who paid for the research behind your medication?

the number that matters most is 87

Almost every article about pharmaceutical funding leads with a percentage. We are going to lead with a denominator instead, because the denominator is the finding.

This audit counts 235 verified study records across 51 medications, substances and topics, harvested from the Resolv evidence library on 2026-08-17 and 2026-08-18. Every citation was fetched live from PubMed and Europe PMC and re-verified in an independent second pass. For each one we tried to read the funding statement off the paper.

The library has kept growing since that harvest: it held 251 published resources on 2026-08-19. Every count on this page is the 235-record harvest, not the library as it stands today — we would rather leave the arithmetic reproducible than quietly restate it against a moving denominator. The current library is free to read in full.

We could read 148 of them. We could not read 87. That is 63% readable, 37% not. Those 87 are not concealed, and we are not implying that they are — they are behind paywalls and access walls. That distinction is the whole point of this page.

Of the 148 studies whose funding statement we could actually read:

• 95 were independently funded — government agencies, research councils, universities, charitable foundations (64% of the 148).
• 25 declared no funding at all (17%).
• 11 were mixed — public or charitable money plus commercial money, usually a grant plus donated study drug (7%).
• 17 were industry-funded — paid for by a party with a commercial interest in the product being studied (11%).

Those four buckets sum to 148, and 148 plus 87 unknown sums to 235. Every table below shows its own denominators for the same reason.

the 11% figure is the wrong number to quote, and here is why

If you stop reading here you will walk away with "only about one in nine psychiatric studies is industry-funded", and that would be a misleading thing to carry around. Our corpus is not a random sample of the psychiatric literature. It deliberately over-samples the top of the evidence pyramid: 127 of the 235 records are syntheses — meta-analyses, systematic reviews, guidelines — because those are the records worth putting in front of a patient.

Syntheses are overwhelmingly publicly funded. The trials they pool are not. Split the corpus by evidence layer and the pattern is stark:

• Of the 48 individual randomised trials whose funding we could read, 14 were industry-funded and 7 were mixed — so 21 of 48, about 44%, carried at least some commercial money.
• Of the 45 meta-analyses of randomised trials whose funding we could read, zero were industry-funded and 3 were mixed.

The syntheses that generate the headline numbers your prescriber quotes are independent. The raw trials underneath them are where the money is. And this dataset genuinely cannot tell you the ratio inside those pooled trials — we did not count them, so we are not going to state a figure for them.

What we can do is quote the syntheses on themselves. Cipriani 2018 in The Lancet — the network meta-analysis of 522 antidepressant trials in 116,477 people, the single most-quoted piece of evidence about whether antidepressants work — is publicly funded by the UK National Institute for Health Research and the Japan Society for the Promotion of Science, is PROSPERO-registered, and states in its own declarations that most of the underlying trials were industry-funded and that several of its authors have taken pharmaceutical-company lecture or consultancy fees. That is what an honest chain of evidence looks like: an independent analysis of a largely commercial literature, saying so out loud.

what industry funding does and does not mean

We are going to be careful here, because this is where these pages usually go wrong in one of two directions.

It is not disqualifying. Some of the most useful studies in this corpus are industry-funded, and several of them produced results their funders would not have chosen:

• EAGLES (Anthenelli 2016, The Lancet, 8,144 smokers) was funded by Pfizer and GlaxoSmithKline, the manufacturers of the drugs studied, with company employees among the authors. It is also the trial that settled whether bupropion causes serious neuropsychiatric harm in people with psychiatric illness. It found no significant increase. Nobody has done better, and a smaller independent trial would not have answered the question.
• BALANCE (Geddes 2010, The Lancet) was part-funded by Sanofi-Aventis, a valproate manufacturer. Its result favoured lithium over valproate.
• VITAL-DEP (Okereke 2021, JAMA, 18,353 adults) was NIH-funded with capsules donated by Pronova BioPharma. It found that omega-3 did not prevent depression and, on one co-primary outcome, was associated with a slightly higher risk — a result unfavourable to the donated product.

It is also not nothing. The clearest demonstration in this corpus is paroxetine Study 329. The original 2001 report, funded by GlaxoSmithKline, concluded that paroxetine was effective and well tolerated in adolescent depression. An independent reanalysis of the same patient-level data — Restoring Study 329 (Le Noury 2015, BMJ, 275 adolescents, declared no funding) — concluded the opposite: paroxetine was no more effective than placebo and was associated with clinically significant increases in suicidal ideation and behaviour. Same data, different analysts, reversed conclusion. The 2001 paper has never been retracted.

On the broader question — does industry funding systematically shift results? — there is a real methodological literature, and its general finding is that industry-sponsored studies more often report results and conclusions favourable to the sponsor than independently funded studies of the same question. We are deliberately not putting a number on that here, because no number for it exists anywhere in this dataset, and the rule for this page is that every figure on it is reproducible by counting our own records. Quoting an effect size we did not compute would be exactly the move this page exists to argue against. What we will say plainly is that funding belongs on the label, next to sample size and study design, and that a reader should weight it — not use it as a veto.

where the 17 industry-funded studies actually are

They are not spread evenly. Eleven of the seventeen are in three places:

• Supplements — 6 of 34 readable records. Every one is funded by, or supplied by, the company selling the exact branded extract tested: Ixoreal Biomed (KSM-66 ashwagandha), Arjuna Natural (Shoden ashwagandha), Taiyo Kagaku (Suntheanine L-theanine), Neurobrands, Ethical Naturals (AlphaWave L-theanine). Three of L-theanine’s four readable records are industry-funded. This is a lane where the trials and the marketing come from the same address.
• Psychedelics and MDMA — 5 of 12 readable records. Compass Pathfinder, the Usona Institute and MAPS all sponsored trials of the product they are seeking to bring to market. We score sponsor-developers as industry regardless of their non-profit status.
• SSRI sexual dysfunction — 2 of 3 readable records. Both are Pfizer-funded sildenafil trials.

The remaining six are single records under bupropion, quetiapine, kratom, magnesium, esketamine and LSD.

Note what is not there. Across the 21 prescription psychiatric medications in this corpus — 102 records, 66 with a readable funding statement — only 2 are recorded as industry-funded. That is not because industry does not fund antidepressant research. It is because the records we selected for those medications are mostly Cochrane reviews and publicly funded network meta-analyses. Read that row as a fact about our sampling, not about the drugs.

why 87 statements were unreadable

Three causes, in order of size.

Paywalls. Funding and conflict-of-interest declarations usually appear in the full text, not the free PubMed abstract. Where the full text costs money, the funding statement costs money. In this corpus, records published in journals we could not open in full — the Journal of Clinical Psychiatry, the American Journal of Psychiatry, Archives of General Psychiatry — are disproportionately unknown. Split by venue tier: 24% of the 112 top-tier-journal records are unreadable, against 57% of the 100 specialty-journal records.

Access walls on aggregators. The Cochrane Library returned HTTP 403 to automated access repeatedly during both harvests, which blocked the support-and-funding sections of several Cochrane reviews. Cochrane reviews are among the most transparently funded documents in medicine; we simply could not read several of them on the day.

Age. Mandatory funding disclosure is a recent norm. Readability rises monotonically with publication year, and the oldest bucket is a clean zero:

Nothing was inferred to fill those gaps. We did not reason from the journal, the authors, the country, or the topic. A trial that is obviously a registration trial for a manufacturer’s indication — BOLDER I for quetiapine in bipolar depression, for instance — is still recorded as unknown, with a note saying sponsor involvement is plausible but unverified. Guessing correctly nine times out of ten would still make the tenth entry a fabrication, and there is no way for a reader to tell which one it was.

how we counted

The full method is in the box below the tables. The short version:

• Unit of counting: the study record. 235 records across 46 source files. Twelve papers appear in more than one file — Cipriani 2018 appears in nine antidepressant files alone — so the 235 records represent 216 distinct papers. Counted by distinct paper instead, the funding split is 79 independent, 24 none, 11 mixed, 17 industry, 85 unknown: 131 of 216 readable, 61% rather than 63%. Both numbers are in this page; neither is hidden.
• Four readable categories plus unknown. Independent (government, research council, university, charitable foundation), none (the paper states it was not funded), mixed (public or charitable money together with commercial money, including donated study drug), industry (a party with a commercial interest in the product studied, including non-profit sponsor-developers).
• Percentages are always over the readable denominator and the denominator is always printed. Where a medication has 5 records and 2 readable funding statements, this page says 2 of 5, not 40% of something.