Matching-adjusted indirect comparison of amivantamab vs mobocertinib in platinum-pretreated EGFR Exon 20 insertion-mutated non-small-cell lung cancer

Approximately 10–15% of patients with non-small-cell lung cancer (NSCLC) in the USA and Europe have mutations in the EGFR gene, the most common actionable driver pathway event in NSCLC [1,2]. EGFR exon 20 insertion (exon20ins) is the third-most common type of mutation of the EGFR gene, accounting for approximately 1–10% of patients with EGFR-mutated NSCLC [1,3]. Tyrosine kinase inhibitors (TKI) are standard treatment for patients with EGFR-positive NSCLC [3], but patients with NSCLC and EGFR exon20ins are generally insensitive to treatment with standard EGFR TKI and have a 75% increased risk of death and a 93% increased risk of a progression-free survival (PFS) event versus patients with common EGFR mutations [4,5]. Other conventional therapies, such as cytotoxic chemotherapy and immuno-oncology agents, also have limited efficacy in this patient population [6], especially after first-line standard of care treatment which is currently platinum-based chemotherapy [7].

The 2023 National Comprehensive Cancer Network clinical practice guidelines for NSCLC recommend amivantamab injection as well as the novel oral EGFR TKI mobocertinib as second-line therapies for patients with EGFR exon20ins with disease progression after first-line chemotherapy or immunotherapy [8]. Amivantamab is the recommended and approved treatment for oncogene-addicted NSCLC in the 2023 ESMO guideline for oncogene-addicted metastatic non-small-cell lung cancer. Amivantamab is a first-in-class, fully humanized, EGFR x mesenchymal epithelial transition (MET), bi-specific antibody with immune cell-directing activity that targets activating and resistant EGFR and MET mutations and amplifications. In May 2021, it became the first treatment approved by the US FDA for the treatment of adults with locally advanced or metastatic NSCLC with EGFR exon20ins who experienced disease progression post-platinum-based chemotherapy [3], based on results of the multi-center, open-label, phase I CHRYSALIS trial (NCT02609776) [7]. In December 2021, amivantamab became the first treatment to receive conditional marketing authorization by the European Medicines Agency for adults with advanced NSCLC with EGFR exon20ins [9].

Mobocertinib is a first-in-class, potent, oral, irreversible TKI designed to selectively target in-frame EGFR ex20ins mutations in NSCLC [10]. It was approved by the FDA in September 2021 for the treatment of patients with metastatic NSCLC with EGFR exon20ins whose disease progressed post-platinum-based chemotherapy [11], based on results from a single-arm, phase I/II study (NCT02716116) [10]. While approved in the UK and Switzerland, mobocertinib is not approved in the European Union.

The relative efficacy and safety between amivantamab and mobocertinib is not well understood as it has not been previously investigated in a head-to-head study. In the absence of randomized clinical studies, an indirect treatment comparison can be used. In this analysis, we assess the relative efficacy and safety of amivantamab versus mobocertinib in patients with NSCLC with EGFR exon20ins mutations who had progressed on prior platinum-based chemotherapy via an indirect comparison of the best available evidence to date.

Methods

As both trials were single-arm studies, unanchored matching-adjusted indirect comparisons (MAIC) were performed using patient-level data from the CHRYSALIS study of amivantamab (NCT02609776) and aggregate data from the single-arm phase I/II mobocertinib trial (NCT02716116). Key aspects of the two trials, including inclusion/exclusion criteria, general study designs, outcome definitions, and baseline characteristics, were broadly comparable.

Unanchored MAICs require that all prognostic factors and effect modifiers that are imbalanced between trials at baseline are accounted for in the analysis. All baseline characteristics reported in the mobocertinib trial were compared with the same characteristics in the CHRYSALIS trial. Clinical experts were asked to identify which of the reported baseline characteristics were prognostic factors and/or effect modifiers and rank each in terms of their expected strength of association with the outcomes of interest (i.e., which variables were most important to adjust for in the analysis). Ten characteristics were deemed clinically important based on input from a panel of four clinicians and selected for matching in the base case analyses: number of prior therapies, Eastern Cooperative Oncology Group (ECOG) performance status, presence of brain metastasis, prior immunotherapy, prior TKI, age (both median and percent ≥65 years), gender, smoking history, median time from diagnosis to advanced disease in months and race. All baseline characteristics included in matching were reported for both trials.

Four additional covariates were considered in an MAIC conducted by Ou et al. (presence of bone metastasis, presence of liver metastasis, prior EGFR exon20ins-targeted therapy, and NSCLC histology) [12]. The same panel of four clinical experts confirmed that these covariates were not prognostic factors nor effect modifiers. However, we conducted sensitivity analyses also including these characteristics.

Study populations

The CHRYSALIS cohorts used in this analysis included patients with locally documented EGFR exon20ins who had progressed on or after prior platinum-based chemotherapy, as per Cohort D+ of the trial. The base case efficacy analysis was based on the supportive efficacy population from CHRYSALIS (n = 114) with a data cutoff of 30 March 2021. This cohort included all patients treated with amivantamab monotherapy at the recommended phase II dose (RP2D; i.e., 1050 mg for body weight <80 kg and 1400 mg for body weight >80 kg) that had ≥3 disease assessments as of June 2020 or discontinued/died prior to that date. The median follow-up of the cohort was 12.5 months. Sensitivity analyses of efficacy outcomes were performed based on the primary efficacy population (n = 81) from CHRYSALIS which formed the basis of the USA prescribing information, and included all patients treated with amivantamab monotherapy at RP2D that had ≥3 disease assessments as of February 2020 or discontinued/died prior to that date. The median follow-up of this cohort was 14.5 months. Patients in CHRYSALIS with missing data on at least one prognostic factor or effect modifier, were excluded.

The efficacy populations from the CHRYSALIS trial (n = 81 and n = 114; March 2021 data cutoff date) were compared with the Platinum Pretreated Patients (PPP) cohort of the mobocertinib trial (n = 114; November 2020 data cutoff date) in the analyses of relative efficacy outcomes. The PPP cohort consisted of patients with EGFR exon20ins-mutant metastatic NSCLC who received prior platinum-based therapy. All patients in the PPP cohort were treated with mobocertinib 160 mg once daily, including six patients who received the RP2D during dose escalation in part 1, 22 patients from the expansion cohort of part 2 and 86 patients from the EXCLAIM extension cohort who were previously treated with platinum therapy from part 3. The median follow-up for the PPP cohort was 14.2 months.

The safety analyses were based on the CHRYSALIS safety population (n = 153; median follow-up 9.9 months), which included patients previously treated with platinum chemotherapy who had received the RP2D as of the March 2021 clinical cutoff, versus the PPP cohort (n = 114; median follow-up 14.2 months). Since the risk of AEs may depend on the follow-up duration, sensitivity analyses were conducted using an additional safety population from CHRYSALIS (n = 81) with longer follow-up (median follow-up of 14.5 months).

Outcome measures

The efficacy outcomes included in the MAIC were: ORR assessed by IRC (ORR-IRC) and investigator (ORR-INV), clinical benefit rate (CBR) assessed by IRC (CBR-IRC) and INV (CBR-INV), PFS assessed by IRC (PFS-IRC) and overall survival (OS). Duration of response (DOR) was not assessed, as matching on the responder subpopulation was not possible due to a lack of published data for mobocertinib. The outcome definitions were comparable between studies, with the exception of CBR. Definitions of, ORR, PFS and OS are reported in the Supplementary Materials. In general, CBR was defined as the percentage of patients achieving confirmed complete or partial response, or stable disease per RECIST v1.1. In the CHRYSALIS trial, CBR included sustained stable disease of at least 11 weeks; however, just 6 weeks of sustained stable disease were required for CBR in the mobocertinib trial. A post hoc version of CBR was derived using the CHRYSALIS trial data to align with the mobocertinib trial outcome definition, defined as the proportion of patients with best overall response of complete or partial response or stable disease of at least 6 weeks. The definition of safety outcomes were comparable across trials (Supplementary Materials).

Statistical analysis

The MAIC balances average baseline differences between two populations by reweighting individual patient-level data from one trial to match the reported summary baseline characteristics of the comparator trial. The weights are in essence propensity scores, estimated by the method of moments.

After population-adjustment, the average (marginal) treatment effect in the comparator population is estimated by comparing the estimated outcome for amivantamab in the weighted CHRYSALIS population with the observed outcome for mobocertinib in NCT02716116. Reconstructed individual patient-level data were derived from published results for mobocertinib, by replicating an individual-level dataset for ORR and CBR and by simulating data for PFS and OS from digitally scanned published Kaplan–Meier curves using the validated algorithm published by Guyot et al. [13].

Relative efficacy and safety for amivantamab versus mobocertinib on binary outcomes were estimated using a weighted logistic regression model. Rather than presenting the odds ratios which are difficult to interpret, we report relative risks (RR), defined as the ratio of the risks for an event for amivantamab versus mobocertinib, with 95% CI that are derived from this model. All hypothesis tests were conducted at the 5% significant level.

Cox proportional hazards regression models were used to estimate the adjusted relative efficacy of time-to-event outcomes (i.e., PFS and OS) based on the MAIC-weighted amivantamab data and the mobocertinib reconstructed individual patient-level data. For both binary and time-to-event analyses, the robust sandwich estimator was used to estimate the variability associated with the adjusted relative efficacies.

Results

Population-adjustment

After population-adjustment, the average baseline characteristics of the amivantamab cohort matched the reported average baseline characteristics of the NCT02716116 population (Table 1). Matching was carried out on all variables in Table 1 and was successful (i.e., patient characteristics after weighting matched those reported in the PPP cohort). Ten of the 114 patients in the efficacy population and 24 of the 153 patients in the safety population were excluded from the MAIC analysis because they had missing data for at least one of the baseline characteristics included in the adjustment (race and exposure to prior TKI therapy). Additionally, matching reduced the effective sample size (ESS) of the CHRYSALIS efficacy population by approximately 27% relative to the unadjusted sample size (ESS = 83); the CHRYSALIS safety population was reduced by 36% following reweighting (ESS = 98). The effective sample sizes for the sensitivity analyses are reported in the Supplementary Materials.

Efficacy

Observed IRC-assessed response rates were higher for amivantamab compared with mobocertinib (43.0 vs 28.1%; Table 2). Patients treated with amivantamab were 1.6-times more likely to respond to treatment versus mobocertinib after population-adjustment (RR: 1.64; 95% CI: 1.12–2.39; p = 0.012). ORR-INV (Table 2), CBR-IRC and CBR-INV (Table 3) did not significantly differ between treatments both before and after population-adjustment. Results of the sensitivity analyses were consistent with that of the primary efficacy analyses (Supplementary Table 1 & Supplementary Figures 1 & 2).

Amivantamab and mobocertinib had similar efficacy for time-to-event outcomes in the base case analysis. The observed median PFS was 6.7 (95% CI: 5.5–9.7) months for amivantamab and after population-adjustment 6.5 (95% CI: 5.4–8.3) months versus 7.3 (95% CI: 5.5–9.3) months for mobocertinib (Figure 1); the adjusted hazard ratio on PFS for amivantamab versus mobocertinib was not statistically significant (1.24; 95% CI: 0.87–1.77; p = 0.244).

The observed median OS was 22.8 (95% CI: 17.5-not reached) months for amivantamab; following population-adjustment the median OS was similar at 23.0 (95% CI: 17.5-not reached) months versus 24.0 (95% CI: 14.6–28.8) months for mobocertinib (Figure 2). The adjusted hazard ratio (HR) for OS was not statistically different between treatments (HR: 0.91; 95% CI: 0.57–1.46; p = 0.698).

Safety

Adjusted safety outcomes are presented in Figure 3, and results of naive comparisons are available in the Supplementary Materials. For nearly all patients in both treatment cohorts, at least one treatment-related adverse event (TRAE) was reported (98 vs 99%; RR: 0.99; 95% CI: 0.96–1.02). In adjusted analyses, the risk of serious AEs (any grade) was lower for amivantamab versus mobocertinib, although not statistically significant (RR: 0.73; 95% CI: 0.53–1.02; p = 0.060). Patients treated with amivantamab were also numerically less likely to experience treatment-emergent AEs (TEAE) of any grade leading to dose reduction versus mobocertinib (RR: 0.57; 95% CI: 0.32–1.03; p = 0.054). Grade ≥3 serious TEAEs were significantly less common with amivantamab compared with mobocertinib (RR: 0.61; 95% CI: 0.41–0.90; p = 0.010).

Of the 23 any-grade TRAEs reported in the mobocertinib study, 19 appeared less common with amivantamab (Figure 3A). 15 of these 19 TRAEs were statistically significantly less commonly reported in the amivantamab study. The largest differences between treatments were estimated for diarrhea (RR: 0.12; 95% CI: 0.06–0.22), decreased appetite (RR: 0.20; 95% CI: 0.09–0.43) and vomiting (RR: 0.23; 95% CI: 0.11–0.50; p ≤ 0.001 for all). Dermatitis (RR: 2.61; 95% CI: 1.68–4.05) and increase in alanine transaminase (RR: 2.17; 95% CI: 1.01–4.64) were the only any-grade TRAEs which occurred significantly more often with amivantamab compared with mobocertinib (p < 0.001 and p = 0.043, respectively).

Of the 18 grade ≥3 TRAEs considered in the MAIC, 15 were less commonly experienced with amivantamab (Figure 3B). Two of the grade ≥3 TRAEs (i.e., diarrhea and lipase elevation) were significantly less common with amivantamab than with mobocertinib in the adjusted indirect comparison; diarrhea was 94% less likely in those treated with amivantamab compared with mobocertinib and elevated lipase was 89% less likely (RR: 0.06; 95% CI: 0.01–0.25; p < 0.001 and RR: 0.11; 95% CI: 0.01–0.93; p = 0.040). Grade ≥3 treatment-related rash, paronychia and dermatitis were more common with amivantamab than with mobocertinib but these differences were not statistically significant. Grade ≥3 TRAEs and grade ≥3 TEAEs were also significantly less frequent with amivantamab, with adjusted RRs of 0.40 (95% CI: 0.26–0.63; p < 0.001) and 0.70 (95% CI: 0.55–0.89; p = 0.002), respectively (Figure 3B). Results of the sensitivity analyses were consistent with that of the primary safety analyses (Supplementary Figures 3 & 4).

Discussion

EGFR exon20ins mutations are historically associated with a poor prognosis due to inherent resistance to EGFR TKIs; however, amivantamab and mobocertinib improve response rates compared with other historical second-line therapies [14]. In the absence of head-to-head randomized trials, comparative effectiveness data comparing treatments provides evidence to help inform therapeutic decision-making for healthcare providers, health technology assessment bodies, payers and other stakeholders. In the current analysis of the primary efficacy population, ORR-IRC was significantly higher for amivantamab. However, analyses of ORR-INV were not significant across all analyses, with findings showing comparable efficacy for amivantamab versus mobocertinib. IRC assessment may be a stronger outcome measure over investigator assessment as more strict criteria are used in a centralized assessment. Both treatments had similar efficacy for all other end points evaluated in the MAIC analyses. Other next-generation EGFR exon20ins-targeted therapies such as sunvozertinib and CLN-081 have shown promising response rates in preliminary results from ongoing clinical trials, but were not included in the scope of this analysis [15–17]; both agents have recently been granted breakthrough therapy status by the FDA.

Amivantamab had a more favorable safety profile versus mobocertinib for most any-grade TRAE outcomes (19 of 23), with the largest differences observed for diarrhea, decreased appetite and vomiting, which resulted in the fewer dose reductions due to AEs in the amivantamab group. Results from the EXCLAIM extension cohort of the mobocertinib trial, showed significant worsening from baseline in EORTC QLQ-C30 symptom scales for both diarrhea (p < 0.001) and appetite loss (p = 0.004). Differences in the rates of these TRAEs, are expected to have a significant impact on health-related quality of life in patients with advanced/metastatic NSCLC [18].

Amivantamab was also associated with a significantly lower risk of overall grade ≥3 TEAEs, grade ≥3 TRAEs and grade ≥3 serious AEs. Risks of any-grade AEs and any-grade serious AEs were comparable between treatments. In comparison with amivantamab, mobocertinib was associated with a significantly lower risk of any-grade dermatitis and increase in alanine transaminase. Although the CHRYSALIS safety population provided shorter median follow-up versus the PPP cohort, the comparative safety results were consistent in a sensitivity analysis evaluating the CHRYSALIS primary efficacy population (median follow-up 14.5 months) versus PPP cohort (median follow-up 14.2 months) over a comparable timeframe. While prescribing information for mobocertinib includes a warning for heart rate-corrected QT prolongation and Torsades de Pointes [10], cardiovascular effects could not be assessed in the safety analysis of this MAIC as patients with cardiovascular comorbidities were excluded from both trials.

A recent MAIC analysis by Ou et al. conducted in 2022 evaluating mobocertinib versus amivantamab between treatments in patients with EGFR ex20ins-mutant NSCLC who have progressed during or after platinum-based chemotherapy [12]. This indirect treatment comparison (ITC) was based on aggregate-level results of an earlier data cut from the CHRYSALIS trial, which led to a more limited sample size and follow-up period as compared with the current work. Ou et al. reported longer DOR with mobocertinib versus amivantamab, although the difference between treatments was not statistically significant [12]. However, any population-adjusted indirect comparisons of DOR are limited by a lack of baseline data from the mobocertinib and amivantamab responder subpopulation, which prevents adjustment for potential differences in responder subgroups between the two trials in both our analysis and that of the Ou et al MAIC. Ou et al. concluded that the efficacy of mobcertininb and amivantamab were similar, but did not consider any of the safety end points investigated in the current study. In this analysis TRAEs, TEAEs, serious TRAEs and TEAEs leading to dose reduction were assessed, enabling a more complete comparison but they did not consider any safety outcomes which have been shown in this analysis to be favorable for amivantamab [12].

The current analysis is strengthened by evaluation of additional efficacy end points: CBR-IRC and CBR-INV. Further analyses were performed to adjust for additional prognostic factors reported in the Ou et al. MAIC which resulted in significant reductions in ESS, with findings similar to the base case. The MAIC presented herein investigated the latest data available and conducted a sensitivity analysis adjusting for a wider set of prognostic factors than was done in Ou et al. Additionally, more granular data in population adjustments for race and histology were used (i.e., matching on all reported categories for each variable). Taking into account the set of adjustment variables considered in the MAIC by Ou et al. updates were made to the previously published version of this MAIC analysis, summarized by Van Sanden et al. [19]. The results of the revised MAIC presented herein were comparable to the original analysis by Van Sanden et al. and the validity of the findings was confirmed when accounting for additional prognostic factors and effect modifiers.

The comparative safety results estimated via the MAIC are expected to translate into long-term economic benefits for amivantamab, which would be of particular relevance in cost–effectiveness evaluations considered by reimbursement bodies. In a recent analysis of US costing data, the costs of managing grade 3 or 4 AEs were lower among patients who received amivantamab versus those who received mobocertinib [8]. Commercial costs of AE management per patient per treatment course were US$4158 with mobocertinib versus $1712 with amivantamab, and commercial costs of diarrhea events were US$2072 with mobocertinib versus $292 with amivantamab. Comparative analyses on safety outcomes in the current analysis showed that amivantamab had a more favorable safety profile than mobocertinib for the majority of AE outcomes, and thus a reduction in predicted resource utilization and cost savings is anticipated.

Limitations

The MAIC method cannot adjust for trial design differences that may affect the study outcomes, such as the timing of tumor assessments. The schedule of disease assessments differed between the two trials, which may result in assessment time bias for unanchored comparisons of PFS. In CHRYSALIS, progression-free status was assessed every 6 weeks; conversely, in NCT02716116, disease assessments were conducted every 8 weeks for the first 56 weeks, and every 12 weeks thereafter. As patients in the CHRYSALIS trial were assessed for progression earlier and on a more frequent basis compared with NCT02716116, the adjusted PFS HRs presented herein are likely conservative, and it's expected that matching the schedule of assessments may reduce the assessment time bias (i.e., current comparison is likely biased in favor of mobocertininb). Both studies assessed patients for progression at 24 weeks (5.5 months) and 48 weeks (11 months), and visual inspection of the Kaplan–Meier curves at these two time points indicates a very similar PFS for both amivantamab and mobocertinib.

Reduction in the effective sample size for the base case analysis was moderate, though matching on additional factors in the sensitivity analysis resulted in a further reduction in ESS. Despite this, no patients were associated with extreme MAIC weights. Furthermore, the findings of the MAIC based on the supportive and updated efficacy populations were consistent in terms of the overall conclusions. Similarly, safety analyses based on the two populations in CHRYSALIS were consistent. All of the above elements confirm the robustness of our analyses.

Analyses of safety outcomes were limited to those reported for mobocertinib, and therefore infusion-related reactions (IRR), which was the second most commonly reported TRAE in CHRYSALIS (any grade observed in 66% of patients, and grade ≥3 in 3%) could not be analyzed using MAIC. All other commonly reported TRAEs in CHRYSALIS, including rash, paronychia, stomatitis and pruritus, could be analyzed. Early results from the ongoing PALOMA trial (NCT04606381), showed that subcutaneous administration of amivantamab in patients with advanced NSCLC (n = 83) was associated with a meaningful reduction in the incidence and severity of IRRs compared with the intravenous formulation (16 vs 67%) [20].

As in any non-randomized comparison, residual confounding cannot be ruled out entirely. For example, patient characteristics for specific position of the insertion mutation were unavailable and thus unable to be matched across the two trials. However, commonly available baseline characteristics allowed us to adjust for the most important clinical prognostic factors, which minimized the risk of biased comparisons to the furthest extent possible.

Conclusion

Unanchored MAIC analyses according to patient-level data from the CHRYSALIS study and aggregate data from NCT02716116 found that amivantamab significantly improved ORR-IRC versus mobocertinib in EGFR exon20ins NSCLC. No significant differences were identified between treatments for the other efficacy end points considered in the MAIC analyses, including ORR-INV, CBR-IRC, CBR-INV, PFS-IRC and OS. Additionally, the results of the MAIC analyses indicate that despite being administered intravenously, amivantamab offers a better safety profile in comparison with oral mobocertinib.

Overall, amivantamab was predicted to have a better safety profile compared with mobocertinib, which indicates a long-term economic advantage for amivantamab. Amivantamab and mobocertinib represent key advances in the treatment of an underrecognized patient population in NSCLC, and future research is ongoing to evaluate these therapies in the first line and in combination with other regimens. In the absence of head-to-head trials, indirect treatment comparison can be used to determine the relative efficacy and safety between two treatments. Results of this analysis confirm and build upon the previously reported comparative efficacy estimates [12], while also highlighting the comparative safety between amivantamab and mobocertinib.