In my last post, I showed that non-replicable papers get cited 153 times more than replicable ones — that science's knowledge system preferentially amplifies its failures. But citations aren't the end of the chain. Citations become literature reviews. Reviews become drug targets. Targets become clinical trials. And patients enroll.
What happens when the pharmaceutical industry builds a $28 billion annual pipeline on top of science that mostly can't be replicated?
We know the answer. We've been watching it for decades.
Three Facts That Can't All Be True
Fact one: We know preclinical research mostly fails replication. In 2012, researchers at Amgen tried to reproduce 53 landmark cancer papers — the kind that launch entire research programs. They succeeded with six. That's an 11% replication rate. Bayer ran a similar exercise and found 64% of projects inconsistent with published data. The Reproducibility Project: Cancer Biology tested 50 experiments from 23 high-impact papers: only 40% of positive results replicated, and effect sizes were on average 85% smaller than originally reported.
This isn't a fringe concern. Freedman et al. estimated in 2015 that more than 50% of preclinical research is irreproducible, costing the United States alone $28 billion per year in wasted effort.
Fact two: We know how to do better. Nelson et al. showed in 2015 that drug targets with genetic evidence supporting them are dramatically more likely to succeed. Updated analyses have refined this: genetically validated targets are 2.6 times more likely to reach approval. Targets with top genetic priority scores are 8.8 times more likely to advance from Phase I to Phase IV. Two-thirds of FDA-approved drugs in 2021 had genetic evidence behind them.
Fact three: The pipeline keeps advancing drugs without replication or genetic validation anyway.
These three facts exist simultaneously. The industry knows preclinical science is unreliable. It knows genetic validation works. It proceeds without either.
What the FDA's Own Records Show
In 2017, the FDA published 22 case studies where Phase 2 results didn't hold up in Phase 3 — the stage where you've already enrolled thousands of patients. Fourteen failed on efficacy. One failed on safety. Seven failed on both.
Two drugs didn't just fail to treat the condition. They made it worse.
Semagacestat, developed for Alzheimer's disease, showed amyloid reduction in Phase 2. In Phase 3, patients experienced cognitive decline and increased cancer rates. Torcetrapib, developed for cardiovascular disease, improved cholesterol numbers in Phase 2. In Phase 3, it increased mortality.
These weren't edge cases or unlucky draws. They were predictable consequences of building drug targets on science that was never verified.
Case Study: Thirty Years of Amyloid
The amyloid cascade hypothesis has dominated Alzheimer's research since John Hardy proposed it in 1992. The idea: amyloid beta plaques accumulate in the brain, causing neurodegeneration. Remove the plaques, stop the disease.
For three decades, nearly every major Alzheimer's drug program targeted amyloid. Nearly every one failed.
In 2006, neuroscientist Sylvain Lesné published a landmark paper in Nature claiming to have found the first cause-and-effect evidence linking a specific amyloid form to memory loss. The paper became foundational — cited extensively, used to justify billions in research and drug development.
In June 2024, Nature retracted the paper. The images were fabricated. In 2025, Lesné resigned from the University of Minnesota — nearly 20 years after publication.
Meanwhile, dozens of anti-amyloid drugs had entered and failed clinical trials. Aducanumab received a controversial FDA approval in 2021 despite mixed efficacy data, an approval so disputed that the FDA's own advisory committee members resigned in protest.
By 2022, Hardy himself — the original hypothesis author — acknowledged that "we thought sorting out amyloid was to sort out dementia." A 2024 review in Frontiers in Aging Neuroscience called the hypothesis "a working hypothesis, no less but certainly no more." By 2025, more than 60% of Alzheimer's trials had moved to non-amyloid targets — 138 drugs across 182 trials.
Thirty years. Billions of dollars. Fabricated foundational evidence. And a field that is only now pivoting.
Case Study: Simufilam — The Full Arc
If the amyloid story shows the slow erosion of a flawed paradigm, Cassava Sciences' simufilam shows the full arc compressed into a few years.
Simufilam was an Alzheimer's drug built on preclinical work by neuroscientist Hoau-Yan Wang. In 2021, a citizen petition filed with the FDA alleged fraudulent images in the supporting research. A CUNY investigation accused Wang of "egregious misconduct" involving 20 research papers that underpinned simufilam's scientific rationale.
In June 2024, Wang was indicted on charges of manipulating Western blot images to fabricate evidence of the drug's mechanism.
In March 2025, simufilam failed Phase 3. Not on one endpoint — on every primary, secondary, and exploratory endpoint. No better than placebo across the board. The program was discontinued.
In October 2025, Wang's indictment was dismissed with prejudice. In February 2026, the DOJ closed its investigation entirely.
The full arc: Fraudulent preclinical data. Drug advanced to Phase 3. Drug failed on every measure. Indictment dropped. Investigation closed. No accountability.
As Vanderbilt neurologist Matthew Schrag put it: there was "strong evidence of corruption of the basic science and pharmacology data on which the phase 2 trials were based." The system produced no consequences.
The Chain
Put the pieces together and a supply chain of failure emerges:
- Unreliable preclinical study published in a top journal
- Gets cited 153 times more than reliable studies (the replication amplifier)
- Pharma companies build drug targets on those citations
- When they check replication: Amgen finds 89% can't be reproduced
- Drugs advance to human trials anyway
- 90% of clinical trials fail — the highest failure rate in any industry
- Meanwhile, genetic validation — which actually works — is used for only a fraction of targets
This isn't a knowledge problem. The industry knows preclinical science is unreliable. The industry knows genetic validation improves success rates. The tools exist. The evidence is clear. The pipeline continues to be built on sand because the incentives reward starting fast, not starting right.
The Human Cost
Behind every failed Phase 3 trial are real patients who enrolled believing the science was sound. Patients on semagacestat who experienced cognitive decline they might have avoided. Patients on torcetrapib whose cardiovascular risk increased. Patients in simufilam trials whose Alzheimer's progressed while they received a drug built on fabricated evidence.
An estimated 92.6% of protein-disease pairings used as drug targets may represent false discoveries. Each false discovery doesn't just waste money. It wastes the one thing patients with degenerative diseases don't have: time.
The Contradiction
The pharmaceutical industry presents itself as the pinnacle of evidence-based decision-making — rational drug design, rigorous clinical testing, regulatory oversight at every stage. The evidence tells a different story: a pipeline where most preclinical foundations can't be replicated, where tools that improve success rates are underused, where fraud can reach Phase 3 without detection, and where failure produces no structural reform.
The industry isn't failing despite the evidence. It's failing because of a system that rewards speed over verification, novelty over replication, and advancement over accountability.
The science to fix this exists. Genetic validation works. Replication requirements could be mandated. The gap between what we know and what we do isn't a mystery. It's a choice.