Sotorasib for previously treated KRAS G12C mutation-positive advanced non-small-cell lung cancer

Non-small-cell lung cancer (NSCLC) is staged from 1A to 4B according to the size of the tumour, location of involved lymph nodes and the presence of distant metastases. NSCLC diagnosed as stage 3 (locally advanced) or stage 4 (metastatic) is advanced. People with locally advanced NSCLC commonly present with a cough. Other symptoms include shortness of breath, coughing up blood and pain. People with metastatic NSCLC may also have headaches, an enlarged liver, changes in mental health, weakness and seizures. KRAS is a protein that helps control normal cell growth and survival. KRAS is one of most frequently mutated genes in cancer, including lung cancer. KRAS mutations are linked with a poorer response to treatment and survival outcomes. The committee concluded that advanced NSCLC can substantially affect health-related quality of life.

The company positioned sotorasib as second- and third-line treatment for people with KRAS G12C mutation-positive advanced NSCLC whose cancer has progressed on, or who are intolerant to, platinum-based chemotherapy or anti-PD-1/PD-L1 immunotherapy, in line with the marketing authorisation. The company's clinical experts advised that an increasing majority of people in the NHS with NSCLC (without mutations for which there are targeted treatments) now have immunotherapy (alone or with platinum-based chemotherapy) as first-line treatment and few have immunotherapy at second line. Because of this, immunotherapies were not considered relevant comparators. The company explained that platinum-based chemotherapy is rarely used after first-line immunotherapy and so it was excluded as a comparator.

Sotorasib for previously treated KRAS G12C mutation-positive advanced NSCLC (NICE technology appraisal 781; TA781) included docetaxel and docetaxel plus nintedanib as comparators. The company's clinical experts and the EAG agreed that docetaxel plus nintedanib is no longer frequently used and that docetaxel is more commonly used alone. This was consistent with evidence from the UK Cancer Analysis System (CAS) database. The clinical experts acknowledged that there could be differences in treatments used across the country. But both confirmed that, in their practices, docetaxel plus nintedanib is used less frequently and that use was declining because of toxicities. So, the committee concluded that docetaxel was the only relevant comparator for this evaluation.

In TA781, the main clinical-effectiveness evidence for sotorasib came from the CodeBreak 100 trial, an ongoing phase 1 and 2, single-arm, open-label trial of sotorasib in people with KRAS G12C mutation-positive locally advanced or metastatic NSCLC. After being recommended through the Cancer Drugs Fund (CDF), new evidence was collected as part of the managed access agreement. The key clinical evidence in the current evaluation came from the CodeBreak 200 trial and the England-based Cancer Analysis System Real World Evidence (CAS RWE) study. All studies included people with NSCLC who had had at least 1 prior treatment.

The CodeBreak 200 trial is an ongoing phase 3, multicentre, open-label, randomised controlled trial comparing sotorasib with docetaxel. The study is being done at 149 centres globally including the UK. The trial recruited 171 people to the sotorasib arm and 174 people to the docetaxel arm. All participants were KRAS G12C mutation positive. The primary outcome was progression-free survival (PFS). Secondary endpoints include overall survival (OS) and health-related quality of life (HRQoL). In the August 2022 data cut, there was a statistically significant improvement for sotorasib compared with docetaxel for PFS (hazard ratio [HR] 0.66, 95% confidence interval [CI] 0.51 to 0.86). A protocol change resulted in a significant reduction in the sample size of the trial. As a result, the trial was no longer powered to detect a statistically significant difference in OS. After adjusting for all treatment switching (switching that happened as per the trial protocol and off-protocol), a post-hoc 2-stage adjustment analysis showed an OS HR of 0.82 (95% CI 0.32 to 1.31).

CAS RWE was a comparative, retrospective cohort study that used real-world evidence from the CAS database in England. CAS RWE included evidence from the Cancer Outcomes and Services Dataset (COSD), the Systemic Anti-Cancer Therapy (SACT) dataset, and the Molecular Diagnostics (MDx) dataset. The study included 394 people who had sotorasib and 1,271 people who had docetaxel. The primary objective of the study was to compare OS in people having sotorasib or docetaxel as second-line or later treatment. The secondary objective was to compare time to next treatment or death (TTNTD). Statistical analyses, including propensity score weighting and standardised mortality ratio weighting, were used to balance and weight baseline characteristics between treatment groups. After these adjustments, there was a statistically significant improvement in TTNTD (HR 0.567 [95% CI 0.49 to 0.66]) and OS (HR 0.63 [95% CI 0.54 to 0.75]). The committee concluded that the CodeBreak 200 trial suggested sotorasib improved PFS. It noted that the CAS RWE study suggested improvements in TTNTD and OS, but that there were some key uncertainties (see section 3.4).

The company used data from the non-randomised CAS RWE study to inform the clinical-effectiveness inputs (TTNTD [used as a proxy for PFS], OS, and time to treatment discontinuation or death [TTDD]) in the economic model. The company said that the CAS RWE data should be used in the model instead of CodeBreak 200 data because it had a larger sample size (n=757 compared with n=345), a longer follow-up (39 months compared with 24 months) and was based in England, whereas the CodeBreak 200 trial was done in 149 centres globally (including the UK). Because of protocol changes in CodeBreak 200, the trial was no longer powered to detect a statistically significant difference in OS (see section 3.3). During the trial, 59 people (34%) having docetaxel were allowed to switch to sotorasib instead. Although the company adjusted the OS results for treatment switching, it said it was concerned the results did not align with those from the CAS RWE study and the US-based Flatiron real-world evidence study. So it preferred to use the CAS RWE data.

The EAG noted these points but preferred to use evidence from CodeBreak 200 to inform the clinical-effectiveness inputs in the economic model. The EAG acknowledged that during recruitment to the CAS RWE study and through statistical and sensitivity analyses, efforts were made to ensure the populations having each treatment were similar. But the CAS RWE study was not randomised and so was at risk of selection bias and confounding that cannot be fully addressed by adjustments. During technical engagement, the company provided a risk-of-bias assessment (ROBINS-1) that showed the study was at serious risk of bias. The EAG noted that the KRAS mutation status of most of the people having docetaxel in CAS RWE was unknown, although the impact of mutation status on treatment effect is unclear. The proportions of people with known and unconfirmed mutation status are considered confidential and cannot be reported here. The mutation status of some people in the sotorasib group was also unconfirmed. But because a KRAS G12C mutation was a requirement for having sotorasib, the committee was reassured that most people having sotorasib had the mutation. Also, using CodeBreak 200 data enabled alignment of sources in the EAG's base-case economic model for baseline characteristics, OS, PFS and TTDD.

The committee considered the strengths and limitations of both studies. It noted that there was a lack of detailed information about the adjustments for treatment switching. So, the committee was uncertain if the adjustments were appropriate. The committee also noted that a lack of detail about the propensity score analysis, that was done on the CAS RWE data, meant that the committee was unable to thoroughly assess if this analysis was suitable or robust. The committee requested more detail about the company's approach to adjusting for treatment switching and about the propensity score analysis. The committee considered the differences in KRAS G12C mutation status between the docetaxel and sotorasib groups as a major limitation. The committee acknowledged the strengths of using the same evidence source to inform all clinical outcomes. In this case, it concluded that it would consider the most appropriate evidence source for each clinical efficacy input (see section 3.6, section 3.8 and section 3.10).

The company's model was in line with the model presented in TA781. The company used a partitioned survival model with 3 health states: progression-free, progressive disease and dead. The committee concluded that the model structure was acceptable for decision making.

The company used HRQoL data from the EQ-5D-5L questionnaire from CodeBreak 200. The results from the questionnaire were mapped onto the EQ-5D-3L to align with the NICE reference case. But, this was mapped using the van Hout mapping function, rather than the Hernandez-Alava mapping function, as stipulated in the NICE reference case. The company preferred using the van Hout mapping function because the Hernandez-Alava function produced higher utilities, which the company thought to be inflated. The company stated this was an artefact of the Hernandez-Alava function and said that this was consistent with findings reported in a study by Maervoet and Bergemann (2024). The EAG preferred to use the Hernandez-Alava function in its base case because this aligned with the NICE reference case. Using the Hernandez-Alava mapping function had a small impact on the incremental cost-effectiveness ratio (ICER). The committee concluded it preferred to use the Hernandez-Alava mapping function stipulated in the NICE reference case.

The company estimated utility values by descriptively analysing the HRQoL data by grouped time periods. This was used to inform its time-to-death (TTD) approach. The EAG preferred a progression-based approach to estimate utilities in line with the health states in the model. The company explored different mixed effects models for repeated measures (MMRM) to model utility. But the company did not provide an MMRM that included both progression and TTD covariates, as requested by the EAG. Instead, the company used a continuous TTD approach. Without this analysis, the EAG adopted the company's approach to estimating utilities in the base case, with the exception of using the Hernandez-Alava mapping function (as discussed in section 3.12). Utilities in the EAG's and company's base cases were capped at UK general population values, adjusted for age and sex.

The company stated that it had considered the EAG's request, but to implement the TTD approach fully, it would have needed to explore all covariates, which was complex. It also noted that this could not be incorporated in a cohort state model and that doing so would introduce more uncertainty than it resolved. The committee acknowledged the company's concerns but noted that providing the MMRM the EAG requested would still be useful to explore the interaction between treatment and progression status.

The company used different utility values depending on the treatment people had. The company explained that this was because people having sotorasib have fewer treatment-related adverse events and the administration method is less burdensome than for docetaxel. The EAG considered this was sufficiently justified. This was because the MMRMs that included the treatment covariate had better statistical fit. The committee agreed that it was appropriate to use treatment-dependent utilities while people were progression-free. But it noted that the company's approach to modelling utilities implied that the lower utility value associated with docetaxel was applied for the whole time horizon, even after people stopped having docetaxel. The clinical experts said that the side effects from sotorasib were usually quickly reversible, whereas the side effects from docetaxel may take weeks. But they explained that once the side effects had stopped, they would expect the utility for people who had had either treatment to be the same after disease progression. So the committee asked the company to provide the MMRM with both progression and TTD covariates to help with decision making. The committee also noted that if the TTD approach is used, it should be limited to 6 months before death. In the absence of this MMRM, the committee thought that treatment-independent utilities may also provide a useful estimate.

The company base-case cost-effectiveness estimates were within the range NICE considers a cost-effective use of NHS resources (£20,000 to £30,000 per QALY gained). The EAG's base-case estimates were above this range. The committee was unable to choose an acceptable ICER because further analyses were needed from the company.

The committee concluded that the company's overall model structure was acceptable for decision making (see section 3.5). But because of the many areas of uncertainty, the committee could not decide on its preferred assumptions and requested further analyses from the company (see section 3.18).

The committee noted the many areas of uncertainty. It would like to see the following additional analyses and information:

• more detail about the company's approach to adjusting for treatment switching and about propensity score analysis (see section 3.4)

• to model OS, applying the inverse crossover adjusted relative treatment effect from CodeBreak 200 to the baseline curves of sotorasib from CAS RWE to generate the docetaxel curves (see section 3.6)

• to model PFS, 2 approaches (see section 3.8):

  • using PFS from CodeBreak 200 to inform PFS directly

  • exploring the relationship between PFS and TTNTD from CodeBreak 200 and using this to adjust the TTNTD curve. This should be applied to TTNTD from CAS RWE to produce a better estimate of PFS in CAS RWE. The relative PFS efficacy from CodeBreak 200 should be applied to the sotorasib baseline curve to get the docetaxel curve.

• to model TTDD, extrapolating data from CodeBreak 200 (see section 3.10)

• to model utilities, providing an MMRM that includes both progression and TTD covariates. TTD should be limited to 6 months before death. Alternatively, using treatment-independent utilities may also provide a useful estimate (see section 3.13).

• using day case unit costs for the administration of docetaxel (see section 3.14).

The clinical-effectiveness evidence suggested that sotorasib improved key outcomes in people with KRAS G12C mutation-positive locally advanced or metastatic NSCLC whose disease has progressed on, or who cannot tolerate, platinum-based chemotherapy or anti-PD-1/PD-L1 immunotherapy. But there are some uncertainties in the clinical-effectiveness data and the economic model. The committee decided that, based on the analyses it had seen, it could not determine the most likely cost-effectiveness estimates for sotorasib. Given this uncertainty, the committee would like to see additional analyses (see section 3.18). So, sotorasib could not be recommended because there was not enough evidence to determine whether sotorasib is value for money in this population.