In Singapore, targeted therapy with EGFR TKIs is a common first-line treatment for patients with EGFR-mutated advanced or metastatic NSCLC, which is also concordant with international guidelines. Osimertinib, a third generation EGFR TKI has been increasingly used as first-line treatment due to recent findings of increased clinical benefit and improved safety profile compared to previous generation EGFR TKIs. To our best knowledge, this is the first study to evaluate the cost-effectiveness of osimertinib in the first line treatment of advanced and/or metastatic NSCLC in the Singapore setting.
Our analysis showed that osimertinib is not a cost-effective treatment given current prices in the local context. The high ICER for osimertinib compared to standard EGFR TKI was largely driven by the high cost of treatment. The lifetime cost of treatment for each patient in the PF state for osimertinib was SG$207,440 compared to SG$48,039 for standard EGFR TKI. While treatment with osimertinib has resulted in a gain of 0.319 QALYs compared to standard EGFR TKI, the increase is unable to offset the high incremental treatment costs at a level likely to be considered cost-effective.
The large incremental difference in the cost per patient in the PF state was driven by both the increased differential drug cost (cost of osimertinib and standard EGFR TKI per 1-week cycle was SG$2042 and SG$633 respectively) and by the longer duration of the PF state associated with osimertinib compared with standard EGFR TKI (21.2 months versus 12.7 months). Scenario analyses investigating different pricing scenarios revealed that even a reduction of 50% in the cost of osimertinib only lowered the ICER to SG$162,013, which is unlikely to be considered cost effective in the Singapore setting.
Our results are consistent with other published cost-effectiveness (CE) studies in other countries comparing first-line treatment of osimertinib with standard EGFR TKI in EGFR-mutated advanced and/or metastatic NSCLC patients. Aguiar Jr et al reported ICERs/QALY of at least US$220,000 and US$162,000 in the United States (U.S) Medicare system and Brazilian private health system respectively [21]. Wu et al compared first-line osimertinib with first-generation EGFR TKIs (erlotinib and gefitinib) from a public payer’s perspective in the U.S. and China and showed ICERs/QALY of US$327,601 and US$43,034 respectively [22]. The substantially lower ICER obtained for China was due to the reduced healthcare costs compared to the U.S. A cost-effectiveness study comparing first-line osimertinib with gefitinib and afatinib from a public payer perspective in Canada obtained an ICER/QALY of $223,123 CAD [23]. Most of these CE studies, however, have used the OS and PFS curves from the interim analysis of FLAURA. Nonetheless, using the same interim data, our analysis generated a comparable ICER of SG$373,768 per QALY, which is about SG$45,000 lower than current base-case. A common finding among these published studies was that the high ICERs were due in large part to the relative high drug cost of osimertinib, which was in line with our study results. Our base-case leveraged on the updated final OS data, with extrapolation predicting a 5-year OS of 24% and 22% for the osimertinib and standard EGFR-TKI arms respectively. With the earlier data cut off, the extrapolated 5-year OS for osimertinib was similar but the survival for patients on standard EGFR-TKI was worse at 17% (see Fig. 6). In other words, projections from the more mature data suggest a convergence in the survival outcome between the standard EGFR-TKI arm and osimertinib arm over time, leading to a reduction in the magnitude of the OS benefits conferred by osimertinib. This was concordant to the observed KM data in the final data analyses for FLAURA which showed an appreciably higher hazard ratio of 0.80 (95% CI 0.64 to 1.00) compared to the interim dataset’s hazard ratio of 0.63 (95% CI 0.45 to 0.88).
By varying input parameters, OWSA was carried out to assess key drivers of the model. Our analysis showed that the model was most sensitive to the time horizon, albeit only to the lower variation of the parameter which was set at 5 years. Our base case time horizon was 10 years, which was consistent with other published CEAs [21–23] and also allowed for all QALYs to be fully captured. Shortening the time horizon to 5 years resulted in an increased ICER of SG$500,917.
Another key driver of the model was the utility values used in the PF state, which were extracted from published literature due to the absence of local utility data for advanced and/or metastatic NSCLC. Moreover, there was a dearth of published data on Asian-specific utility weights for the different health states during NSCLC disease progression. We thus adapted utility values from a multinational HRQoL survey on advanced NSCLC patients, and applied AE-associated disutility values sourced from a UK study. Although extracted from another published study [24], utility values in the PF state were also one of the key drivers in models of similar published cost-effectiveness studies [22] .
Scenario analyses showed that when using the joint fitting approach where the shape parameter for both treatment arms was jointly estimated, the resultant ICER/QALY was reduced to SG$337,380. The lower ICER value was primarily driven by the increased OS benefit from osimertinib treatment. The hazard ratio applied was assumed to hold true for the entire duration of the time horizon. Such an assumption contributes to higher uncertainty associated with the resultant ICER, especially in light of FLAURA’s final data set suggesting a converging trend in the OS benefits between osimertinib and the standard EGFR-TKI arms, challenging the constant proportional hazards assumption.
The results of our economic model must be interpreted within the context of our study limitations. Firstly, the published HRQoL studies where we extracted our utility data from were not representative of the Singapore population as the study populations were predominantly from western countries. The HRQoL study by Chouaid et al is a multinational study in hospitals across Europe, Canada, Australia, and Turkey [19], while Nafees et al published a UK HRQoL study which had 9% Asian representation [20]. Although Nafees et al later published an international study [24] on NSCLC health state utilities that included Asian countries such as China, Taiwan, and South Korea, the cultural differences from physician-patient interactions in those settings resulted in utility values that were not representative of our NSCLC population in Singapore.
Secondly, because we did not have access to individual patient data from the FLAURA trial, digitization of the published survival curves was utilised to reconstruct survival data. Extrapolation of the survival data was required to estimate OS and PFS curves over a 10-year time horizon. We had to examine the variability of extrapolated survival curves from candidate parametric functions, namely Weibull, exponential, log-normal, log-logistic, Gompertz, and generalised gamma distributions. For OS, we selected Weibull and Gompertz (for the osimertinib and standard EGFR TKI curves respectively) as our base case after assessing the parametric functions’ goodness-of-fit using AIC, visual inspection of the curves against actual empirical data, and local clinical expert opinion.
Thirdly, we did not incorporate cost data for significant AEs that did not require hospitalisation, which in this case were paronychia and rash that would require treatment in an outpatient setting. Although proportions of patients who had paronychia were similar across treatment arms, there were more patients in the standard EGFR TKI arm who reported having grade ≥ 3 rash in the FLAURA trial. Excluding the cost of dermatology consultations may have overestimated the ICER. However, given the low proportion of reported grade ≥ 3 rash and local clinical expert opinion that only half of those would seek outpatient treatment with a dermatologist, we expect the omission of these cost data to be negligible.
Further limitations stemmed from the nature of cost-effectiveness analysis, which carries inevitable assumptions of certain homogenous transition probabilities that deviate from the heterogeneous dynamics of the real world setting. In this context, we assumed that all patients who entered the model underwent tissue biopsy, all patients who progressed to second-line treatment in the standard EGFR TKI arm underwent T790M testing, and all patients who tested negative in T790M testing had undergone repeat tissue biopsy. In addition, we did not take into account biopsy complication rates, which we assumed to be uncommon. Our proposed treatment algorithm, although rigid, represented the most realistic treatment pathway for EGFR-mutated advanced NSCLC patients in the local clinical setting. Lastly, the analysis was carried out using data from the overall trial population and not the Asian subgroup, to which the results would have been more generalizable to Singapore’s context. The reason for this was that only the OS, and not the PFS curves, for the Asian subgroup were available from the published final data set. Hence for consistency, the model only utilised data from the entire trial population. The final OS curves for the Asian subgroup revealed no statistically significant survival benefits for osimertinib [hazard ratio of 1.00 (95% CI 0.75 to 1.32)][25] and would likely have increased the base-case ICER substantially owing to the expected lower QALY gains, if any, in comparison to the incremental QALYs estimated in the main population.