Overview
Programmes fail for structurally different reasons, and which one it was decides what to do next. A model that mispredicts, a drug that never arrives, an endpoint nobody accepts — these are not variations on bad luck, and they call for entirely different responses. Five diseases, five distinct failure modes.
Parkinson's disease
Ten disease-modifying programmes cleared their animal model, and every one failed in patients. The models predict symptom relief well and disease modification not at all — and the disease still has no accepted way to measure whether a drug is slowing it.
Read the pageMyotonic dystrophy (DM1)
The cause was settled in 1992 and has not been in serious doubt since. The first programme stopped because the drug could not get into muscle in useful quantity — the hypothesis was never actually tested. A delivery problem, diagnosed precisely, and now apparently engineered around.
Read the pageRheumatoid arthritis
Induced in the wrong species by the wrong trigger, with none of the autoimmunity that defines the human disease — and seven independent mechanisms reached approval regardless. The counterexample that bounds every other page here: construct validity was not what decided the outcome.
Read the pageCrohn's strictures
Every approved Crohn’s drug targets inflammation; the scarring it leaves behind still sends patients to surgery. On every available instrument, inflamed bowel and scarred bowel both look thick — so no trial can demonstrate that a drug worked, whatever it is doing.
Read the pageFSHD
A trial took a molecular surrogate for the causal gene as its primary endpoint and missed it, while the functional measures moved in the drug’s favour. The next trial made function the primary — and missed that too. Both readouts, in opposite directions, then nothing.
Read the pageThese five are a sample chosen for contrast, not a ranking. The full list of indications covers several more, including two cases where one disease is solved in one organ and unsolved in another.