Investigation
5 flagged titles in 902
A screen for negative findings across every record title this corpus holds returned 5 candidates out of 902. That measures the screen as much as the literature, and the first version of the screen was wrong about roughly half of what it found.
Published 13 October 2026Subject: 902 record titles across 77 compoundsCounted 13 October 2026
- 5 of 902 titles were flagged as reporting an absent, failed or negative effect: 0.55%, on 5 of 77 compounds.
- 72 compounds returned nothing. The honest reading of that is "nothing in the titles", not "nothing negative exists".
- Only titles were read. A null result reported inside a paper whose title is neutral is invisible to this method, and most titles are neutral.
- The first version of the screen was roughly half false positives, because "heart failure" and "kidney failure" name conditions rather than failed studies. Three corrections followed; the current marker list is printed in full below.
- A further 1 record was separated out rather than counted as negative: a non-effect on a pharmacokinetic or tolerability measure is usually a reassuring result.
What the screen is
Every record this corpus carries has a title. A short list of negation markers was matched against all 902 of them. A match makes the record a candidate: a title that reads as reporting an absent, failed or negative effect. It is not a finding about the compound, and it is not a judgement on the study.
Across 77 compounds, the screen returned 5 candidates, on 5 compounds. 72 compounds returned nothing. That is 0.55% of titles screened.
| Marker | What it matches | Titles caught |
|---|---|---|
| no effect | a title stating no effect, with or without a qualifier such as apparent or significant | 1 |
| does not act | does, did or do not, followed within two words by an effect verb | 4 |
| failed to | the phrase failed to, which cannot be a condition name | 0 |
| without effect | without effect, benefit or improvement | 0 |
| negative result | negative result, finding, trial or study | 0 |
| no benefit | no benefit, improvement, difference, association or change | 0 |
| ineffective | ineffective, or lack of effect or efficacy | 0 |
| not associated | not associated, superior or effective | 0 |
| unaffected | unchanged or unaffected | 0 |
| retraction | retracted or retraction | 0 |
Two of the ten markers caught anything. The other eight returned nothing across all 902 titles, which is itself a result about the screen.
The five
These are the five candidates in full. The titles are quoted exactly as the records carry them. Read them as what the screen found, not as what the literature says: a title is a sentence an author wrote about one experiment, and the piece makes no claim about what any of these papers concluded or how they were received.
| Named in the title | Marker | Year | Record |
|---|---|---|---|
| Capromorelin | does not act | 2025 | The 4 prokinetic drugs metoclopramide, cisapride, pyridostigmine, and capromorelin do not increase fecal output or food intake in healthy New Zealand rabbits (Oryctolagus cuniculus) PMID 40179972 |
| Mechano growth factor | no effect | 2014 | Mechano-growth factor peptide, the COOH terminus of unprocessed insulin-like growth factor 1, has no apparent effect on myoblasts or primary muscle stem cells PMID 24253050 |
| Noopept | does not act | 2019 | Drug with Neuroprotective Properties Noopept Does Not Stimulate Cell Proliferation PMID 30788746 |
| Thymosin beta 4 | does not act | 2012 | Thymosin beta 4 treatment after myocardial infarction does not reprogram epicardial cells into cardiomyocytes PMID 21907210 |
| Vasoactive intestinal peptide | does not act | 2008 | Vasoactive intestinal peptide causes marked cephalic vasodilation, but does not induce migraine PMID 18254893 |
Each row is one bibliographic record retrieved by the compound's own literature query. No full text was read.
The first pass was wrong about roughly half of what it found
This matters more than the count. The first version of the screen used a bare match on failure and a bare match on does not. It was roughly half false positives, and the reason is specific: in clinical titles "failure" almost always names a condition rather than a failed study.
It flagged a trial of semaglutide in patients with obesity-related heart failure, and a trial comparing early vasopressin with norepinephrine on kidney failure. Neither is a negative result. Both are titles containing the word. A reader shown that output would have learned, correctly, to ignore the whole signal.
Three corrections followed, and they are the reason the current list looks the way it does in the table above. The marker failure was removed entirely, leaving only failed to, which cannot name a condition. A bare does not was required to be followed within two words by an effect verb. And a non-effect on a pharmacokinetic or safety measure was moved out of the negative category altogether.
The record that was moved out
The screen moved 1 record out of the negative category: it matched a marker in a pharmacokinetic and tolerability context, so it is reported separately rather than as a negative finding.
| Named in the title | Year | Record |
|---|---|---|
| Cagrilintide | 2026 | Renal or Hepatic Impairment Does Not Affect Pharmacokinetics, Safety, or Tolerability of Subcutaneous Cagrilintide PMID 42228334 |
A title reporting that something does not alter exposure or tolerability is usually a reassuring result. Counting it as discordance would invert its meaning, so the dataset carries it in a separate category column.
What a rate of this size can and cannot mean
0.55% of titles is a measurement of the screen at least as much as of the literature. Three things would each produce it on their own, and this analysis cannot tell them apart.
A null result can be reported in a paper whose title is neutral, which is the normal case: most titles name the subject and the model, not the direction of the result. A null result can go unpublished, which is a well-documented pattern in the research literature generally and is not something a title screen can observe. Or the records here can simply be the most-cited work under each name, which is what the retrieval was sorted by, and heavily cited papers are not a random sample of outcomes.
So the defensible statement is narrow, and it is the one this publication makes: of the 902 titles this corpus carries, 5 are phrased as reporting an absent or negative effect. Anything stronger would need the full texts, and this publication does not read them.
What this cannot show
- No full text was read, so no row here states what any study concluded. A paper whose title reads as negative may report a positive result in a subgroup, and the reverse.
- Only titles were screened. Abstracts, results sections and supplementary material were not, so the true number of null results among these records is unknown and is certainly higher.
- The screen cannot distinguish a literature with few negative findings from a literature whose negative findings are phrased neutrally, or from one whose negative findings were never published.
- The titles screened are a citation-ranked sample capped at twelve per compound, not the literature. A negative finding outside that sample is not counted here.
- A marker that caught nothing is not evidence that nothing matches it. Eight of the ten markers returned zero, which may mean the phrasing is rare, or that the sample is small.