Clinical failure rates over the decades: yikes
Clinical drug development continues to see roughly 90% of candidates fail in human trials, prompting debate over whether this rate reflects deep gaps in our understanding of biology or simply the intrinsic difficulty of the problem. Commenters contrast drug R&D with more mature engineering fields, arguing that human biology is far harder to model, preclinical systems often don’t predict human outcomes, and regulatory and safety standards have tightened over time. Others contend that high failure rates may be economically or scientifically “optimal” for a high‑risk, high‑reward domain, while pointing to misaligned incentives, limited impact of AI, and the ethical and financial costs of failed trials as reasons to seek better approaches.
Clinical failure rate and its meaning
- Many see ~90% clinical failure as unsurprising given biological complexity and safety constraints.
- Some argue the stability over decades suggests an equilibrium: as tools improve, projects get more ambitious, keeping the success rate flat.
- Others see the rate as “insanely” high, implying our models and methods are still poor, not that the system is optimal.
Comparisons to other fields
- Several commenters call comparisons to cars/airplanes “bizarre” or misleading:
- Engineering prototypes quietly fail at high rates before products ship.
- Finished drugs correspond more to final products than to prototypes.
- Others propose better analogies: early vacuum tubes before understanding electrons, or deep learning’s trial‑and‑error era.
Why drug development is so hard
- Biology is messy, poorly constrained, and hard to simulate; animal and cell models often fail to predict human outcomes.
- You can’t “patch” a molecule after a phase 3 miss; failure often means a full restart.
- There’s extensive preclinical triage, yet many “good” candidates still fail in humans for efficacy, endpoints, or safety.
Economics, incentives, and “optimality”
- Some argue ~10% success is economically rational: if success gets easier, more marginal, riskier projects are funded until rates drop again.
- Others dispute this, noting per‑trial costs are massive and firms have strong reasons to kill losers early, but internal incentives (career rewards for advancing programs, not killing them) may distort decisions.
- Debate over whether high failure is desirable exploration vs. waste of scarce resources.
Regulation and safety trends
- Regulatory hurdles have tightened (e.g., screening for cardiac risks like hERG inhibition), eliminating candidates that historically might have reached market.
- This improves safety but raises costs and reduces apparent success.
AI/ML and modeling
- Long‑running skepticism that “AI will fix” the failure rate; ML has been used in pharma for decades with only incremental impact.
- Hype about AI discovering safe drugs in a few years is viewed as tech‑bro reductionism.
Alternatives, precision, and side topics
- Some mention drugs that “fail” overall but work for identifiable subgroups; without reliable pre‑selection, this isn’t yet a true alternative model.
- Discussions touch on personalized medicine, copycat or incremental drugs (better side effects, dosing, routes), and ethical issues around animal use and long‑term genetic impacts of medicine.