a job whose purpose is to know who already tried this and failed, so the knowledge survives only by accident, in whoever happens to be standing nearby when it matters.
The scale of what never gets written down
This is not a story about one unlucky fellow. It is the modal outcome of clinical research, and the numbers are not close.
A 2025 Cochrane Review found that 47% of clinical trial results are never made public at all, drawing on 165,135 trials across 204 studies. Where results are eventually published, the median time to publication runs past two years, and positive results are published both more often and faster than negative ones. A separate analysis found that among trials with published protocols between 2011 and 2022, roughly 36% never published main results at all.
The bias compounds specifically around funding source. A 2025 Clinical Pharmacology and Therapeutics study of 1,542 trials found that exclusively industry-funded trials were roughly one-third as likely to report a negative outcome as trials without that funding structure (odds ratio 0.31). The share of published negative trials across four major journals, BMJ, JAMA, Lancet and NEJM, crept from 27.6% in 2000 to just 37.4% by 2020, a two-decade improvement of under ten percentage points, despite the entire modern registry infrastructure being built and matured across that exact window.
And even where a trial does get published, the profession is not reliably learning from what came before it. A 2002 JAMA analysis of trial reports in general medical journals found that 27 of 30 showed no systematic attempt to place their results in the context of prior trials. The field is not just failing to publish negatives. When it does publish, it is frequently not even looking backward at what was already known.
What this actually costs
The most cited estimate in this literature, from Chalmers and Glasziou's 2009 Lancet analysis, put avoidable waste across the full research pipeline (question choice, design, reporting, publication) at an estimated 85% of biomedical research investment. That is not a claim that 85% of studies fail. It is a claim that across every stage from choosing what to study through publishing what was found, the large majority of potential value is lost to preventable inefficiency, and unindexed negative results are a central piece of that loss.
Put a number on just one slice of it, at the scale of a single institution. Ten fellows a year, each spending six months on a project that turns out to replicate an already-abandoned attempt somewhere else in the field, is five FTE-years of research time, on the order of $500,000 in fully loaded cost, recurring every single year, at one hospital. Scale that finding across the roughly $47 billion NIH budget plus industry research spend, and even a 1% reduction in redundant work driven by better visibility into prior failed attempts is worth hundreds of millions of dollars annually.
There is a second cost that is not financial. Every redundant trial enrolls real patients in a study whose answer, somewhere, a colleague already privately knows. That is not just wasted money. It is exposure without the possibility of new knowledge, for people who consented to a study believing its outcome was genuinely uncertain.
Why the knowledge stays private
Ask any investigator why they did not publish a null or abandoned finding and the honest answer, almost universally, is some version of: it was not worth it, and it might have cost me something.
Reporting a negative result is professionally costly and largely unrewarded. Journals have historically preferred positive, novel findings, editors know it, and investigators internalize the preference before they ever submit anything. A negative result does less for a CV, a grant renewal, or a promotion case than a positive one, even when the negative result is scientifically just as informative.
The one journal built specifically to fix this closed. The Journal of Negative Results in BioMedicine, founded explicitly to give null findings a publication home, ceased operations in 2017. A structural attempt to solve this exact problem through the normal mechanism of academic publishing did not survive.
Registry compliance captures existence, not learning. ClinicalTrials.gov shows that a trial was registered. It does not, in the majority of cases, capture why a trial stalled, what the investigators would do differently, or what the honest lesson was. A registry entry that a trial exists is not the same information as a colleague telling you, over coffee, exactly where it went wrong.
And there is a real fear of being seen as a failed investigator, which is a distinct, more personal cost than the professional incentive problem. Admitting a study did not work, in a public, permanent, searchable record, feels different from admitting it quietly to a trusted peer who will not use it against you.
The result is that this knowledge exists in exactly the four places research knowledge always ends up when there is no formal channel for it: individual memory, private conversation at conferences, mentor-to-fellow oral tradition inside a single institution, and, occasionally, the kind of accidental hallway encounter that saved the fellow in the opening scene, and that cannot be relied upon to happen for anyone else.
Two different searches, constantly conflated
Most discussion of this problem collapses two genuinely different questions, and that conflation is part of why the fixes that exist do not work.
The first question is: does the published literature already answer this? That is what PubMed searches for, and it works reasonably well for questions with a robust positive literature.
The second question is: has someone already tried this and quietly stopped, for reasons that would change my plan? That is a completely different search, and nothing in the current research infrastructure answers it. ClinicalTrials.gov can tell you a trial with a similar design was registered. It cannot tell you it collapsed at month six because of a recruitment problem specific to a patient population very much like yours, which is exactly the piece of information that would have saved the fellow in the opening scene two months of work.
Registries capture the first kind of question adequately, if imperfectly. They were never designed to capture the second kind at all, and no other institution has stepped in to build that capability, because the second kind of knowledge is reputationally sensitive in a way the first kind is not.
The structural failure: this is a trust problem, not a publishing problem
Here is the diagnosis that the standard fixes keep missing. Publication bias will not be fixed by more publication, because the reason negatives go unpublished is not that there is no venue. It is that sharing a negative result publicly, permanently, and searchably carries a real professional cost that sharing it privately, to a trusted colleague, does not.
That means the fix has to be built around reputational safety, not around openness. A confidential, verified channel where "I tried this and it did not work, here is why" is safe to say, because the audience is limited to verified peers who cannot use the admission against the person who made it, would capture exactly the information that public registries and journals structurally cannot.
This is the same shape of failure documented elsewhere in this series around morbidity and mortality conferences, and around informal peer consultation after a complication: the knowledge that would help the next person is not secret, it is simply confidential by necessity, shared only under conditions of trust that no open, indexed, public system can offer.
What would actually work
A confidential attempt registry, not another public database. A place where an investigator can log an idea, a design, and why it stopped, visible only to verified peers with a genuinely matching question, not to the public and not to funders directly.
Routed by question, not browsed by category. The value is not a searchable archive someone stumbles onto. It is a system that actively connects "did anyone try this?" to the specific person who already knows the answer, the way the visiting attending in the opening scene did by accident.
A credit mechanism for answering, not just for asking. The investigator who logged an abandoned attempt years ago should get something (a professional credit, a reciprocity signal, a track record) for being the one whose failure saved someone else's eighteen months, because right now there is no reward at all for having tried and failed.
No patient data, no manuscripts, protected the way M&M conference discussion is protected. The content has to be narrow enough (idea, design, what stopped it, what would be different) that it carries genuinely low reputational risk to share, while still being the exact information a colleague needs.
A public aggregate signal without exposing individuals. An annual figure, something like "X% of members abandoned a study in the last three years, and the leading cause was recruitment," would demonstrate the scale of the problem without asking any individual investigator to bear public exposure for their own attempt.
A pre-award check built for funders. A funder or IRB reviewing a proposed study should be able to ask, confidentially, whether this exact approach has already been quietly tried and abandoned elsewhere in the field, before committing budget to it.
What you can do now
If you are an investigator or fellow
Before you commit real time to a design, ask a wider circle than your own mentor whether anyone has tried something similar. The visiting attending in the opening scene was a lucky accident. Deliberately asking colleagues at other institutions, not just your own department, replicates some of that luck on purpose.
Write down your own abandoned attempts, even if you never publish them. The two-page account of what you tried, why it stopped, and what you would change is exactly the document that would help the next person, and right now it almost never exists anywhere, even privately.
Ask specifically about failures when you reach out to a peer about a new project. "What have you tried on this that didn't work" produces dramatically more useful answers than "what's the state of the evidence," and most people will answer it candidly if asked directly and confidentially.
If you mentor or lead a research division
Normalize discussing abandoned projects in lab meeting or division conference, not just completed ones. Most divisions already have an informal version of this. Making it explicit and routine, rather than incidental, is a low-cost way to move some of this knowledge out of individual memory and into shared department memory.
Ask new fellows, before they finalize a design, who else in the broader field might have already tried it. This is a five-minute question that the current system gives fellows no structured way to ask on their own.
If you fund or govern research
Look seriously at what a confidential pre-award check would be worth. Even a small reduction in redundant, already-known-to-fail study designs, at the scale of a national research budget, is worth pursuing with real infrastructure rather than leaving to chance encounters.
Consider funding the negative result directly, as its own deliverable. No journal will prioritize an honest account of an abandoned study, and no conference will feature it, which is precisely why it needs direct funding rather than encouragement, the same way negative results in industry trials sometimes get reported only because a regulator requires it.
Frequently asked questions
What percentage of clinical trials are never published? 47% of clinical trial results are never made public, according to a 2025 Cochrane Review covering 165,135 trials across 204 studies. A separate analysis found roughly 36% of trials with published protocols between 2011 and 2022 never published main results at all.
Why are negative clinical trial results not published? Journals have historically favored positive, novel findings, and investigators face little professional reward and some professional risk in publishing a null result. Exclusively industry-funded trials were about one-third as likely to report a negative outcome as other trials in a 2025 study of 1,542 trials, and the share of published negative trials across four leading journals rose only from 27.6% in 2000 to 37.4% in 2020.
What is the Journal of Negative Results in BioMedicine, and does it still exist? It was a journal created specifically to publish null and negative findings in biomedicine. It ceased operations in 2017, illustrating that a dedicated publication venue alone did not solve the underlying incentive problem.
How do I find out if my research idea has already been tried? Currently, mainly through PubMed, ClinicalTrials.gov, and informal conversation with mentors or colleagues, none of which reliably surfaces abandoned or unpublished attempts. ClinicalTrials.gov shows that a trial was registered but rarely explains why it stopped or what investigators learned, which is usually the more useful information.
What is research waste in medicine, and how large is it? A widely cited 2009 Lancet analysis by Chalmers and Glasziou estimated that an estimated 85% of biomedical research investment is avoidably wasted across question selection, study design, reporting and publication, a figure that has anchored two decades of subsequent research-waste literature including the 2014 Lancet REWARD series.
Is there a database of failed clinical trials? Not in the sense of documenting why trials failed or what investigators learned. ClinicalTrials.gov records registration and, increasingly, results reporting compliance, but has no mechanism for capturing informal lessons from abandoned or null studies, which remain confined to personal networks and institutional memory.
The bottom line
The fellow in the opening scene was one conversation away from repeating eighteen months of work that someone else had already quietly done and abandoned. Most investigators never get that conversation. They spend the eighteen months, generate the same disappointing result, and, following the same incentives that suppressed the first attempt, quietly do not write it up either. The knowledge dies twice: once when the original team decided it was not worth publishing, and again when the next team never heard about it at all.
The numbers describing this are not ambiguous. Nearly half of all trial results never reach the public record. An estimated 85% of research investment is avoidably wasted across the full pipeline. Two decades of registry infrastructure moved the published-negative-trial share less than ten percentage points. This is not a problem that better publication practices, on their own, have been able to fix, because the actual barrier is not the absence of a venue. It is that the information is only safe to share under trust, and nothing has ever built the confidential, verified, routed channel that trust requires at scale.
Somewhere in the country right now, in a hospital not affiliated with the one currently designing a very similar study, is an investigator who already knows exactly why this approach does not work, and would say so immediately if anyone verified and trustworthy asked.
Part of a series on the missing professional infrastructure of healthcare. Previously: Deadline Expertise
Evidence note: the 47% non-publication figure and trial count are from a 2025 Cochrane Review as reported via TranspariMED, drawing on the underlying Cochrane methodology review (MR000011). The 36% protocol-to-publication gap figure is from a 2024 JAMA Network Open-linked analysis covering 2011 to 2022 registered protocols. The industry-funding and negative-result likelihood figure (odds ratio 0.31, n=1,542) and the 27.6%-to-37.4% published-negative-trial trend are from a 2025 Clinical Pharmacology and Therapeutics study. The finding that 27 of 30 trial reports showed no systematic contextualization against prior trials is from a 2002 JAMA study and describes a small, dated sample; whether this has improved materially since 2002 was not independently re-verified here. The 85% research-waste estimate is the widely cited Chalmers and Glasziou figure from a 2009 Lancet analysis; it is an estimate across a complex, multi-stage pipeline and should be read as an order-of-magnitude framing rather than a precise, individually verifiable statistic. The $500,000 single-institution cost example and the "hundreds of millions" national extrapolation are illustrative calculations built from these published figures, not independently published estimates themselves. The Journal of Negative Results in BioMedicine's 2017 closure is a documented fact about that specific publication.
