A Phamacoeconomic Analysis of Personalized Dosing vs Fixed Dosing of Pembrolizumab in Firstline PD-L1-Positive Non–Small Cell Lung Cancer

Abstract

Background

In October 2016, pembrolizumab became the new standard of care for firstline treatment of patients with metastatic non–small cell lung cancer (mNSCLC) whose tumors express programmed death ligand 1 in at least 50% of cells. The US Food and Drug Administration–recommended dose is 200 mg every three weeks. Multiple studies have demonstrated equivalent efficacy with weight-based doses between 2 mg/kg and 10 mg/kg. The objective of this study was to compare the economic impact of using personalized dosing (2 mg/kg) vs fixed dosing (200 mg) in the firstline setting of mNSCLC.

Methods

We performed a budget impact analysis from the US societal perspective to compare fixed dosing with personalized dosing. We calculated the target population and weight of patients who would be treated with pembrolizumab annually in the firstline setting. Using survival curves from the KEYNOTE 024 trial with Weibull extrapolation, we estimated the mean number of cycles that patients would receive. Using the Medicare average sales price, we calculated the difference in cost between personalized and fixed dosing.

Results

Our base case model demonstrates that the total annual cost of pembrolizumab with fixed dosing is US $3 440 127 429, and with personalized dosing it is US $2 614 496 846. The use of personalized dosing would lead to a 24.0% annual savings of US $825 630 583 in the United States.

Conclusions

Personalized dosing of pembrolizumab may have the potential to save approximately $0.825 billion annually in the United States, likely without impacting outcomes. This option should be considered for the firstline management of PD-L1-positive advanced lung cancer.

Despite recent advances with the use of targeted therapy, 158 080 people die from lung cancer annually in the United States (1). In October 2016, the results of the pivotal KEYNOTE 024 trial were published comparing the safety and efficacy of pembrolizumab with conventional chemotherapy in the firstline treatment of patients with metastatic non–small cell lung cancer (mNSCLC) whose tumors express programmed death ligand 1 in at least 50% of cells (PD-L1 ≥ 50%) (2). In these patients, pembrolizumab conferred a statistically significant benefit in terms of progression-free survival (PFS), with a hazard ratio of 0.5 (95% confidence interval [CI] = 0.37 to 0.68, P < .001). The hazard ratio for death (overall survival) was 0.60 (95% CI = 0.41 to 0.89, P = .005). As a result, pembrolizumab has become the new standard of care for the firstline management of patients with PD-L1-positive mNSCLC.

The dose for pembrolizumab used in KEYNOTE 024 was 200 mg for all patients (fixed dosing) every three weeks, and as a result this is the dose approved by the Food and Drug Administration (FDA) for this indication. However, in prior clinical trials of pembrolizumab, alternative personalized dosing strategies have been used with equivalent safety and efficacy. The initial phase I trial aimed to identify the maximal tolerated dose and to assess pharmacodynamics with a wide range of doses (3). This study demonstrated complete target engagement with full saturation at a dose of 1 mg/kg, which was durable for at least 21 days, and there was no difference in pharmacodynamics with alternative doses of 1, 3, or 10 mg/kg. Translational models of intratumor exposure predicted robust responses at doses of 2 mg/kg every three weeks, leading to a recommended phase II dose of 2 mg/kg every three weeks. The initial FDA approval of pembrolizumab was for the treatment of metastatic melanoma with a dose of 2 mg/kg every three weeks (4). Another trial in melanoma demonstrated equivalent levels of efficacy with pembrolizumab irrespective of whether the dose was 2 mg/kg or 10 mg/kg every three weeks (5). In the expansion cohorts of the phase I trial in mNSCLC (KEYNOTE 001), patients were treated with either 2 mg/kg or 10 mg/kg every three weeks or 10 mg/kg every two weeks (6). The response rate and toxicity profile was similar regardless of dose and schedule. Furthermore, an exposure-response model based on KEYNOTE 001 demonstrated no differences in efficacy irrespective of whether the dos was 2 mg/kg or 10 mg/kg every three weeks (7). KEYNOTE 010 compared pembrolizumab at 2 mg/kg and 10 mg/kg every three weeks with docetaxel in previously treated patients with mNSCLC whose disease expressed PD-L1 in more than 1% of cells. This study again demonstrated equivalent efficacy with pembrolizumab irrespective of the dose (8).

We suspected that using the FDA-approved dose of 200 mg for all patients may be an unnecessarily high dose, given that the average weight in the United States is 82 kg (9). In an era of value-based cancer care, avoidance of drug wastage is of paramount importance. If dosed at 2 mg/kg, an appropriate dose for the average American adult would be 164 mg. The objective of this study was to estimate the difference in financial impact using personalized dosing (2 mg/kg) vs fixed dosing (200 mg) for patients with PD-L1-positive mNSCLC.

Methods

Methodological Overview

We performed a budget impact analysis according to the guidelines set forth by the International Society of Pharmacoeconomics and Outcomes Research (10). We performed this analysis from the societal perspective of payers in the United States. We first estimated the population of patients expected to receive pembrolizumab annually. We then estimated the range of weights for these patients. We then calculated the total annual volume of drug required using personalized or fixed dosing. Finally, we calculated the cost using the two different methods of dosing. The description and justification for each variable is provided below. All variables and their ranges are provided in Table 1.

Table 1.

Variables used in the model*

Description	Value	Minimum	Maximum	Distribution	References
No. of lung cancer deaths nationally per year	158 080	142 272	173 888	Gamma	Siegel et al., 2016 (1)
% of lung cancer patients with SCLC	13.0	11.7	14.3	Beta	Howlader et al., 2013 (11) Govindan et al., 2006 (12)
% of NSCLC patients with EGFR mutation	23.0	20.7	25.3	Beta	Midha et al., 2015 (13)
% of NSCLC patients with ALK translocation	4.4	4.0	4.8	Beta	Gainor et al., 2013 (14)
% of NSCLC patients ineligible for systemic therapy	35.0	31.5	38.5	Beta	Goulart et al., 2013 (15) Warren JL et al., 2015 (32)
% of non-EGFR/ALK mutated NSCLC patients with PDL1 > 50%	30.2	27.2	33.2	Beta	Reck et al., 2016 (2)
No. of cycles of pembrolizumab	17.9	13.4	35	Weibull	Reck et al., 2016 (2)
Patients’ weight, kg	76	68.4	83.6	Normal	Emory University
Drug cost (ASP + 4.2%), $ per mg	48.45	43.605	53.295	Gamma	Medicare

*

NSCLC = non–small cell lung cancer; SCLC = small cell lung cancer.

Target Population

Figure 1 demonstrates how we estimated the target population of patients. We started with lung cancer deaths and then removed subpopulations of patients who would not receive pembrolizumab in the firstline setting of mNSCLC. The Surveillance Epidemiology and End Results Program (SEER) estimates 158 080 deaths due to lung cancer in the United States in 2016 (1). Thirteen percent of patients with lung cancer have small cell disease, and this percentage has been constant since 2009 (11,12). A systematic review of 16 US-based studies comprising 6663 patients with NSCLC demonstrated a rate of 23.0% (13). The frequency of ALK rearrangement has been previously estimated to be 4.4% (14). The number of patients with mNSCLC who do not receive systemic therapy is difficult to estimate. An analysis using data from SEER and Medicare including 157 638 patients with stages III and IV NSCLC (2000–2005) reported that 45.0% of patients with mNSCLC did not receive systemic chemotherapy (15). There are many reasons that patients do not receive treatment; these may include lack of insurance, poor performance status, or concerns related to adverse events. We propose that 45.0% is an overestimation for our model considering that systemic chemotherapy has changed since 2005 with the approval of pemetrexed (16) and the adherence to guidelines recommending platinum-based chemotherapy for Eastern Cooperative Oncology Group (ECOG) performance status 2 patients (17). Additionally, we speculate that in the future patients with poor performance status may be more likely to receive systemic therapy because the option of pembrolizumab has a favorable toxicity profile (2). We therefore used 35.0% as our estimate in the analysis. In the KEYNOTE 024 study, 30.2% of patients screened had PD-L1 tumor proportion scores of 50% or higher; thus they were eligible for enrollment (2).

Figure 1.

Target patient population in base case analysis.

Patients’ Weight

The average weight of an American adult is 82 kg (9); however, because of the combination of tobacco use and cachexia, we suspected that the weight of patients with mNSCLC may be lower. We therefore analyzed the weights of all patients with mNSCLC in the course of one year (2015) at Emory University Hospital in Atlanta, Georgia, with prior institutional review board approval. Written informed consent was not required because of the retrospective nature of the study. We collected the weights of 77 deidentified patients who were eligible to receive firstline systemic therapy. There was a normal distribution, with a median of 76.5 kg and mean of 76.0 kg. In the base case model, we used a value of 76 kg.

Cost of Pembrolizumab

We used the average sales price (ASP) for pembrolizumab from the fourth quarter of 2016 ($46.495 per mg) and added 4.2% to represent the current pattern of Medicare reimbursement.

Number of Cycles of Pembrolizumab

In the KEYNOTE 024 study, patients were assigned to receive pembrolizumab for a maximum of two years (35 cycles) or until disease progression or unacceptable toxicity. At the time of trial data analysis (May 9, 2016), the median duration of follow-up was 11.2 months and 48% of patients in the pembrolizumab group were still receiving therapy. The median number of cycles received at that point was 10.5. In order to capture the subsequent cycles received by durable responders following publication, we used the Nelder-Mead Simplex Method (18) to fit a Weibull distribution to the Kaplan-Meier curve. This Weibull distribution was truncated at two years to represent the maximum number of cycles (Figure 2). The Weibull distribution was then used to estimate the mean number of cycles received.

Figure 2.

Truncated Weibull distribution of treatment duration up to two years. Progression-free survival curve from the pembrolizumab arm of the KEYNOTE 024 trial and the fitted Weibull distribution of treatment duration truncated to 24 months. Weibull distribution survival function: ${S\left(t\right)=\hspace{0.17em}\text{exp}\hspace{0.17em}(-\lambda t}^{\gamma })$⁠.

Sensitivity Analyses

We created a range for all parameters in order to perform sensitivity analyses. For all parameters except treatment duration, we applied +/− 10% to all base case parameters. For treatment duration, the range was based on the Weibull distribution. Details of distributions are available in Table 1. We performed a univariate sensitivity analysis to assess the variables for which the range had the greatest impact on cost savings.

Model Structure

We used Matlab (The MathWorks) to run a Monte Carlo Simulation of the budget impact analysis. The model was run 100 000 times, using the variables included in Table 1. All parameters were inserted into the model to capture target population, weights, drug cost, and treatment duration. The distributions used as inputs for the model are demonstrated in the Supplementary Materials (available online).

Results

In the base case analysis, the total annual cost of pembrolizumab with fixed dosing is US $3 440 127 429, and with personalized dosing it is US $2 614 496 846. The use of personalized dosing would lead to an annual savings of US $825 630 583. With 100 000 simulations of our model, the mean potential annual savings with personalized dosing from the US societal perspective is US $869 310 000, with a standard deviation of US $601 680 000. The results of the Monte Carlo simulation are available in Figure 3.

Figure 3.

Probabilistic sensitivity analysis of potential cost savings as a result of 100 000 simulations of the model using Monte Carlo methods. This graph demonstrates the results of 100 000 simulations of the model. In each individual simulation, different values for each variable are randomly selected from the related distribution of their respective ranges. This leads to a different potential cost savings for each simulation. As a result, there is a different probability of different cost savings. For example, there is a 50% probability of saving a minimum of $800 000 000.

The univariate sensitivity analysis (Figure 4) demonstrates that the model variables with the greatest potential impact on the differences in annual savings are patients’ weight and number of cycles of pembrolizumab. With this analysis, the most conservative estimate of savings is $0.564 billion annually when the average weight is considered to be at the upper end of the range (83.6 kg).

Figure 4.

Univariate sensitivity analysis demonstrating the variables with the greatest impact on potential cost savings. NSCLC = non–small cell lung cancer.

Discussion

The results of KEYNOTE 024 were paradigm shifting in the management of mNSCLC, and a new era has arrived in the management of patients with lung cancer. Coupled with advances in cancer care are economic challenges related to the ability to pay for such advances. An integral component of value-based cancer care is the avoidance of waste. In this study, we have demonstrated that by using personalized dosing it may be possible for US society to save $0.825 billion annually and that this shift is unlikely to impact clinical outcomes.

The relevance of our analysis is dependent on the clinical equivalency between fixed dosing and personalized dosing. Several prior studies have demonstrated equivalent efficacy irrespective of whether a dose of 10 mg/kg or 2 mg/kg was used (4–8). The question of 200 mg fixed dosing vs personalized dosing is ultimately a question of 3 mg/kg vs 2m g/kg. As a result, it appears that the efficacy is equivalent irrespective of dose. Early clinical trials and models suggest that the exposure-response relationship reaches a plateau beyond 2 mg/kg, with no statistically significant improvement in efficacy across doses of 2 to 10 mg/kg (3,7,19,20). Furthermore, an analysis comprising samples from 2993 patients on KEYNOTE studies 001, 002, 006, 010, and 024 found that the pharmacokinetics of pembrolizumab is consistent across weight-based and fixed-dose regimens and demonstrated a flat exposure-response relationship (21). These findings were confirmed in randomized clinical trials demonstrating no statistically significant differences in outcomes (objective response rate (ORR), PFS, and overall survival) between the two dose options (2 vs 10 mg/kg) in melanoma and mNSCLC (5,8,22).

In practical terms, which stakeholders should be responsible for ensuring personalized dosing and minimizing waste? Perhaps physicians should be responsible for ensuring that they use only the minimum necessary amount of drug required. Conversely, should public and private payers reimburse only personalized doses of pembrolizumab? However, physicians and payers may be reluctant to prescribe and reimburse a dose that is not the FDA indicated dose. We believe that the most appropriate first step to avoid waste is to change or add personalized dosing to the FDA label. If the FDA changes the label, clinicians and payers will be more comfortable prescribing and reimbursing personalized dosing. The FDA has previously demonstrated their willingness to change a label based on such an issue. The label for nivolumab, a similar checkpoint inhibitor, was recently changed by the FDA from personalized dosing to fixed dosing based on pharmacokinetic studies and dose/exposure-response analyses (23). These demonstrated the comparability of the pharmacokinetics exposure, safety, and efficacy of the two dosing strategies. The FDA determined that the overall exposure at a 240 mg fixed dose of nivolumab was similar (less than 6% difference) to 3 mg/kg and any differences in exposure were unlikely to have a clinically meaningful effect on safety and efficacy. Although it would be unusual for the FDA to make dose changes to a label independent of any request by the company, we believe that it may be possible. Of note, the FDA has no mandate to consider cost, so any label change would be solely on the grounds of equivalent clinical efficacy, not cost.

There are substantial practical challenges in realizing an annual saving of $0.825 billion. An important component for potential cost savings with personalized dosing will be the potential for vial sharing. Bach et al. recently demonstrated the problem of oversized vials leading to large volumes of wastage of cancer drugs in the United States (24). In the case of pembrolizumab, there are two important issues, namely vial size and the ability of dispensing pharmacies to share vials between patients. Pembrolizumab was initially approved in the United States in 2014, and it was sold in 50 mg vials. However, in 2015, 100 mg vials were introduced and distribution of 50 mg vials ceased in the US market. Meanwhile, pembrolizumab was approved in Europe in 2015 and was sold in 50 mg vials (24). The addition of a 10 mg vial has previously been proposed (24). We believe that it is incumbent on the manufacturer to provide multiple vial sizes in order to minimize the opportunity for wastage. The Inspector General of the US Department of Health Human Services has also recently stated that vial sizes will be a focus of future investigation (25). However, if there are insufficient vial sizes available, the vials should be shared between patients. Once a vial of pembrolizumab is opened and diluted, it may be stored at room temperature for up to six hours or refrigerated for up to 24 hours (26). We propose that a possible option to maximize vial sharing is for infusion centers to schedule pembrolizumab treatments only on certain days of the week. The current reimbursement structure also provides a financial incentive for doctors to prescribe a higher volume of expensive drugs, and this may be a further obstacle to implementation of personalized dosing. In addition, the current policy of payment for wastage is a disincentive for reducing wastage.

The pricing of drugs has come under great scrutiny in recent years. To truly pay for value, ideally there should be a system of indication-specific value-based pricing (27). In this way, the price of a drug would reflect the clinical benefit that it provides (28). However, currently in the United States, prices are neither indication specific nor value based, and they reflect more strongly what the market will bear. While policy discussions continue in the United States related to value, the current pricing and reimbursement system leaves little opportunity for payers to control expenditure on pharmaceuticals. Using dosing manipulations such as that described for pembrolizumab is one way that payers can seek to improve value within the current system. Although some cancer drug prices have increased in recent years (29), the price of pembrolizumab has remained fairly constant every quarter for the past six quarters at around $46 per mg (30). Close attention will be paid to any changes in the ASP of pembrolizumab in the future.

Our study is not without limitations. All economic models are based on estimates, which limits the accuracy of the results. Our analysis was based on data from the KEYNOTE 024 study, which included only patients with ECOG performance status of 0 and 1. In the real-world setting, it is likely that patients with poorer performance status will also be treated with pembrolizumab, which may negatively impact the progression-free survival, and thus also the duration of treatment. Also, we estimated the number of patients with mNSCLC using the incidence of lung cancer deaths. This number may potentially include a small number of patients who die of treatment without metastatic disease. We excluded patients with EGFR-mutated or ALK-rearranged mNSCLC as they are usually treated with targeted therapy. Although these patients were excluded from the KEYNOTE 024 study, they may be offered PD-L1 testing and pembrolizumab after progression on targeted therapy. We used the average sales price plus 4.2% to simulate Medicare reimbursement; however, it is possible that alternative public and private payers may pay less or more, respectively. We used the weights of patients in Georgia, which may not be equivalent to the whole nation. However, the prevalence of overweight adults in Georgia is 35%, which is the same as for the whole nation (31). The study did not account for waste when using personalized dosing because we propose that with appropriate vial sizes and vial sharing, waste should be minimal. We used PFS to estimate treatment duration, which may be imprecise; however, the practice of treatment beyond progression may mean this is an underestimate, while the cessation of therapy due to toxicity may mean that this is an overestimate. The ranges used in the sensitivity analyses incorporate the uncertainty of our estimations. Bach et al. previously projected $943 million of sales of pembrolizumab in 2016, of which $197 million was a result of wastage (24). This analysis incorporated other indications such as second-line lung cancer and melanoma and was based on the dose of 2 mg/kg for all patients. The results of our analysis are substantially different because they relate only to the recent firstline approval for lung cancer and to the dosing change.

The era of personalized immunotherapy in lung cancer has arrived and is expanding in other malignancies and will undoubtedly lead to improved outcomes for some patients. Personalized dosing has the potential to decrease costs while maintaining efficacy. While personalized dosing should be encouraged, its rapid adoption and potential cost savings may be dependent on cooperation between the FDA, physicians, patients, payers, and the manufacturer.

Note

The authors have no conflicts of interest to disclose.