In the summer of 2021, Ferdinandos Skoulidis of MD Anderson Cancer Center was lead author on the New England Journal of Medicine paper that introduced the field to sotorasib, the first drug aimed at a KRAS mutation long written off as undruggable. The FDA had granted it accelerated approval in May 2021, and Amgen sold it as Lumakras.
- 2021Accelerated approval on a single-arm trial
- 2023Advisers doubt confirmatory PFS data; full approval declined
- 2028Deadline for a new confirmatory study
This month, at the Presidential Symposium of the 2026 ESMO Congress in Madrid, Skoulidis presented the trial in which a Roche pill beat a control arm built almost entirely around it. At the first interim look at survival, patients randomized to Roche’s divarasib had lived a median 17.7 months, against 12.0 months for patients randomized to the older KRAS G12C drugs, nine in ten of whom were taking Lumakras. Skoulidis concluded that divarasib “represents a new standard of care” for previously treated patients with the mutation.
The trial, Krascendo 1, randomized 338 people with KRAS G12C-mutated non-small cell lung cancer who had already been through platinum chemotherapy and a PD-1 or PD-L1 immunotherapy. Half got divarasib, and the other half got whichever approved KRAS G12C inhibitor their doctor picked. The doctors overwhelmingly picked sotorasib: according to the full readout, 156 of the 171 control patients took it and 15 took adagrasib, the drug Bristol Myers Squibb acquired with Mirati. On paper the comparison was against two drugs, but in practice it was a fight with Lumakras.
The primary endpoint, progression-free survival judged by blinded central reviewers, ran a median 9.8 months on divarasib and 5.7 months in the control arm, a hazard ratio of 0.58 (95% confidence interval 0.44 to 0.76). Overall survival, next in the testing hierarchy, came in at a hazard ratio of 0.56 (95% confidence interval 0.41 to 0.78, p=0.0005), and at one year 67.1 percent of divarasib patients were alive against 49.2 percent of controls, roughly 18 more people per 100. By the May 19 data cutoff, 63 patients had died on divarasib and 93 in the control arm.
| Result | Value (HR) | Lower bound | Upper bound |
|---|---|---|---|
| Overall survival | 0.56 | 0.41 | 0.78 |
| Progression-free survival | 0.58 | 0.44 | 0.76 |
Median follow-up was 14.3 months, which is short enough that the confidence interval around divarasib’s median survival still runs from 16.3 months to a number nobody can yet estimate. The trial was also open-label. Death is about as hard to talk yourself into as an endpoint gets, and because crossover was barred, no control patient could switch to divarasib, though whatever treatments patients received after leaving the study can still tilt a survival curve and will need checking when the full paper appears.
Tumors responded more often, too. The confirmed response rate was 58.4 percent against 32.3 percent, and responses lasted a median 11.9 months against 6.9. Complete responses ran 4 percent in both arms, which is worth remembering when the marketing begins: divarasib shrinks more tumors and holds them back longer without making many more of them disappear.
Five years on a response rate
Lumakras itself reached the market with nothing randomized beside it. Skoulidis’s 2021 study was a single-arm trial of 126 previously treated patients, with a 37.1 percent response rate and a median survival of 12.5 months. Its confirmatory trial, CodeBreaK 200, finally put it against docetaxel and produced a median progression-free survival of 5.6 months against 4.5, along with no significant difference in overall survival: 10.6 months on sotorasib, 11.3 on chemotherapy.
In October 2023 the FDA’s Oncologic Drugs Advisory Committee voted 10 to 2 that even the progression-free result could not be reliably interpreted, after reviewers flagged heavier dropout in the docetaxel arm and patients crossing over to sotorasib before their progression had been assessed. “I had to vote no on the reliability of the PFS benefit from this study,” Ravi Madan said at the meeting. That December the agency declined to grant full approval, left the accelerated approval standing, and required Amgen to complete a new confirmatory study by February 2028.
Adagrasib, the other option on the control side, beat docetaxel in its phase 3 trial, KRYSTAL-12, on progression-free survival, 5.5 months against 3.8, though its survival data were not yet mature at that analysis.
So for nine control patients in ten, Krascendo 1 measured divarasib against a drug that had been on sale for five years without a randomized trial showing it extended survival beyond what chemotherapy offered. Divarasib beat that arm on survival. Lumakras, in its own confirmatory trial against docetaxel, did not.
Roche earns credit for picking an active targeted drug as its opponent instead of running another round against docetaxel, and for winning on the endpoint patients care about most. The result also reads as an after-the-fact verdict on how Lumakras was handled: a drug admitted on a response rate, a confirmatory trial whose progression data the agency’s own advisers voted they could not reliably interpret, and then a new deadline rather than a withdrawal. This randomized survival win came from Roche’s trial against Amgen’s drug, not from Amgen’s trial against chemotherapy.
Divarasib is not gentle on the gut. In the safety data presented in Madrid, diarrhea hit 66 percent of patients against 38 percent on the older drugs, nausea 53 percent against 20, and vomiting 39 percent against 8, mostly at grade 1. On the heavier categories the newer drug came out ahead, with grade 3 or 4 treatment-related adverse events in 24.6 percent against 34.5 percent, serious treatment-related events in 7.8 percent against 15.2, and discontinuations for treatment-related side effects in 4.8 percent against 11.7. ALT elevations, a familiar problem with sotorasib, ran 15 percent against 23, and neither arm recorded a treatment-related death.
Patients also stayed on divarasib for a median 11.3 months against 5.0 on the comparators, so those gastrointestinal rates belong to a drug people were taking for most of a year. The trial was designed to measure what that felt like, with time to confirmed deterioration in patient-reported outcomes as its last formal endpoint, and the published accounts of the Madrid presentation give no result for it. For a patient weighing a longer median survival against daily nausea, that is the missing number.
In its July announcement of the topline result, Roche said it would submit the data to health authorities, and noted that divarasib already holds an FDA Breakthrough Therapy designation from 2022 and an Orphan Drug designation for KRAS G12C lung cancer from 2026. Chief medical officer Levi Garraway said the results “should establish divarasib as a new standard of care.” The company puts the G12C mutation at about 14 percent of non-small cell lung cancers, in a disease that kills about 1.8 million people a year.
The FDA gave Amgen until February 2028 to finish its new confirmatory study of Lumakras. Sixteen months ahead of that deadline, a competitor that owed the agency nothing on the subject has put a randomized survival result beside the drug.
Sources
- Targeted Oncology: Krascendo 1 results at the ESMO 2026 Presidential Symposium
- OncoDaily: Krascendo 1 phase III efficacy and safety readout
- Roche: Krascendo 1 topline announcement, July 2, 2026
- Targeted Oncology: FDA rejects full approval of sotorasib after CodeBreaK 200
- CancerNetwork: FDA issues complete response letter for sotorasib
- The ASCO Post: KRYSTAL-12, adagrasib vs docetaxel
- The ASCO Post: sotorasib single-arm trial survival analysis (Skoulidis, 2021)