A systematic review of the cost-effectiveness of targeted therapies for metastatic non-small cell lung cancer (NSCLC)

Eleven studies analyzed the cost-effectiveness of the treatment with erlotinib compared to BSC or chemotherapy [12‐22]: One study focused on first-line treatment [16], three on first-line maintenance treatment [13‐15] and seven studies on subsequent treatment lines [12, 17‐22]. Wang et al. [16] analyzed the cost-effectiveness of first-line erlotinib monotherapy compared to carboplatin-gemcitabine combination therapy in patients with advanced EGFR mutation-positive NSCLC. They conclude that erlotinib is cost-effective from the perspective of the Chinese health care system. However, their ICER calculation is based on differences in progression free survival (PFS). Vergnenegre et al. [13] estimated the cost-effectiveness of first-line maintenance treatment with erlotinib vs. BSC in a German, France and Italian setting. The authors conclude that erlotinib is cost-effective in first-line maintenance therapy at a time horizon of five years. Restricting patients to those with EGFR wild type, Walleser et al. [14] also assessed the cost-effectiveness of first-line maintenance treatment with erlotinib over a lifetime horizon. They deduce cost-effectiveness of erlotinib for the country settings of UK, Germany, France and Italy. Klein et al. [15] focused on the cost-effectiveness of first-line maintenance therapy with pemetrexed from a US payer perspective; however, the authors have also considered erlotinib in their calculations. They showed that the treatment with erlotinib causes lower costs and lower efficacy when compared to pemetrexed at a time horizon of three years. These results applied to patients with nonsquamous and patients with squamous and nonsquamous histology. Therefore, no ICER was calculated. Furthermore, the reported values indicate that erlotinib is hardly cost-effective when compared to BSC in patients with metastatic nonsquamous NSCLC.

Araújo et al. [17] analyzed erlotinib in subsequent treatment lines versus docetaxel, pemetrexed, and BSC for the Portuguese health care system. They conclude that erlotinib is dominant with lower cost and higher efficacy than docetaxel as well as pemetrexed. However, the ICER per QALY compared to BSC was $ 236,924. Carlson et al. [12] drew a similar conclusion concerning erlotinib vs. docetaxel or pemetrexed for a US setting. Erlotinib seems to be dominant, since more QALYs were gained at lower costs. The cost-effectiveness of erlotinib vs. docetaxel was also analyzed for the setting of the National Health Service (NHS). Lewis et al. [18] concluded that erlotinib is dominant and cost-effective. In contrast, Thongprasert et al. [19] and Cromwell et al. [20] did not reveal cost-effectiveness for erlotinib vs. docetaxel for the Thai and Canadian setting, respectively. Two years costs were higher with only slightly better efficacy for erlotinib in the second-line treatment of advanced NSCLC patients from a Thai payer perspective [19]. However, Cromwell et al. [20] did not discover statistical difference in terms of costs and overall survival (OS) for the second-line treatment with erlotinib. The cost-effectiveness of erlotinib vs. BSC was assessed somewhat opposing in the studies by Cromwell et al. [21] and Bradbury et al. [22] for the Canadian setting. Cromwell et al. [21] revealed a potential cost-effectiveness of erlotinib for the third-line treatment compared to BSC. By contrast, Bradbury et al. [22] reported no cost-effectiveness of erlotinib for second and third-line treatment due to a substantial lower benefit in terms of OS. However, they calculated an increasing effectiveness of erlotinib when the analysis was restricted to a second-line treatment setting.

Only two studies were identified that evaluated the cost-effectiveness of gefitinib [19, 23]. Zhu et al. [23] considered EGFR mutation status in their cost-effectiveness model for first-line gefitinib maintenance therapy. Only patients with EGFR mutation-positive NSCLC received gefitinib. They concluded that gene-guided gefitinib maintenance treatment is indicated as a cost-effective option compared to routine follow-up in China. From a Thai payer perspective, Thongprasert et al. [19] revealed that gefitinib is dominant compared to docetaxel in the second-line treatment without taking pharmacogenomic profiling into consideration.

Three other studies addressed the cost-effectiveness of treatments including erlotinib [24‐26]. The objective of a study by Chouaid et al. [24] was to assess the cost-effectiveness of erlotinib followed by chemotherapy after disease progression, compared to the reverse strategy from the perspective of the French health care system. Chemotherapy included docetaxel and gemcitabine. They focused on a highly specific population of fit elderly; no significant differences in patient outcomes were identified. However, first-line treatment with chemotherapy was slightly more expensive. Chouaid et al. [25] replicated this study design for frail elderly patients and, once again, could not detect meaningful differences in terms of cost-effectiveness. Carlson et al. [26] evaluated the cost-utility of implementing epidermal growth factor receptor (EGFR) testing before initiating second-line therapy with erlotinib. Two testing strategies were compared. The EGFR protein expression test and the EGFR gene copy number test. Within the testing strategies, erlotinib was given to patients with high expression or a high copy number and docetaxel to those with low expression or a low number, respectively. Erlotinib monotherapy, without testing, was used as the comparator. The analysis showed that EGFR testing has the potential to improve quality-adjusted life expectancy in NSCLC. However, the improvement could only be achieved at high costs and the results had a high uncertainty.

Five studies evaluated the cost-effectiveness of bevacizumab in the first-line treatment of advanced non-squamous NSCLC [15, 27‐30]. Four of these studies compared bevacizumab plus chemotherapies to chemotherapies alone [27‐30]. Giuliani et al. [27] analyzed bevacizumab in combination with cisplatin and gemcitabine vs. pemetrexed in combination with cisplatin. They conclude that the bevacizumab-based therapy can be considered as cost-effective in Italy. Comparing the same treatment regimens for the Korean and Taiwanese setting, Ahn et al. [28] revealed similar results. Both, Ahn et al. [28] as well as Giuliani et al. [27], used an indirect comparison to obtain efficacy data since no head-to-head trials existed. However, the results of both studies were completely different in terms of LYG. Goulart et al. [29] assessed the cost-effectiveness of bevacizumab plus chemotherapy compared to chemotherapy alone for the US setting. Chemotherapeutic agents were carboplatin and paclitaxel. They concluded that bevacizumab is not cost-effective when added to chemotherapy; neither when considering LYG nor when taking QALY’s into account. The cost-effectiveness of bevacizumab added to carboplatin and paclitaxel in comparison to pemetrexed plus cisplatin was further analyzed by Klein et al. [30]. Marginal benefits in terms of LYG and QALY’s can only be achieved at very high costs. Hence, bevacizumab based treatment is supposed to be not cost-effective. Beyond, Klein et al. [15] concluded that bevacizumab monotherapy is dominated by pemetrexed in the first-line maintenance treatment due to lower efficacy and higher costs.

The results of the quality assessment using the QHES instrument are presented in Table 5. The table shows how often each criterion was met by the 19 studies. The quality of the included studies is at a fair level (Mean QHES Score: 66.5 SD: 17.2). More than 40% of the studies are classified as high quality and almost half of the studies are of fair quality. Only two studies are classified as poor quality [24, 25].

The study objectives as well as the perspective of the analysis are clearly presented by all studies (Question 1 & 2). Only three studies [20‐22] justify the data selection e.g. through a systematic literature review. The remaining studies may not have used data from the best available source (Question 3). As mentioned before, using data from RCT without any justification was not considered appropriate. Klein et al. [15] used data from a subgroup analysis, which, however, the authors do not describe properly (Question 4). Uncertainty is not handled sufficiently by more than half of the studies (Question 5) [13, 15, 18‐21, 24, 25, 27, 30]. Conducting probabilistic sensitivity analysis usually fulfilled this criterion. All studies perform an incremental analysis, since this was an inclusion criterion for this review. However, Klein et al. [15, 30] do not undertake a full incremental analysis between all available comparators (Question 6). Detailed information on the methods used to derive data or parts of the data is not reported in eight studies (Questions 7) [13, 15, 19, 23‐25, 27, 30]. All studies handle time horizons and discounting correctly, except for Klein et al. [30] who do not discount in the base case as well as Chouaid et al. [24, 25] who do not discount nor clarify the time horizon of their model (Question 8). The costs were not measured appropriately in two thirds of the publications (Question 9) concerning the statement of the price year [13, 14, 28, 30], inclusion of all relevant costs [13, 24, 27], description of the estimation of quantities [15, 26, 30], and details on expert panels [17‐19]. In seven studies it remains unclear if all long term costs were included (Question 10), due to no consideration of subsequent treatment costs [13, 16, 22, 29], overall insufficient reporting [24, 25], and incomplete cost consideration of the societal perspective [26]. The results by Walleser et al. [14] might be sensitive to the usage of a QoL measurement, whereas Wang et al. [16] have given no justification for using an unusual health outcome measure (Question 11). While three studies [16, 20‐22] use no economic model, the model is displayed in an intransparent manner by three studies [18, 24, 25] and the analysis reported selectively by one study [17] (Question 12). No justification for the choice of the model was given by four studies [13, 14, 23, 25] (Question 13). The direction and magnitude of potential bias is not discussed by eleven studies [13‐19, 24, 25, 27, 28] (Question 14). With one exception, the conclusions stated by the authors are based on the respective study results and appear to be reasonable. Only Araújo et al. [17] and Wang et al. [16] seem to have selectively reported the results of their models (Question 15). Finally, Walleser et al. [14] as well as Thongprasert et al. [19] do not disclosure the source of funding for the study (Question 16).