Cost-effectiveness of everolimus vs sunitinib in treating patients with advanced, progressive pancreatic neuroendocrine tumors in the United States

Abstract

Background:

Everolimus (Afinitor) and sunitinib (Sutent) were recently approved to treat patients with advanced, progressive pancreatic neuroendocrine tumors (pNETs). (Afinitor is a registered trademark of Novartis Pharmaceuticals Corporation, East Hanover, NJ, USA; Sutent is a registered trademark of Pfizer Inc., New York, NY, USA.) This analysis examined the projected cost-effectiveness of everolimus vs sunitinib in this setting from a US payer perspective.

Methods:

A semi-Markov model was developed to simulate a cohort of patients with advanced, progressive pNET and to estimate the cost per life-year gained (LYG) and per quality-adjusted life-year (QALY) gained when treating with everolimus vs sunitinib. Efficacy data were based on a weight-adjusted indirect comparison of the agents using phase 3 trial data. Model health states included: stable disease with no adverse events, stable disease with adverse events, disease progression, and death. Therapy costs were based on wholesale acquisition cost. Other costs such as physician visits, tests, hospitalizations, and adverse event costs were obtained from literature and/or primary research. Utility inputs were based on primary research. Sensitivity analyses were conducted to test the model’s robustness.

Results:

In the base-case analysis, everolimus was associated with an incremental 0.448 LYG (0.304 QALYs) at an incremental cost of $12,673, resulting in an incremental cost-effectiveness ratio (ICER) of $28,281/LYG ($41,702/QALY gained). The ICER fell within the cost per QALY range for many widely used oncology drugs. Sensitivity analyses demonstrated that, overall, there is a trend that everolimus is cost-effective compared to sunitinib in this setting.

Limitations:

Results of the indirect analysis were not statistically significant (p > 0.05). Assumptions that treatment patterns are the same across therapies may not represent real-world practice.

Conclusions:

While the analysis is limited by its reliance on an indirect comparison of two phase 3 studies, everolimus is expected to be cost-effective relative to sunitinib in advanced, progressive pNET.

Keywords::

• Pancreatic neuroendocrine tumors
• Everolimus
• Sunitinib

Introduction

Pancreatic neuroendocrine tumors (pNETs) are tumors that arise from well-differentiated mature endocrine cells or multipotent stem cells within the pancreas1. The overall annual incidence of pNETs in the US is estimated to be 3.2 per million, of which 64% are diagnosed in an advanced stage2. Advanced NET, defined as unresectable and/or metastatic disease, is a rare, progressive, and fatal malignancy2. The median survival for patients with well- to moderately-differentiated pNET with distant metastases is 27 months2.

Until recently, only streptozocin had been approved by the US Food and Drug Administration (FDA) for the treatment of pNET. The lack of approved therapies has contributed to inconsistent treatment approaches and the use of toxic agents such as chemotherapy. Everolimus (Afinitor; a registered trademark of Novartis Pharmaceuticals Corporation, East Hanover, NJ), recently approved by the FDA for treatment of progressive pNETs, represents a significant clinical advance in the treatment of advanced NET by controlling disease progression through inhibition of the mammalian target of rapamycin (mTOR) pathway. The pivotal RADIANT-3 trial demonstrated that everolimus 10 mg/day administered orally plus best supportive care was associated with a 65% reduction in disease progression risk (hazard ratio [HR] = 0.35; 95% confidence interval [CI]: 0.27, 0.45; p < 0.001) in patients with advanced pNET when compared to placebo. The median progression-free survival (PFS) was 11.0 months (95% CI: 8.4, 13.9) for everolimus-treated patients and 4.6 months (95% CI: 3.1, 5.4) for placebo-treated patients. Median overall survival (OS) was not reached at the time of analysis3. Sunitinib (Sutent; a registered trademark of Pfizer Inc., New York, NY), an anti-vascular endothelial growth factor (VEGF) therapy, was also recently approved by the FDA to treat progressive, well-differentiated pNET in patients with unresectable locally advanced or metastatic disease. Its pivotal trial (A6181111) demonstrated that sunitinib 37.5 mg/day administered orally was associated with a 58% reduction in the risk of disease progression or death (HR: 0.42; 95% CI: 0.26, 0.66; p < 0.001). The median PFS was 11.4 months (95% CI: 7.4, 19.8) for sunitinib-treated patients and 5.5 months (95% CI: 3.6, 7.4) for placebo-treated patients. Early OS data reported in a recent New England Journal of Medicine publication of the A6181111 trial suggested that treatment was associated with a 59% reduction in the risk of death (HR: 0.41; 95% CI: 0.19, 0.89; p = 0.02)4. However, subsequent analyses submitted to the European Medicines Agency (EMA) and FDA of more mature data were no longer statistically significant (HR: 0.74; 95% CI: 0.47, 1.17)5.

As there were no head-to-head trials comparing everolimus and sunitinib in advanced, progressive pNET, an indirect comparison was performed based upon the sunitinib A6181111 trial and a matched sub-set of the RADIANT-3 population6. A comparable study was not available for streptozocin, and thus an indirect comparison could not be performed against everolimus. The weight-adjusted indirect comparison against sunitinib was performed using a method of moments estimator, as previously used in a Signorovich et al.7 study. The analysis demonstrated a trend towards improved PFS and OS with everolimus compared to sunitinib. To date, there have been no published economic evaluations of these treatments in pNET. Consequently, based on the results of the indirect survival comparison, this analysis examines the potential cost-effectiveness of everolimus vs sunitinib from a US payer (i.e., insurer) perspective.

Methods

Model structure

A semi-Markov model was developed in Microsoft Excel® (Microsoft Corp, Redmond, WA) to simulate two hypothetical patient cohorts with advanced, progressive pancreatic NET: one treated with everolimus and the other with sunitinib. The cohorts were modeled over a 20-year time horizon in monthly (30.4-day) cycles; the 30.4-day cycle length was calculated by dividing the number of days per year (365) by the number of months per year (12). A 20-year time horizon was chosen in order to allow 99.99% of patients in both arms to reach the absorbing state. Extrapolation of the efficacy data had produced a long tail for progression-free and overall survival, thus necessitating a long time horizon. Four health states were included in the model: stable disease with no adverse events, stable disease with adverse events, disease progression, and death (Figure 1).

Figure 1.  Semi-Markov model health states.

All patients started in the health state ‘stable disease with no adverse events’ and transitioned to the remaining health states according to PFS and OS estimates obtained from the indirect analysis6. Movement among health states was unidirectional: patients who transitioned from stable disease to disease progression were unable to return to the former state. The probability of health state membership was estimated using a partitioned survival analysis, which calculates the mean time spent in a health state from the area under the appropriate survival curve (Figure 2a and b). This type of analysis has been described by Glasziou et al.8,9. In the model presented herein, Weibull cumulative distribution curves for the everolimus arm were constructed from the PFS (shape: 1.195, 95% CI: 0.990–1.429; scale: 447.188, 95% CI: 223.632–897.847) and OS (shape: 1.379, 95% CI: 1.111–1.724; scale: 1427.245, 95% CI: 765.095–2670.444) data for the matched sub-set of the RADIANT-3 population (derived in statistical software [SAS 9.2, SAS Institute, Cary, NC USA]); the area under the curve was calculated to derive the health-state membership at each cycle. As health states were mutually exclusive, the disease progression health-state membership was calculated as the complement of the summation of the stable disease and death states. The Weibull distribution was used because it is uniquely relevant in the clinical setting10.

Figure 2.  Partitioned survival analysis of (a) everolimus patients and (b) sunitinib patients.

Weibull curves for sunitinib were constructed by applying the hazard ratios for PFS and OS (0.84 [95% CI: 0.460–1.530] and 0.81 [95% CI: 0.490–1.310], respectively) reported in the indirect analysis6 to the resultant transition rates11 from the everolimus-based parametric curves. As the survival curves incorporated in the model were time-dependent, the transition probabilities (and the transition rates) varied according to the cycle. A similar methodology utilizing Weibull distributions fitted to Kaplan-Meier curves for the baseline therapy, and hazard ratios to estimate the expected disease progression in the comparator was used in the Peninsula Technology Assessment Group’s (PenTAG) assessment of bevacizumab, sorafenib tosylate, sunitinib, and temsirolimus for renal cell carcinoma12.

The proportion of everolimus-treated patients who experienced an adverse event (AE) in each cycle was calculated from the RADIANT-3 data. For the weight-adjusted matched population used in the indirect comparison with sunitinib, the proportion of everolimus patients who experienced an adverse event was derived by calculating the AE rate ratio between the pre- and post-matched everolimus population and applying the inverse to the former. The proportion of sunitinib patients who experienced an adverse event at each cycle was then derived by calculating the ratio of AE rates between the post-match everolimus population and the sunitinib population, and applying the inverse to the cycle-specific adverse event rate of the post-matched everolimus arm. It was assumed that presence in ‘stable disease with adverse events’ occurred for a single cycle as many of the AEs could be resolved within a hospitalization visit and did not lead to treatment discontinuation; presence in this state was cross-sectionally assessed for each model cycle. The adverse events included in the analysis were based on the grade 3/4 AEs that occurred in at least 2% of everolimus patients (irrespective of treatment)13 and at least 2% of sunitinib patients5, as well as any AEs that had significantly different rates (p < 0.05) in the patient populations (of which there were eight), as identified in the indirect analysis6. Adverse event status did not affect likelihood of progression or death, but was used to calculate the associated quality-of-life (QoL) reductions and increased costs.

Resource utilization

To estimate resource utilization, a survey of physicians with experience in NET treatment in the US was conducted14. Physicians reported treatment practice patterns on patients (40 across all physicians) during three disease stages: baseline, initial progression, and second progression.

The cost-effectiveness analysis incorporated the resource utilization estimates from the initial progression stage for the model’s stable disease state as the RADIANT-3 patient population consists of persons with advanced, progressive NET. The analysis incorporated the survey’s second progression results in the model’s disease progression state (i.e., when the patients represented in the model progress a second time). The model did not incorporate the resource utilization data captured for the survey baseline disease stage, as patients in this category have not yet progressed. Results were analyzed using statistical software (Cary, NC).

Cost inputs

In the RADIANT-3 trial and in the A6181111 study, patients received, respectively, everolimus 10 mg/day or sunitinib 37.5 mg/day (or placebo in both studies) until disease progression or dose reduction/discontinuation due to tolerance issues. The model used the patient-level data to estimate the treatment dose intensity, calculated to be 85.9% for everolimus13 and 91.3% for sunitinib4. The drug cost of everolimus and sunitinib was based on 2011 wholesale acquisition cost, and was estimated to be $221 per day and $230 per day, respectively (Table 1)15.

Table 1.  Drug acquisition costs.

Drug	Dose and frequency	Cost	Cost per monthly cycle
Everolimus	10 mg daily	$221 per dose^a	$5768^b
Sunitinib	37.5 mg daily	$230 per dose^a	$6374^c

^aDrug unit costs from an electronic drug reference¹⁵; mg, milligram.

^bAssumed a 85.9% dose intensity.

^cAssumed a 91.3% dose intensity.

Patients also received supportive care, which included symptomatic care (somatostatin analogs), physician visits, imaging and lab tests, hospitalizations, treatment for AEs (i.e., AE-specific hospitalizations), post-progression therapy, and end-of-life care (Table 2). Drug costs were derived from an electronic drug reference (Red Book for Windows 2011, Thomson Reuters, Greenwood Village, CO USA15); medical resource utilization costs were taken from procedure codes (Physicians' Fee and Coding Guide, MAG Mutual, Atlanta, GA USA16); and the Thomson Reuters© MS-DRG 2010 Handbook17. Costs of AEs were based on the Healthcare Cost & Utilization Project18 and secondary literature19–21. End-of-life care costs were calculated from literature22. As this analysis examined the cost-effectiveness of the two oncologics from a third-party payer’s perspective, other societal costs, such as patient time or productivity losses, were not considered.

Table 2.  Health state medical costs.

Health state	Cost per cycle ($USD)^a
Stable disease with no adverse events	$3744
Symptomatic care	$1546
Procedures/tests	$471
Physician visits	$318
Hospitalizations	$1410
Stable disease with adverse events^b	$9491 (everolimus)
	$10,427 (sunitinib)
Symptomatic care	$1546
Procedures/tests	$471
Physician visits	$318
Hospitalizations	$1410
Adverse events	$5747 (everolimus), $6683 (sunitinib)
Disease progression	$3448 (entry tunnel state)
	$1980 (follow-up state)
Symptomatic care	$50
Procedures/tests	$222
Physician visits	$935 (entry tunnel state), $149 (follow-up state)
Hospitalizations	$1,250 (entry tunnel state), $568 (follow-up state)
Post-progression treatments	$991
Death	$9956^c
End-of-life care	$9956

^aDrug costs were derived from an electronic drug reference15; medical resource utilization costs were taken from procedure codes16 and MS-DRG codes17. Costs of adverse events were based on the Healthcare Cost & Utilization Project18 and secondary literature19–21. End-of-life care costs were calculated from literature22.

^bAdverse events included: abdominal pain, anemia, anorexia, asthenia, diabetes mellitus, diarrhea, fatigue, hand-foot syndrome, hyperglycemia, hypoglycemia, hypertension, leukopenia, neutropenia, peripheral edema, pyrexia, rash, stomatitis, and thrombocytopenia.

^cThis cost is applied upon entry into the death state.

USD, United States dollars.

Health state utility values

Health state utility scores (between 0 = death, and 1 = perfect health) were used to calculate quality-adjusted life-years (QALYs). A time trade-off (TTO) study was conducted in the UK to estimate the utility value of the stable disease and disease progression health states23. Disutility values associated with specific adverse events were obtained from the study, and were multiplied by the relative frequency of the corresponding event to yield a weighted average disutility value for each comparator’s adverse event profile (Table 3). It should be noted that the time trade-off study did not provide disutility values for all adverse events present in the model.

Table 3.  Health state utility values.

Health state	Mean utility value (SD)
Stable disease with no adverse events	0.768 (0.230)
Stable disease with adverse events:
Everolimus	0.656 (0.239)
Sunitinib	0.629 (0.240)
Disease progression	0.612 (0.240)
Death	0.00 (0.00)

*average of utility score for stomatitis, rash, diarrhea, nausea, vomiting, hyperglycemia, thrombocytopenia, hand & foot syndrome, and pneumonitis).

SD, standard deviation.

Base-case analysis

The effectiveness and total costs for each cohort were assessed and compared based upon the point estimates of the hazard ratios for PFS and OS from the indirect comparison. The analyses estimated the total and incremental life-years (LYs) and QALYs. Both costs and survival estimates were discounted annually. The model incorporates a discount rate of 3%, as supported by pharmacoeconomic guidelines published by the Academy of Managed Care Pharmacy24. Discounting began during the first year of the model. The ICERs were calculated as the cost per life-year gained and cost per QALY gained using the discounted total costs and survival time in the final model cycle (cycle 240).

Sensitivity analyses

One-way sensitivity analyses were performed to identify parameters driving the ICERs. Parameters tested included: discount rate of the cost and efficacy values, model time horizon, adverse event care costs, post-progression treatment costs, end-of-life care costs associated with death, health state utility values, and PFS and OS hazard ratios. Additionally, the consistency of the results across scenarios was evaluated by a probabilistic sensitivity analysis (PSA). The cost parameters were assumed to follow a gamma distribution, utility parameters to follow a beta distribution, the Weibull survival curve parameters to follow a normal distribution, and the hazard ratio parameters to follow a log-normal distribution (Table 4). The PSA was run for 1000 iterations. Results of the PSA were used to generate a cost-effectiveness scatterplot.

Table 4.  Model parameters varied in the PSA and their distributions.

Parameter	Distribution (range)	Rationale
Health state utilities	Beta	Standard convention33.
Costs	Gamma	Standard convention33.
Dose intensity	Normal Note: This range was truncated so that values < 0 could not be generated	The normal distribution was truncated in order to prevent generation of negative numbers and to avoid a skew towards higher values; the range of possible values is symmetrical across the mean. A beta distribution cannot be used as mean daily dose may be greater than 100%. A gamma distribution is not recommended as it has a long tail to the right, which is unlikely to reflect variance in mean daily dose.
Weibull parameters: shape, scale	Normal	Standard convention33.
Hazard ratios	Log-Normal	Standard convention33.

Results

The base-case analysis projected everolimus and sunitinib to yield mean survival times of 3.298 life-years and 2.850 life-years, respectively. Everolimus treatment would incur an incremental cost of $12,673, 23% of which comprised active treatment costs associated with the predicted increase in PFS implied by the indirect comparison (Table 5). Estimated post-progression treatment costs were higher for the everolimus cohort than for the sunitinib cohort, which is also a result of the implied additional survival of the everolimus patients per the indirect analysis. Incremental cost-effectiveness ratios for the base-case analysis were $28,281/LYG and $41,702/QALY.

Table 5.  Base case results.

	Everolimus	Sunitinib	Incremental
Life-years	3.298	2.850	0.448
Quality-adjusted life-years	2.186	1.822	0.304
Mean duration of treatment (months)	11.896	10.967	0.928
Mean PFS duration (months)	14.348	12.512	1.836
Mean post-progression duration (months)	29.024	24.623	4.402
Active treatment costs	$77,516	$74,580	$2935
Symptomatic care costs	$22,077	$19,210	$2867
Physician visits costs	$8829	$7757	$1072
Procedures and tests	$12,137	$10,510	$1627
Hospitalization costs	$34,363	$29,866	$4497
Adverse events treatment costs	$4271	$8070	($3799)
Post-progression treatment costs	$25,889	$22,283	$3606
End-of-life care costs	$8973	$9105	($132)
Total	$194,053	$181,381	$12,673
Cost per LYG			$28,281
Cost per QALY			$41,702

PFS, Progression-free survival; LYG, life-year gained; QALY, quality-adjusted life-year. Parentheses indicate a negative value.

The one-way sensitivity analyses are summarized in Table 6. The PFS hazard ratio point estimate greatly influenced model results. Active treatment dose intensity, post-progression treatments costs, and adverse event costs were also influential on the results. The PSA yielded results that were consistent with the deterministic analysis. The probabilities that the ICER fell below $50,000/QALY and $100,000/QALY were 53% and 69%, respectively. The PSA yielded some instances when everolimus was dominated by sunitinib, as well as instances when everolimus was dominant (Figure 3); however, the probability that everolimus had a gain in QALYs over sunitinib was 81%.

Table 6.  One-way sensitivity analysis results.

	Base case	Sensitivity analysis	ICER (cost/QALY)
Base case	n/a	n/a	$41,702
General
Discounting (costs and benefits)	3.0%	0%	$42,510
		5%	$41,122
Time horizon	20 years	10 years	$41,474
Costs
End-of-life care cost	$9956	$0	$42,137
AE cost	$5747 (everolimus) $6683 (sunitinib)	$0 (both treatment arms)	$54,202
Post-discontinuation treatment	$991 per patient per cycle	$0	$29,837
Active treatment dose intensity	86% (everolimus) 91% (sunitinib)	100%	$60,185
Health state utilities
Stable disease without adverse events	0.768	0.726 (lower 95% CI)	$42,787
		0.809 (upper 95% CI)	$40,695
Stable disease with adverse events	0.656 (everolimus) 0.629 (sunitinib)	0.608 (everolimus, lower 95% CI) 0.564 (sunitinib, lower 95% CI)	$41,228
		0.701 (everolimus, upper 95% CI) 0.690 (sunitinib, upper 95% CI)	$42,174
Disease progression	0.612	0.564 (lower 95% CI)	$43,800
		0.659 (upper 95% CI)	$39,833
Hazard ratio estimates
PFS	0.844	0.460 (lower 95% CI)	$144,562
		1.530 (upper 95% CI)	($226,908)
OS	0.805	0.490 (lower 95% CI)	$40,406
		1.310 (upper 95% CI)	$37,687

N/A, not applicable; ICER, incremental cost-effectiveness ratio; QALY, quality-adjusted life-year; CI, confidence interval; PFS, progression-free survival; OS, overall survival. Parentheses indicate a negative value.

Figure 3.  Probabilistic sensitivity scatter plot. Note: The ellipse demonstrates the 95% confidence interval for the PSA results and the center of the triangle represents the base-case ICER.

Discussion

While everolimus treatment of patients with advanced, progressive pNET has proven to be clinically efficacious, model results indicate that it is also likely to be cost-effective compared to sunitinib: specifically, in 81% of simulated projections, everolimus had an added benefit in terms of QALYs gained. Overall, the results of this analysis remained consistent across one-way sensitivity analyses (Table 6). The PFS hazard ratio has the greatest impact on model results. While it may seem counterintuitive that the lower bound of this parameter (the maximum expected efficacy for PFS) produced an ICER of $144,652, while the upper bound (the lowest expected efficacy for PFS) produced a dominant ICER, these results can be explained by the fact that the PFS and OS are not assumed to be correlated and are modeled independently in the analysis. Therefore, as the HR for PFS decreases, the PFS duration (and corresponding active treatment cost) for everolimus increases relative to sunitinib, while OS remains unaffected. This results in a higher ICER despite the improved efficacy. This result also reflects the fact that the change in QALYs in the denominator of the ICER is only slightly affected by improved efficacy in terms of PFS and is largely driven by OS. Overall, based on the available data in comparison to sunitinib (along with the associated variability of these data), everolimus has a high probability (69%) of being considered cost-effective at a willingness-to-pay threshold of $100,000/QALY.

The conclusions from this analysis should be viewed in light of several limitations. Although the indirect analysis demonstrated a trend towards improved PFS and OS for everolimus vs sunitinib, the calculated HRs were not statistically significant (p > 0.05). Despite this uncertainty, the estimates indicate that it is more likely that everolimus will provide added benefit and be cost-effective compared to sunitinib, than the converse. The fact that the HRs were not statistically significant may explain why everolimus was dominated by sunitinib in rare instances of the PSA and was dominant in other instances.

Furthermore, due to limitations in the availability of adverse event reporting between the trials, the model accounts for AE costs in a manner that is likely biased against everolimus. The RADIANT-3 trial captured AEs irrespective of the relationship to the study treatment while the sunitinib trial captured treatment-emergent AEs: therefore, the model may be over-estimating the AE costs for the everolimus arm. As expected, sensitivity analysis results demonstrate that everolimus remains cost-effective compared to sunitinib when AE costs for both arms are assumed to be $0.

Regarding resource utilization, the analysis assumes that all patients in stable disease (both those with and without adverse events) have the same resource utilization rates, aside from the inclusion of adverse event hospitalizations for the former. It is also assumed that treatment patterns are the same across therapies, which may not represent real-world practice. Furthermore, as many of the physicians were unable to report data for the model’s ‘disease progression’ state, the majority of responses for this group were based on hypothetical treatment scenarios. Although the survey results may differ from real-world treatment patterns of patients with pNET, they offer informed assumptions. The limitations of this study are expected to affect both model treatment arms similarly.

Study results also depend on the selected utility values. This analysis used values from a time trade-off study executed through fielding a survey of healthy individuals’ responses to NET-related vignettes. This method may be limited as the instrument may lack sensitivity between disease state utilities because healthy individuals may not fully comprehend the changes in QoL. Furthermore, the utility values for ‘stable disease with adverse events’ were based on a weighted average accounting for nine AEs. Therefore, the entire adverse event profiles were not represented, which may lead to over- or under-estimation of the state utility value. The utility value-related limitations, however, likely apply equally to both arms. As noted above, results from the sensitivity analyses indicate that this variability did not greatly affect the overall results.

In light of the uncertainty, there exists the question of whether it is worth conducting more research to narrow the parameters’ confidence intervals, thus providing a higher level of confidence in results. Value-of-information analyses can address the benefits and costs of gathering more data: however, as argued by Claxton, if a decision must be made immediately, a decision maker’s best choice is generally based on the estimated direction and magnitude of the current differential, even when the difference is not statistically significant based on conventional practice25. In principle, for physicians and payers to make more informed decisions regarding the use of everolimus and sunitinib, they would need to have improved efficacy data. As previously mentioned, it would be ideal to have a head-to-head trial comparing the therapies. Both parties would also benefit from access to more comparable safety data between the therapies; physicians would benefit from a more accurate view of patient safety, and payers would benefit from the more precise AE cost estimates. Similarly, both groups would require more precise utility scores in order to make better informed decisions. This information would help physicians better assess patients’ QoL while on treatment, and it would help payers determine more accurately the incremental cost per QALY. A value-of-information analysis would be a useful next step to compare the costs of conducting this trial, with the potential benefits.

The use of benchmarks or willingness-to-pay thresholds in healthcare economic analyses remains under debate26–31. Some economists have proposed a threshold of $50,000/QALY28; this benchmark assumes that economic efficiency from a societal perspective is the primary concern of decision-makers. However, multiple studies demonstrate that such a threshold may be inconsistent with societal preferences in the US; estimates of the willingness-to-pay threshold may approach or exceed $100,000/QALY30,31. Practicing clinicians have a different perspective. Nadler et al.28 suggest that oncologists are willing to prescribe regimens that cost much more for the QALYs gained. Regarding bevacizumab use, for example, physicians indicated an average implied willingness-to-pay threshold of $320,000/QALY. Many economists would also argue that the average threshold should be higher. For example, Murphy and Topel32 estimate the willingness-to-pay threshold peaks at $350,000 per life-year for a 50-year-old. Willingness-to-pay thresholds must be considered with the understanding that the extension of survival leads to a longer period during which patients incur healthcare expenditures. Therefore, it is important to consider the lifetime survival gain estimation of economic efficiency, as is done in this analysis.

The probabilistic sensitivity analysis results project that everolimus will be cost-effective compared to sunitinib under a range of acceptable willingness-to-pay thresholds. While further studies are warranted in broader populations, providers and decision-makers should take these results into consideration when treating patients with advanced, progressive pNET.

Conclusion

Although this analysis is limited due to its reliance on an indirect comparison of two phase 3 studies, everolimus is expected to be cost-effective relative to sunitinib in advanced, progressive pNET. Sensitivity analyses demonstrate that, overall, there is a trend that everolimus is cost-effective compared to sunitinib in patients with advanced, progressive pNET.

Transparency

Declaration of funding

Novartis Pharmaceuticals Corporation provided funding for this study. The sponsored research did not put limits on freedom to publish or the content of the publication.

Declaration of financial/other relationships

Roman Casciano, Maruit Chulikavit, and Allison Perrin are employees of The LA-SER Group, a consultancy that received compensation for the overall economic study design, the analysis, and preparation of this manuscript. Zhimei Liu and Xufang Wang are employees of and own stock in Novartis Pharmaceuticals Corporation. They also contributed to the analysis and manuscript preparation. Louis P. Garrison is a consultant for Novartis Pharmaceuticals Corporation and has received compensation for his contributions to analysis and the preparation of the manuscript.

Acknowledgments

No assistance in the preparation of this article to be declared. This study was presented in part (Figure 1 and Tables 1–3, 5, and 6) at the American Society of Clinical Oncology, 2012 Gastrointestinal Cancers Symposium; Science and Multidisciplinary Management of GI Malignancies; January 19–21, San Francisco, CA. The poster was titled, Cost-Effectiveness of Treating Patients with Advanced Progressive Pancreatic Neuroendocrine Tumors with Everolimus Versus Sunitinib in the United States.