This study found that ceritinib is not cost-effective as a following treatment among NSCLC patients with ALK + who were previously treated with chemotherapy and crizotinib, from the US health payer perspective. To our best knowledge, this is the first study conducted in the US setting and considering ceritinib as a second-line treatment, instead of a first-line treatment.
This finding was robust to model choice as both Markov and PartSA models, demonstrated that ceritinib was not cost-effective when compared to chemotherapy in the study population. Both models were most sensitive to the same factors in the one-way sensitivity analysis.
The study finding is not consistent with previous studies. The Canadian study found ceritinib to be cost-effective among patients who have progressed or are intolerant to crizotinib, comparing with the willingness-to-pay threshold for end-of-life cancer drugs10. This result may not be comparable because of the heterogeneity in study samples, drug costs, parameters, and willingness-to-pay threshold. The Canadian study used a pooled patient population of ASCEND-1 and ASCEND-2 in patients previously treated with crizotinib, while we used ASCEND-5, the most recent appropriate phase 3 clinical trial. Costs for active treatments in the Canadian study also included associated costs of concomitant medications, including dexamethasone, NSAIDs, bisphosphonate, and morphine for chemotherapy, and folic acid associated with pemetrexed, while we did not consider these, because information on specific concomitant medications used in ASCEND-5 trial was unavailable. In both strategies there will be concomitant medication expenditures, so we are unable to anticipate if this will bias our findings in favor of one particular strategy. In addition, which adverse events were included in the Canadian study were not explicitly specified, so we are unable to predict how our selection of adverse events may affect the results. Further, as both arms in our study are from the same clinical trial, we were able to extract adverse events from a single study for both arms. In comparison, the Canadian study pulled data on the two strategies from different clinical trials which may bias that study in favor of one arm over the other if there are differences between the two. The Canadian study compared certinib to pemetrexed monotherapy, whereas we followed the ASCEND-5 trial and compared certinib to a combination of pemetrexed and docetaxel. And lastly, the Canadian study was conducted from the Canadian public healthcare perspective which has lower prices than in the US perspective used here.
Prior work has shown ceritinib was cost-effective as a first-line treatment among NSCLC patients in the US, compared with chemotherapy, using ASCEND-4 clinical trial data11. However, when ceritinib is used for a continuing treatment option after crizotinib, it is assumed that the patients have already experienced cancer progression or intolerant on crizotinib. Thus, the progression rate or progression free survival is different among patients using ceritinib as first-line treatment and patients with ceritinib as a following treatment after crizotinib. The comparator platinum doublet (with maintenance) in ASCEND-4 included pemetrexed in combination with cisplatin, or carboplatin followed by pemetrexed maintenance therapy. The chemotherapy combination is different from what we adapted from ASCEND-5 trial in our study as mentioned above. In addition, comparing the predicted PFS and OS curves for ceritinib and chemotherapy from ASCEND-4 trial adapted in the US study (Fig. 2 from Zhou et al., 201811), the proportions of patients remaining alive under ceritinib and platinum doublet treatments at 60 months from ASCEND-4 were higher than the proportion that we predicted from ASCEND-5 trial. This may be able to explain the difference in cost-effectiveness of ceritinib when it is used at different lines.
In addition, the PFS curves and OS curves in eFigure1 supported our findings. Even the PFS from ceritinib is longer than the PFS from chemotherapy (median time of 5.4 months from ceritinib vs. median time of 1.6 months from chemotherapy, the OS from ceritinib is significantly shorter than the OS from chemotherapy (median time of 11.3 months from ceritinib vs. median time of 53.1 months from chemotherapy). Thus, ceritinib may be cost-effective in short term, but it may lead to the opposite direction in long term.
One of our strengths is that two models for the same cost-effectiveness analysis were performed in this study. Ceritinib was not cost-effective in both models. And both models were most sensitive to the same parameters including medical cost for PD state, cost of ceritinib, and utility of PFS state in ceritinib arm in the one-way sensitivity analysis, even the calculated ICERs were different. Another strength is that the survival rates (i.e., progression free survival and overall survival) were extrapolated from clinical trials, instead of using transition probabilities from published literature, using survival rates from real-world clinical trials provided more reliability in our data inputs.
While ceritinib performs more poorly on ultimate survival than chemotherapy, patients on ceritinib spend less time on the progressed disease state, which is an expensive state to maintain. The PartSA model finds increased costs of certinib over chemotherapy (compared to the Markov model) since patients in the PartSA model spend more time in the progression free state, magnifying the additional costs patients accrue at that stage under ceritinib. Since the transition from PFS to death is not allowed in the PartSA model, while it is allowed in Markov model, more subjects transited out from PFS to other states in Markov model in each cycle. Though we note that both models found numerous iterations both above and below cost-effectiveness thresholds, so this difference may not be statistically significant.
There are some limitations in this study. First, we extracted survival rates from graphs published in the ASCEND-5 trial. Ideally our study would be based on individual patient data. This extract process may cause measurement error in the survival rates at different cancer stages. Second, the OS curves of ceritinib and chemotherapy were extracted from two different clinical trials, ASCEND-2 and PROFILE-1007. These two trials had different study population selection criteria. However, the efficacy observed in ASCEND-5 was consistent with the reported results in ASCEND-2, and ASCEND-2 and PROFILE-1007 have similar patient populations with progressed ALK + NSCLC, except for the previous treatment of crizotinib criteria. Median age in both trials was about 50 years, and gender and race distributions were similar. The PROFILE 1007 trial had similar median overall survival as was reported in the ASCEND-5 trial. The populations in these trials most closely matched our study population. Third, this study only included direct medical cost in the analysis, indirect costs such as cost due to loss of productivity and transportation were not considered. The estimated ICER may change accordingly when considering the indirect costs. Ceritinib is taken orally and may be taken at home without physician supervision and thus is accompanied by minimal travel costs. Chemotherapy is taken intravenously once per cycle (e.g. once per month) and thus requires regular travel costs to locations where the injections can be administered. This difference in where drugs can be administered may cause ceritinib to be more cost-effective, potentially reversing our findings if these travel costs and lost productivity costs are high enough. Fourth, although we used the most appropriated parameters to our study population, all model inputs in this study were not all based specifically on NSCLC patients previously treated with crizotinib and chemotherapy. Other factors, such as previous treatment options or numbers of lines of treatments, that may have impacts on utilities were not controlled for in our analysis.