The Pill That Beat the Pills
In a phase 3 trial, divarasib cut the risk of death by 44 percent versus sotorasib or adagrasib in previously treated KRAS G12C lung cancer. Median survival was 17.7 months versus 12. It is not approved yet.
Divarasib kept people alive longer than the KRAS drugs already in use. It is not approved yet.
Divarasib is a pill for non-small cell lung cancer that carries a mutation called KRAS G12C. Two pills for that mutation are already approved: sotorasib and adagrasib. Krascendo 1 asked a ruder question than most cancer trials. It did not ask whether the new pill beats chemotherapy. It asked whether it beats the pills already on the shelf.
The phase 3 trial enrolled people whose cancer had already been treated and then grown. Three hundred and thirty-eight patients were randomised. One hundred and sixty-seven received divarasib. One hundred and seventy-one received a control pill: 156 of them sotorasib, 15 adagrasib. Median follow-up was 14.3 months.
People on divarasib lived longer. Median overall survival was 17.7 months, against 12.0 months on the older pills. The hazard ratio for death was 0.56. The 95 percent confidence interval ran from 0.41 to 0.78. The p-value was 0.0005. Roche described that as a 44 percent reduction in the risk of death.
The cancer also stayed quieter for longer. Median progression-free survival was 9.8 months versus 5.7. The hazard ratio was 0.58, confidence interval 0.44 to 0.76, p less than 0.0001. At 12 months, 44 percent on divarasib had not progressed, against 24 percent on control.
Of 328 patients who could be assessed for a response, a confirmed response came in 58.4 percent on divarasib and 32.3 percent on control. The median response lasted 11.9 months versus 6.9. Ten patients had no measurable disease at the start and were left out of that response set.
The side-effect table did not go the wrong way. Grade 3 or 4 treatment-related adverse events were 24.6 percent on divarasib and 34.5 percent on control. Reuters reported that 4.8 percent stopped divarasib because of a treatment-related event, against 11.7 percent on control. Genentech said no new safety signal appeared, and that the common treatment-related events were mostly low grade.
Genentech said the G12C mutation is in about one in eight people with this lung cancer. Ferdinandos Skoulidis of MD Anderson presented the data at ESMO in Madrid on 25 October 2026. The paper is in the New England Journal of Medicine the same day. The trial number is NCT06497556.
Levi Garraway, Roche's chief medical officer, said the company is talking to health authorities. Talking is not an approval. The patients in this trial had already been treated. A person hearing the diagnosis for the first time was not the person in this randomisation.
Yesterday's episode carried a melanoma vaccine that cut recurrence and could not yet show a survival gain. This is a different cancer, a different drug, and a survival result. It is not a sequel.
Key Facts
- Phase 3 Krascendo 1, NCT06497556, previously treated KRAS G12C non-small cell lung cancer
- 338 patients: 167 divarasib, 171 control (156 sotorasib, 15 adagrasib)
- Median follow-up 14.3 months
- Median overall survival 17.7 versus 12.0 months, hazard ratio 0.56, p=0.0005
- Median progression-free survival 9.8 versus 5.7 months, hazard ratio 0.58
- Confirmed response 58.4 percent versus 32.3 percent in the response-evaluable set
- Grade 3 or 4 treatment-related adverse events 24.6 percent versus 34.5 percent
- Presented 25 October 2026 at ESMO and published the same day in NEJM
Why This Matters
First-generation targeted pills often win by shrinking tumours and then lose the argument on survival. This one was tested against those pills, not against an older chemotherapy, and the survival curves separated. About six months at the median, in people who had already been through treatment, is the number a clinic can say out loud. It is still a median. It is not a promise to a named patient.
What We Don't Know Yet
Whether regulators approve it, on what line of therapy, and in which countries. Follow-up is 14.3 months, shorter than either survival median, so later deaths can move the estimate. Fifteen control patients received adagrasib. The comparison is mostly against sotorasib. It says nothing about people whose cancer does not carry G12C, and nothing about using divarasib first.