Charles Jiang, Medical Oncologist at UT Southwestern Medical Center, shared a post on X:
“Run the same adjuvant trial twice, in two different groups of patients, and you can produce one celebrated drug and one failure. The drug does not change. The recurrence rate of the people you enrolled does. A thread on why the P value is the least useful number on the slide.
Take a drug that cuts recurrence by 30%.
- Give it where 10 in 100 recur. You prevent 3. You treat 33 people for one to gain.
- Give it where 50 in 100 recur. You prevent 15. Now you treat 7. Same drug, same 30%.
The second trial gets the standing ovation.
Oncology has been doing this for fifty years. Stage came first. Then nodes, grade, receptor status, RX scores, nomograms. ctDNA is the newest one. All of them do one job. Fill the trial with people who will have events. The marker changes every decade. The idea is same.
A P value reports the population you chose as faithfully as it reports the drug you tested.
Now ask who in the treated arm actually gained. Three groups sit in every adjuvant trial. Surgery has already cured some. Some will recur whatever you give them. A small third group is cured by the drug. Only that third group gains, and nobody can have critical ball.
So price the win. To prevent 15 you treated 100 . Eighty five were never going to benefit and paid anyway. The neuropathy that does not resolve. Thyroid replacement for life. Six months of infusion chairs. The copays. The job.
Number needed to treat reaches the slide sometimes. Number needed to harm almost never does. Last question, and the one that gets skipped. Did the drug remove the recurrence or postpone it?
Curves that split early and rejoin by year five bought time. Curves that stay apart cured someone. A three year readout cannot tell you which one you are holding. The asterisk says the effect is real. Every question worth asking starts after that.”
Other articles featuring Charles Jiang on oncoDaily.