ideasIdea
A machine-readable taxonomy of why cancer drugs fail
Drugs fail for very different reasons: the target was wrong, the drug did not reach it, the side-effects were too bad, or the trial was badly designed. Recording which reason each time would show where the system is broken.
Analyses of attrition (for example those published by AstraZeneca and Pfizer on their own pipelines) show that failure reasons are learnable but rarely recorded consistently. A shared taxonomy (target biology, exposure, safety, efficacy in unselected population, biomarker failure, design, commercial) applied to every discontinued oncology programme in public pipeline databases would enable system-level diagnosis and comparison across sponsors and decades.
Hypothesis
Applying the taxonomy to a decade of discontinuations will show that at least a third of efficacy failures were preceded by inadequate target-validation or exposure evidence, quantifying an avoidable loss.
Rationale
Aviation and surgery improved by classifying failure. Pharma's 'five Rs' framework (right target, tissue, safety, patient, commercial) is a start that has not been applied openly.
What would test it
Curate 500 discontinued oncology programmes with two independent coders; publish inter-rater reliability and the distribution of causes.
Maturity
speculative
Who has to act
data
Cost to try
Small (under $1M)
Years to first evidence
2
Bottlenecks it attacks
- Failures are hidden · Negative trials, failed drugs and abandoned programmes are rarely published, so the same mistakes are repeated.
- The valley of death between lab and product · Most academic discoveries die before anyone tests them in people because nobody funds the middle step.