CodeBreaK or code blue? Assessing sotorasib's vital signs in metastatic colorectal cancer

The ability to therapeutically target Kirsten rat sarcoma viral oncogene homolog (KRAS)–long considered an undruggable oncogenic driver of nonsmall cell lung cancer (NSCLC) and gastrointestinal (GI) cancer—is one of the most important biomedical discoveries of the past decade. Therefore, sotorasib, the first KRAS glycine-to-cysteine substitution at codon 12 (G12C) mutation inhibitor to reach the clinic was heralded as a triumph of precision oncology.¹ Yet the path of sotorasib from preclinical discovery to US Food and Drug Administration (FDA) approval offers a cautionary tale for a field increasingly immersed in molecularly targeted therapies. Although heralded as a success of precision oncology, precision oncology is not merely the right drug for the right target. It also entails the right dose at the right time for the right duration, to which one could arguably add right reporting. However, after reading the results from the recent CodeBreak 300 trial of sotorasib in colorectal cancer (ClinicalTrials.gov identifier NCT05198934), we believe sotorasib has fallen short in many domains of this definition. The case of sotorasib also highlights how regulators have not encouraged the conditions needed to deliver the real promise of precision oncology to patients.

CodeBreaK 300 is a phase 3 trial evaluating the efficacy and safety of sotorasib in combination with panitumumab versus treatment of investigator's choice (TIC) in patients with advanced colorectal cancer harboring a KRAS G12C mutation who progressed on chemotherapies.² Although the included patient population and choice of control arm in this trial were appropriate, a striking feature of this trial is the uncertainty about the appropriate dose level for sotorasib. Doses for registration trials are typically established in early phase clinical trials to minimize the number of participants exposed to potentially ineffective or toxic dosages. Thus it is notable that this phase 3 trial appears to conduct a dose-finding exercise, with separate arms for the 240-mg and 960-mg doses, respectively. The justification of nonlinear pharmacokinetics, although not untrue, is perhaps only a part of the story.

Having shown promise in early phase trials, sotorasib received FDA Accelerated Approval for use in KRAS G12C mutant NSCLC in 2021. However, the FDA noted that preliminary trials were “hampered by a lack of robust dose exploration,” mandating further exploration in postmarketing studies³ and highlighting the tension between speed and certainty in drug development. Appropriately, the 960-mg and 240-mg dosages underwent further evaluation, with no statistically significant difference in efficacy but with higher treatment-related adverse events for the 960-mg dose.⁴ Although published in 2024, evidence from patent applications suggests that this was known to the sponsor as early as March 2022, before CodeBreaK 300 commenced enrolling.⁵ Although the trial appears to address these uncertainties, the hierarchical statistical approach—testing 960 mg versus TIC followed by 240 mg versus TIC—prohibits inferential comparisons between both doses of sotorasib, further clouding the dose-selection issue.

These design decisions have moral, clinical, and economic implications. Study investigators have a moral imperative to minimize research participants' exposure to harm: this is a common tenet of Good Clinical Practice. Although a trial designed to directly compare efficacy across both drug dose levels would need to be much larger—indeed, approximately twice the number of participants would be required—this robust dose exploration appears to be exactly what the FDA had previously asked for.³ The choice to spend participants' time, effort, and welfare, as Kimmelman suggests, on a noninformative dose comparison, in our view, is morally questionable—particularly considering suggestions that the sponsor possessed information unsettling equipoise about dosage.^{6, 7} If this information was not yet known, CodeBreak 300 should not have proceeded without adequately resolving uncertainties about optimal dosing.

Second, the chosen design raises concerns about the intent behind dose selection: by prioritizing the 960-mg dose for hierarchical testing, the trial may have supported regulatory decisions favoring higher cost dosing with no significant clinical benefit, greater potential harms, and higher economic burden for the health system. Finally, inflating enrolment to include a third arm incurred a significant opportunity cost, reducing the statistical power available to answer more clinically meaningful questions regarding overall survival (OS). Consequently, the trial was underpowered to detect differences in patient-centered outcomes despite enrolling a sizeable population of participants with this relatively infrequent mutation.⁸ A more patient-centric approach would have been either to do the hierarchical testing but in the opposite direction-that is, first test 240mg and then test 960mg.

The primary end point of CodeBreak 300 was met, with sotorasib 960 mg–panitumumab showing a statistically significant progression-free survival (PFS) advantage over the control arm (hazard ratio [HR], 0.48; 95% confidence interval [CI], 0.30–0.78; p = .005), although the 240-mg dose failed to do so (HR, 0.59; 95% CI, 0.37–0.95; p = .036; negative because of alpha splitting).^{9, 10} The higher dose also led to a higher overall response rate (ORR; 26.4% vs. 5.7%). Both of these PFS and ORR outcomes must have been factored into the FDA decision to approve the 960-mg dose. These comparisons contrast with prior dose-finding studies, which demonstrated more equivocal differences between the 240-mg and 960-mg doses.⁴ Notwithstanding these differences in ORR and PFS outcomes between the two doses versus the control arm—the two doses were not compared against each other for any outcome—they do not answer the questions that are most important to patients, clinicians, and regulators who must make challenging therapeutic decisions for a last-line therapy, which is whether sotorasib improves survival and quality of life.

In this trial, PFS was chosen as the primary end point, with OS and health-related quality of life relegated to secondary end points. CodeBreak 300 was not even powered to detect differences in OS between treatment arms. This is concerning for several reasons. First, OS and quality of life are consistently identified as being of primal importance to patients living with advanced solid tumors.^11-16 Second, PFS is not a valid surrogate measure of OS in second-line and later line colorectal cancer trials.^{17, 18} We recognize that there are situations in which intermediate end points like PFS may be justifiable, for example, in diseases with long survival or when prior surrogacy has been established. However, these conditions are not met in this last-line trial population.¹⁸ Usually, the appeal of PFS is shortening the time to results readout. However, for last-line therapy trials, OS results are usually available quickly, as in this case, I which OS results were available with just 6 months of additional follow-up after PFS results were available.² Notably, the control arms of this trial—trifluridine/tipiracil or regorafenib, as well as other recently approved drugs in this space, such as trifluridine/tipiracil plus bevacizumab or fruquitinib—were all approved based on OS improvement.^{19, 20} Indeed, no other drugs have been approved for last-line colorectal cancer based on PFS alone. The allure of the biology of targeting a novel mutation should not compensate for the need to improve OS in this group of patients. It is striking that, in CodeBreak 200, the last-line trial in lung cancer, the initial primary end point of OS was subsequently changed to PFS according to regulatory feedback to limit patients exposed in the control arm.²¹ This shows that even regulators are not protected from the allure of precision oncology.

Despite the initial promising results of this trial, the fundamental design flaws of CodeBreak 300 leave it fighting for life amidst the competitive landscape of precision oncology trials. However, it is bias in reporting and approval of the drug that ultimately triggers a code blue for sotorasib in advanced colorectal cancer. Despite what could be otherwise described as an exploratory analysis of survival outcomes, the conclusions of this study assert that sotorasib 960 mg and panitumumab constitutes a new standard of care for patients with KRAS G12C mutations. Speculation about potential survival gains from a nonstatistically significant, underpowered analysis should not be misrepresented as clinical insight: even if it is not statistically different, an improvement in OS could be particularly meaningful. Such statements do not carry evidentiary weight and risk misleading clinicians and patients alike.

We are quite surprised that the FDA accepted the trial with all of these flaws and approved the drug for this indication.²² Unlike the NSCLC indication, the colorectal cancer indication was granted regular approval—not even Accelerated Approval with a safety net of the requirement to confirm clinical benefit in the future. The FDA approved the 960-mg dose, although there were no statistical comparisons between the two dose levels—OS outcomes for both doses were similarly unimproved against the control arm, with HRs of 0.70 (95% CI, 0.41–1.18) and 0.83 (95% CI, 0.49–1.39), respectively, and p values of .2 and .5, respectively.² Granting full regular approval based on PFS for a last-line therapy trial in patients with a poor prognosis, despite negative OS, no quality-of-life improvement, and at a potentially higher dose with increased toxicities, is inexplicable. Sadly, after two large trials of this agent, the field—and, more importantly, our patients—are left with fleeting signals of efficacy, no survival benefit, and a cautionary reminder of the need to temper our excitement for innovation with the need for epistemic humility. Poorly designed or underpowered trials like CodeBreak 300 miss important opportunities to answer critical questions, at great human and economic cost. When regulators rubberstamp these trials with approvals, there is even less incentive to do things the right way.

To realize the full potential of precision oncology, we must move beyond molecular enthusiasm to methodological discipline. Only rigorously designed trials can separate promising ideas from proven benefit. Although there may yet be a place for sotorasib in the treatment of colorectal cancer, CodeBreak 300 does not provide sufficient evidence that its' time has arrived. The new FDA has signaled that they will require OS data for patients in later lines of therapy.^{23, 24} We hope that this stance results in positive change. Rewarding drugs that fail to improve survival and that seek registration at higher doses—despite the exciting biology and allure of precision oncology—will impede innovation that matters to patients.

Samuel X. Stevens: Conceptualization, writing–review and editing, and writing–original draft. Bishal Gyawali: Conceptualization and writing–review and editing.

Bishal Gyawali reports personal/consulting fees from Vivio Health outside the submitted work. Samuel X. Stevens disclosed no conflicts of interest.