University of Southern California Dissertations and Theses

Value in oncology care and opportunities for improvement

(USC Thesis Other)

Value in oncology care and opportunities for improvement

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

VALUE IN ONCOLOGY CARE AND OPPORTUNITIES FOR IMPROVEMENT

by

Xiaohan Hu

A Dissertation Presented to the
FACULTY OF THE GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the Requirements for the Degree
DOCTOR OF PHILOSOPHY
(HEALTH ECONOMICS)

December 2019

2
Value in Oncology Care and Opportunities for Improvement

ABSTRACT

This dissertation evaluates the value of oncology services in the current oncology lexicon from
different perspectives. As we see tremendous resources dedicated to treating cancer and the
growth trajectory is projected to continue, the value of the oncology care delivered has become a
crucial part of decision-making. Recent debates have focused on the value of new, innovative
oncology treatments as there has been a striking increase in cancer drug prices. The definition of
value may vary depending on the stakeholder involved, but nearly all shares the concept of
clinical improvement achieved relative to cost. Therefore, correctly measuring the clinical
benefits and costs associated with the oncology services is critical in recognizing the true value
in oncology care; in addition, discerning the determinants of improved care efficiency from
different levels, such as policy- or patient-level, could help identify effective solutions for
achieving better patient outcomes. This dissertation attempts to shed light on these issues by
examining the economic value of a recently approved immunotherapy and trend in value of
cancer treatments over time as well as by assessing the relationship between oncology spending
and patient survival at the clinical practice level.
Chapter I. evaluates the economic value of pembrolizumab, an immune check-point inhibitor,
which was approved as a first-line treatment for patients with PD-L1 positive, late-stage non-
small cell lung cancer. A Markov model was adopted to simulate the course of disease among
patients treated by pembrolizumab or traditional chemotherapy. Informed by the clinical data

3
from the phase III, randomized clinical trial and utility data from the late-stage lung cancer
patients in the real world, the model projected a life expectancy difference between
pembrolizumab and chemotherapy group of 1.32 years and a quality-adjusted life-year gain of
0.83 years. Cost parameters were sourced from published literature and government national
formulary. Over the projected lifetime, patients treated with pembrolizumab incurred £72,465
more than those treated with chemotherapy, yielding an incremental cost-effectiveness ratio of
£86,913 per quality-adjusted life-year gained. Sensitivity analyses revealed that a discount of
50% or more is needed in order to make pembrolizumab meet the UK health care system’s
willingness-to-pay threshold of £50,000 per quality-adjusted life-year gained. Although
pembrolizumab does not appear to be cost-effective at its current list price in the UK from this
analysis, it showed potential in providing patients with untreated PD-L1 positive advanced
NSCLC significant survival benefits. Taking into account the uncertainty around the relative
overall survival data associated with pembrolizumab as well as the fact that patients’ response to
immunotherapy vary in extent and timescale in the real world, risk-sharing contracts and other
forms of innovative payment schemes may provide an opportunity for all stakeholders to ensure
patient access under a constrained budget.
Chapter II. examines trends in economic value of cancer drugs approved between 1995 and
2017. The economic value was define using two measures of survival gain – price per median
life-year gained and price per mean life-year gained. The sources of the survival gain data were
survival differences reported in clinical trials and life expectancy differences projected in
economic models, i.e. cost-effectiveness or cost-utility analyses. Using a fractional polynomial
model, the adjusted cancer drug prices per median life-year gained was shown to have been
increasing until 2013. But from 2013 to 2017, the observed positive trend in the adjusted cancer

4
drug price per median life-year gained stopped. The adjusted price per mean life-year gained
exhibited a slower rise. By investigating the trends in price and life-year gained separately, it was
found that gains median life-year gained increased at a faster rate of 22% per year, which
suggested that median life-year gained had outstripped prices increases from 2013 to 2017.
Taken together, we find that the price per median life-year gained increased considerably more
than price per mean life-year gained among cancer drugs approved between 1995 and 2017.
These findings suggest that different measures of a drug’s clinical benefit may general different
implications for assessing value.
Chapter III. begins the investigation of variation of oncology spending at the practice level and
its association with patient survival. Data from a privately insured population with metastatic
cancer were used for this analysis. Substantial variation in oncology spending at the practice
level was demonstrated after accounting for differences in patient demographic and clinical
characteristics, suggesting impact from the practice-level factors. A model at the practice level
revealed that the use of targeted therapy and number of oncologist visit per patient were two
significant drivers of oncology spending at the practice level. Whether higher practice-level
spending is associated with better patient outcomes, including two-year all-cause mortality, five-
year all-cause mortality and survival duration, were then investigated. Results from the
generalized linear models showed that patients who were treated at practices in the quintile of
highest risk-adjusted practice-level spending had significantly lower two-year and five-year all-
cause mortality rates compared to those treated at practices in the quintile of lowest risk-adjusted
practice-level spending. Results from the Cox model further confirmed the findings by
demonstrating the patients treated in the highest risk-adjusted spending quintile experienced
significantly longer survival compared to those treated in the lowest risk-adjusted spending

5
quintile. While evidence is limited as to the full spectrum of potential contributing practice-level
factors, the use of targeted therapy was identified as one significant driver of higher oncology
expenditures. Given the ongoing discussion on the overuse or misuse of new, innovative
oncology drugs, the relationship between spending and cancer patient survival is an important
topic for investigation. As improved survival outcomes were observed in patients treated at
practices in highest-spending quintile, the survival differences might be partly explained by the
differences in the cancer treatment choice.

6
TABLE OF CONTENTS

ABSTRACT……………………………………………………………………………………….2
Chapter I. First-line pembrolizumab in PD-L1 positive non-small-cell lung cancer: a cost-
effectiveness analysis from the UK healthcare perspective……………………………………….8
Abstract……………………………………………………………………………………9
1. Introduction…….….….….….….….….….….….….….….….….….….….….….…...10
2. Material and methods…….….….….….….….….….….….….….….….….….….…...12
2.1. Model overview…….….….….….….….….….….….….….….….….….….12
2.2. Clinical parameters………………………………………………………….12
2.3. Cost inputs…………………………………………………………………..14
2.4. Sensitivity analyses………………………………………………………….16
3. Results…………………………………………………………………………………17
3.1. Base case results…………………………………………………………….17
3.2. Sensitivity analyses………………………………………………………….17
3.3. Scenario analyses……………………………………………………………18
4. Discussion……………………………………………………………………………..18
5. Tables and figures…………………………………………………………………….22
6. References……………………………………………………………………………..29
Chapter II. Trends in the Price per Median and Mean Life-Year Gained Among Newly Approved
Cancer Therapies 1995 to 2017………………………………………………………………….33
Abstract…………………………………………………………………………………..34
1. Introduction……………………………………………………………………………35
2. Study data and methods……………………………………………………………….37
2.1. Data………………………………………………………………………….37
2.2. Statistical analysis…………………………………………………………...40
3. Study results…………………………………………………………………………...41
3.1. Sample characteristics……………………………………………………….41
3.2. Overall trends………………………………………………………………..42
4. Discussion……………………………………………………………………………..45
5. Conclusion…………………………………………………………………………….48
6. Tables and figures……………………………………………………………………..49
7. Supplemental materials………………………………………………………………..55

7
8. References……………………………………………………………………………63
Chapter III. Variation in oncology spending at the practice-level and its association with patient
survival in a privately insured population with metastatic cancer……………………………….72
Abstract…………………………………………………………………………………..73
1. Introduction…………………………………………………………………………....75
2. Methods………………………………………………………………………………..76
2.1 Data…………………………………………………………………………..76
2.2 Study sample…………………………………………………………………77
2.3 Study outcomes……………………………………………………………....77
2.4 Statistical analysis……………………………………………………………78
3. Results…………………………………………………………………………………80
3.1. Sample characteristics……………………………………………………….80
3.2. Practice-level drivers of spending…………………………………………...80
3.3. Association between practice-level spending and patient outcomes………..81
4. Discussion……………………………………………………………………………..81
5. Tables and figures.………………………………………… …………………………86
6. References……………………………………………………………………………..93

8

Chapter I. First-line pembrolizumab in PD-L1 positive non-small-cell lung
cancer: a cost-effectiveness analysis from the UK healthcare perspective

Full citation: Hu X and Hay JW. First-line pembrolizumab in PD-L1 positive non-small-cell lung cancer: a cost-effectiveness
analysis from the UK healthcare perspective. Lung Cancer. 2018 Sep;123:166-171. doi: 10.1016/j.lungcan.2018.07.012. Epub
2018 Jul 11.

9
Abstract
Background: Pembrolizumab has shown significant survival benefits in treating chemotherapy-
naïve non-small-cell lung cancer patients (NSCLC) with increased level of PD-L1 expression.
This analysis aimed to evaluate the cost-effectiveness of pembrolizumab as a first-line treatment
for patients with PD-L1 positive NSCLC from the UK health care perspective.
Methods: A Markov model with progression-free, progressive disease and death states was
developed. Clinical parameters were informed by the KEYNOTE-024 trial. Utility values were
sourced from published literature. Cost data including drug acquisition costs, disease
management costs, and adverse event costs were derived from British National Formulary and
published literature. Model was run until 99% patients died. Both health outcomes and costs
were discounted at an annual rate of 3.5%. Deterministic and probabilistic sensitivity analyses
were performed to address the uncertainties around model parameters.
Results: In the base case, pembrolizumab is projected to increase patient’s life expectancy by
1.32 life-years over chemotherapy (2.45 vs. 1.13) and 0.83 QALYs (1.55 vs. 0.71) at an
additional cost of £72,465, yielding an incremental cost-effectiveness ratio of £86,913/QALY.
When parameters were varied in the deterministic sensitivity analyses, results are most sensitive
to duration of median overall survival in both groups.
Conclusion: Using a willingness-to-pay threshold of £50,000, pembrolizumab is not cost-
effective at its current list price and a discount of 50% or more is required for it to be cost-
effective comparing to commonly prescribed chemotherapy.

10
1. Introduction
Pembrolizumab is the first immune check-point inhibitor approved as an initial therapy for
patients with locally advanced or metastatic non-small-cell lung cancer (NSCLC) expressing
high level of PD-L1. Before the introduction of immune check-point inhibitors, first-line
treatment options for patients with locally advanced or metastatic NSCLC were primarily
cytotoxic chemotherapies. A small portion of patients who have mutations of epidermal growth
factor receptor (EGFR) or abnormal fusion of the anaplastic lymphoma kinase (ALK) are also
eligible for tyrosine kinase inhibitors (TKI). Pembrolizumab is a humanized monoclonal
antibody works by blocking the interaction between the programmed death 1 (PD-1) and
programmed death ligand 1 (PD-L1).
1,2,3
In January 2017, the European Medicine Agency
granted marketing authorization for pembrolizumab as a first-line treatment for adult patients
with locally advanced or metastatic NSCLC, whose tumors express PD-L1 tumor proportion
score of 50% or higher with no EGFR or ALK positive tumor mutations.
4

The incremental survival benefits and better safety profile associated with pembrolizumab was
demonstrated in KEYNOTE-024, an international, randomized, phase III clinical trial,
comparing pembrolizumab to platinum-based chemotherapy in patients with untreated PD-L1
positive NSCLC.
5
In October 2017, an updated analysis of KEYNOTE-024 trial reported data
from a longer follow-up time (median follow-up = 25.2 months). Over half of patients have had
overall survival events at time of the analysis in both groups and patients treated with
pembrolizumab had significantly longer overall survival compared to those treated with
chemotherapy.
6
Although only a small proportion of advanced NSCLC patients, approximately
23-28%,
2,3
may be eligible in the real-world, results from the clinical trial showed that it has the

11
potential to provide significant and durable benefits for this targeted patient population, which
could transfer into considerable survival benefits.
The development of this kind of new immunotherapies and targeted therapies marked the
exciting new era that cancer research has come into. The hefty price tags that come along with
these novel therapies, however, have posed a major challenge to the health care systems and
have thus attracted extensive criticisms from health policy makers globally. A 2015 study
analyzed the price trend of cancer drugs from 1995-2013 in the United States, and observed an
annual increase of 10% after adjusting for inflation and median survival benefits reported in the
clinical trials.
7
Although this type of studies signals the potential unsustainability of cancer drug
prices, benefits spanned over patient’s lifetime that cannot be demonstrated in clinical trials and
the value associated with the quality of life improvement still need to be accounted for. To
enable patient access and assess economic implications, cost effectiveness analyses are useful in
quantifying the societal benefits as well as the potential cost associated with the new therapies
through making use of the best available evidence.
In the UK, pembrolizumab is currently priced at £1,315 per 50 mg vial.
8
It will cost £5,260 per
cycle for each patient at its recommended dose of 200 mg per 21 days. Taking in to account the
huge unmet need for an effective treatment for metastatic NSCLC and increased patient survival,
this could transfer into significant financial burden on the UK health care system. After an initial
rejection due to concerns of high cost, pembrolizumab is now recommended by the National
Institute for Health and Care Excellence (NICE) for use within the Cancer Drugs Fund, under a
special patient access scheme in which the manufacturer will provide a discounted price to the
UK government.
9
In the manufacturer’s submission to NICE, an incremental cost-effectiveness
ratio (ICER) of £41,213 of pembrolizumab vs. chemotherapy was reported in the base case

12
analysis. However, the model was criticized for several key assumptions including the
extrapolation of overall survival data, choice of utility values, and a 2-year treatment stopping
rule in the pembrolizumab group.
9
Therefore, in this analysis, we sought to address some of these
issues and to further evaluate the cost-effectiveness of pembrolizumab in patients with untreated
PD-L1 positive NSCLC using the most recently reported survival data from a longer follow-up
of patients enrolled in KEYNOTE-024.

2. Material and methods
2.1 Model overview
We used a Markov model with three mutually exclusive health states, progression-free,
progressive disease, and death, to simulate the course of disease. Patients start receiving
pembrolizumab or chemotherapy in the progression-free state, and can stay in or move to
progressive disease or death state at a cycle length of 21 days based on their assigned transition
probabilities. Treatment decision following an event occurrence (progression or adverse event)
were based on the KEYNOTE-024 trial or published literature. Only direct medical costs were
considered in the model and are presented in 2017 British pounds. Both costs and health
outcomes were discounted an annual rate of 3.5% according to NICE guidelines. The model was
run until 99% of patients die.
The primary outputs of the model are quality-adjusted life-years (QALYs), total costs, and
ICERs. The model was developed and run in Excel 2013. (Microsoft Corporation, Redmond,
WA, USA)
2.2 Clinical parameters

13
Patient characteristics and treatment assignments were based on the KEYNOTE-024 trial.
5
We
calculated the between-state transition probability using the DEALE methods by assuming a
declining exponential function for survival.
10
The transition probabilities between progress-free
and progressive disease states were based on the median progression-free survival (PFS) reported
in KEYNOTE-024 (10.3 [6.7 – not reached] vs. 6.0 [4.2 – 6.2] months, pembrolizumab vs.
chemotherapy).
5
The probability of death is the sum of the probabilities of transitioning from
either state to death state, and was calculated based on the median OS data (30.0 [18.3 – not
reached] vs. 14.2 [9.8 – 19.0] months, pembrolizumab vs. chemotherapy).
6

Currently there is no clinical guideline on subsequent treatment options for patients failing
immune checkpoint inhibitors. Thus, we assumed that patients failing first-line pembrolizumab
received docetaxel as their second-line treatment. This assumption is based on the NHS clinical
guidance that docetaxel should be considered standard-of-care after disease progression on first-
line chemotherapy.
11
For patients on chemotherapy, in KEYNOTE-024 trial, those who failed
first-line chemotherapy could switch to receiving pembrolizumab upon physician’s approval as
their second-line treatment. During a median follow-up time of 25.2 months, 54.3% of patients in
the chemotherapy group crossed over to pembrolizumab. Therefore, in this model, we assumed
that 54.3% patients who progressed on first-line chemotherapy received pembrolizumab as their
second-line treatment, and the rest received docetaxel as their second-line treatment. Of note, this
may had diminished the incremental benefits of pembrolizumab as a first-line therapy in the
model, because pembrolizumab had been demonstrated to be superior to docetaxel as a treatment
of relapsed or recurrent NSCLC.
3
Nevertheless, we did not account for additional benefits from
the second-line pembrolizumab in the chemotherapy group, i.e. we attributed the survival
benefits from subsequent pembrolizumab to the first-line chemotherapy in the model. As

14
pembrolizumab is now recommended for advanced NSCLC patients who have failed a
chemotherapy, this likely reflects the real-world prescribing pattern in advanced NSCLC
management. Figure 1 presents the treatment assignment in this model.
Adverse events of ≥grade 3 and frequency ≥5% as reported in KEYNOTE-024 were included in
the model. Non-immune-mediated adverse events considered in the model included anemia,
neutropenia, decreased platelet count and thrombocytopenia. Immune-mediated adverse events
of grader 3 or higher were infrequent (<5%) in both pembrolizumab and chemotherapy groups in
the clinical trial, thus excluded from this analysis.
Utility values measuring individual’s quality of life were assigned to patients by health state as
well as by line of treatment. Utility values by line of treatment were obtained from published
literature.
12,13,14
Disutilities associated with adverse events were also applied based on the
frequency reported in KEYNOTE-024.
14,15
Utility values used in the base case are presented in
Table 1.
2.3 Cost inputs
Direct medical costs of treating the disease were considered in this model. Specific cost
components included drug acquisition cost, drug administration cost, disease management cost,
adverse event cost, PD-L1 testing cost, and terminal care cost.
For drug acquisition cost, resource use was based on the dosing schedule reported in
KEYNOTE-024. Pembrolizumab was administered at a dose of 200 mg every 3 weeks for 35
cycles in the base case. Chemotherapy group consisted of five platinum-based chemotherapy
regimens: carboplatin plus pemetrexed, cisplatin plus pemetrexed, carboplatin plus gemcitabine,
cisplatin plus gemcitabine, or carboplatin plus paclitaxel, and was administered every 3 weeks

15
for 4 to 6 cycles depending on specific regimen choice. Unit prices of the regimens were
obtained from the electronic market information tool (eMIT) for regimens other than
pembrolizumab and pemetrexed which were obtained from the British national formulary (BNF)
as they were not yet available in eMIT at the time of analysis.
8,16
Because patients randomized to
the chemotherapy group received one of the five chemotherapy regimens, the drug acquisition
cost for the chemotherapy group was calculated as a weighted average cost based on the number
of patients on each regimen in KEYNOTE-024. Patients’ body surface area (BSA) used to
calculate the dose needed for each cycle was based on the estimated average UK population’s
BSA and the gender distribution in KEYNOTE-024.
17
We assumed no vial-sharing in this
analysis as it was suggested that the practice of vial-sharing was still very limited in clinical
practice.
18

Administration costs were calculated according to the dosing and frequency of infusion from
drug labels and assumptions used in previous NICE guidance.
19,20,21
Unit costs were obtained
from the NHS reference costs by Healthcare Resource Group (HRG) code.
22

Patients incurred disease management costs in both progression-free and progressive disease
states. We assumed the same resource use for both arms by health state, and the resource use per
cycle were based on a systematic literature review of clinical effectiveness and cost-effectiveness
of chemotherapies in advanced NSCLC in a first-line setting.
23
Unit costs of individual resource
use were obtained from the NHS reference costs by Health Resource Group (HRG) code as well
as the Personal and Social Services Research Unit.
22,24

Costs associated with adverse events were computed based on the estimated incidence per cycle
and per event costs. Per event costs including medications, hospital visits and/or hospital stays

16
were obtained from published literature and previous NICE guidance and inflated to 2017 British
pounds.
15,23,25

Subsequent treatment was incorporated as a one-time cost including drug acquisition cost and
administration cost for both groups. We assumed the acquisition cost of docetaxel for
pembrolizumab for those receiving it as a second-line treatment in the chemotherapy group.
Terminal care costs were also included as a one-time cost in both groups. PD-L1 testing costs
were applied to the pembrolizumab group only.
9
Clinical and cost inputs used in the base case are listed in Table 2.
2.4 Sensitivity analyses
To determine the impact of each parameter on model outputs, one-way sensitivity analyses were
carried out. The value of parameters was varied one at a time by +/- 20% other than the discount
rate, for which we imposed a minimum value of 0 and a maximum value of 6.0%. Probabilistic
sensitivity analyses were also performed to further address the uncertainty in parameters and
evaluate the robustness of the model. All parameters were varied over its defined distribution
simultaneously each time the model was run. Beta distribution was used for values with a range
between 0 and 1, gamma distribution was used for median OS, median PFS and cost parameters,
and uniform distribution was used for discount rate. The model was run 1,000 times. The range
and distribution of the parameters used in the sensitivity analyses are presented in Table 2.
2.5 Scenario analyses
Finally, we also evaluated the ICER of pembrolizumab vs. chemotherapy in several scenarios
that likely reflected the real-world situations. First, current NICE guidance suggests that
pembrolizumab should be stopped after 35 cycles if there is no recorded disease progression or

17
intolerable toxicities. There is no direct evidence from the clinical trial, however, directly
supporting such stopping rule. Recommendations from NICE guidance were primarily based on
clinical experts’ and patient advocates’ opinion.
9
Therefore, in the first scenario, we removed this
restriction and allowed patients on pembrolizumab as long as they were progression-free and
experienced no adverse events. In the second scenario, we used BNF prices for all regimens
instead of prices from eMIT as in the base case. Lastly, to reflect the discounts of
pembrolizumab price offered by the manufacturer to the UK government, we applied several
common discounts to the list price of pembrolizumab and observed the direct effect of
pembrolizumab price on ICER, other factors constant.

3. Results
3.1 Base case results
In the base case, pembrolizumab yielded a survival benefit of 2.45 life-years (LYs) in PD-L1
positive metastatic NSCLC patients comparing to 1.13 LYs in the chemotherapy group,
providing an additional survival benefit of 1.32 LYs. After adjusting for quality of life, patients
in the pembrolizumab group survived 0.83 QALYs longer than the chemotherapy patients (1.54
QALYs vs. 0.71 QALYs). Total costs incurred was £92,833 in the pembrolizumab group and
£20,368 in the chemotherapy group, yielding a cost difference of £72,465. The ICER for
pembrolizumab vs. chemotherapy was £86,913 per QALY. (Table 3)
3.2 Sensitivity analyses
Results from the one-way sensitivity analyses are shown in the tornado diagram (Figure 2).
When varying each parameter individually, the median OS in the pembrolizumab group had the

18
greatest influence on the ICER, followed by the median OS in the chemotherapy group. ICER
also appeared to be sensitive to the utility values in the progression-free health state. Within the
+/- 20% range of each variable, ICER remained > £60,000 per QALY.
The average ICER computed from the 1,000 iterations run in probabilistic sensitivity analyses
was £78,440 per QALY, with an average QALY gain of 0.86 and an incremental cost of
£62,571. Figure 3 shows the cost-effectiveness acceptability curve. At a willingness-to-pay level
of £50,000 per QALY, the probability of pembrolizumab being cost-effective is approximately
29.4%, and this probability rose to 40.7% if the willingness-to-pay level is £70,000 per QALY.
3.3 Scenario analyses
In the scenario analyses, removal of the 35-cycle stopping rule increased ICER by £10,982 per
QALY to £97,895 per QALY. The impact of using BNF prices was minimal, only decreasing
ICER by £1,799 per QALY. The ICERs of applying a 25% or a 50% discount on pembrolizumab
list price resulted in an ICER of £67,756 and £48,600 per QALY, respectively.

4. Discussion
We developed a Markov model to evaluate the cost-effectiveness of pembrolizumab in untreated
PD-L1 positive, locally advanced or metastatic NSCLC from the UK health care perspective.
Using the current NHS willingness-to-pay threshold of £30,000 - £50,000 per QALY, results
from our analysis indicated that pembrolizumab is not cost-effective comparing to standard
platinum-based chemotherapies in this patient population at its current list price. Pembrolizumab
provided a survival benefit of 0.83 QALYs over chemotherapy at an additional cost of £72,465.
One-way sensitivity analyses revealed that the cost-effectiveness of pembrolizumab is most

19
sensitive to median OS in both groups, and the probabilistic sensitivity analyses showed that at a
willingness-to-pay threshold of £50,000, the probability of pembrolizumab being cost-effective
is approximately 29.4%. With a discount of 50% or more on the list price, other factors constant,
pembrolizumab can be a cost-effectiveness therapy comparing to chemotherapy for untreated
PD-L1 positive NSCLC patients.
The cost-effectiveness of pembrolizumab as a first-line therapy for PD-L1 positive metastatic
NSCLC was previously evaluated in the US setting from a third-party payer perspective.
26
The
US-based analysis found a 1.05-QALY gain associated with pembrolizumab at an additional cost
of $102,439 compared with chemotherapy, resulting in a base case ICER of $97,621 per QALY
over a 20-year time horizon. Using an ICER range of $100,000 to $150,000 per QALY
recommended by Neumann et al. in the US setting, the study concluded that pembrolizumab can
be considered cost-effective in treating PD-L1 positive NSCLC patients with no prior
chemotherapy.
26,27
Two factors may have contributed to this difference in the estimated QALY
gain between the US-based analysis and our analysis. Firstly, the US-based analysis adopted a
different model approach with ours – for the first 38 weeks with observed individual patient-
level data, a partitioned-survival model was fit to build the survival function for the first phase
and then an exponential distribution was used to model OS up to 5 years of follow-up. Our
analysis adopted a Markov model structure assuming a constant progression/mortality rate,
which was calculated from the aggregate survival data reported from a longer follow-up time of
25.2 months (approximately 109 weeks). Secondly, the utility values used in the US-based
analysis were based on the EQ-5D-3L data collected from the patients enrolled in KEYNOTE-
024 trial, whereas we sourced the utility values from real-world surveys of patients with
advanced NSCLC. Patients who enrolled in a clinical trial may differ from a real-world

20
population in several ways, therefore, the different utility values may explain part of the
difference QALY gain estimated in the two analyses.
One major strength of this analysis is that we used the most recently reported long-term survival
data from the KEYNOTE-024 trial after a median follow-up time of 25.2 months. Many
innovative drugs are approved based on immature but significantly superior survival data to
facilitate early patient access. Therefore, earlier cost-effectiveness analyses of pembrolizumab
extrapolated patient survival based on survival data at a shorter follow-up time, and thus the
results might be influenced by the model assumptions to a greater extent. Another strength is that
to adjust for life-years gained, we applied utility values that were obtained from the quality of
life data reported by those representative of the general patient population with advanced
NSCLC. Quality of life has become one of the primary endpoints in clinical trials that evaluate
the effectiveness of cancer treatment, as it measures the general well-being of patients. It is
particularly important for those with advanced diseases.
28
Using estimated utility values from a
general patient population allows the policy makers to make better inferences about the
effectiveness of pembrolizumab in a real-world situation.
There are several limitations of this analysis, mainly governed by data availability and model
assumptions. Firstly, we did not have access to patient-level data and thus we relied on the
aggregate survival data reported from the clinical trial to model patient survival. We performed
model validation in order to alleviate this problem. We compared the survival rates yielded by
the model to the survival rates reported by the clinical trial at 6 months, 12 months, and 18
months and ensured that similar rates were seen at these important time points. Secondly, there
was still uncertainty around the OS data from KEYNOTE-24 despite a longer follow-up time.
Although median OS was reached at 30.0 months for the pembrolizumab group, the average

21
QALY gain could be largely influenced by those who survive long. Given this uncertainty
around the extrapolation of patient overall survival, longer-term and/or real-world patient
survival data is still needed in order to model patient survival more accurately and to better
reflect the benefits associated with the treatment.
Using the findings from the clinical trial, we estimated that patients who have untreated locally
advanced and metastatic NSCLC and express high PD-L1 level could on average gain 0.83
QALYs over their lifetime at an additional cost of £72,465 if they received pembrolizumab as
their first therapy as opposed to platinum-base chemotherapy. Although pembrolizumab does not
appear to be cost-effective at its current list price in the UK from this analysis, it showed
potential in providing patients with untreated PD-L1 positive advanced NSCLC significant
survival benefits. Taking into account the uncertainty around the relative overall survival data
associated with pembrolizumab, risk-sharing contracts and other forms of innovative payment
schemes may provide an opportunity for all stakeholders to ensure patient access under
constrained budget. Studies like this could serve as a start of such discussion.

22
Table 1. Utility Inputs.
Progression-free (SE) Progressive disease (SE)
First-line pembrolizumab 0.71 (0.24) 0.67 (0.20)
First-line chemotherapy 0.68 (0.24) 0.67 (0.20)
Second-line
pembrolizumab
0.67 (0.18) 0.59 (0.34)
Second-line
chemotherapy
0.65 (0.18) 0.59 (0.34)

23
Table 2. Clinical and Cost Parameters.
Base-case
Value
Min Max 95% CI Reference for resource use/unit
cost
Clinical inputs
Median OS –
pembrolizumab (months)
30.0 24.0 36.0 (18.3 – not
reached)
Brahmer J, et al, 2017
6
Median OS –
chemotherapy (months)
14.2 11.4 17.04 (9.8 – 19.0) Brahmer J, et al, 2017
6
Median PFS –
pembrolizumab (months)
10.3 8.2 12.4 (6.7 – not
reached)
Reck M, et al, 2016
5
Median PFS –
chemotherapy (months)
6.0 4.8 7.2 (4.2 – 6.2) Reck M, et al, 2016
5
Cost per cycle
Drug acquisition cost
Pembrolizumab £5,260 £4,208 £6,312 (£4,229 - £6,291) Reck M, et al, 2016
5
; BNF
8
Chemotherapy* £1,133 £907 £1,360 (£911 - £1,355) Reck M, et al. 2016
5
; eMIT
16
; BNF
8
Drug administration cost
Pembrolizumab £184 £147 £220 (£148 -£219) Reck M, et al, 2016
5
; NHS reference
costs 2015-2016
22
Chemotherapy* £550 £449 £659 (£442 - £657) Reck M, et al, 2016
5
; NHS reference
costs 2015-2016
22

Docetaxel £383 £306 £460 (£307 - £458) Reck M, et al, 2016
5
; NHS reference
costs 2015-2016
22

Disease management cost
Progression-free state £121 £97 £145 (£97 - £145) Reck M, et al, 2016
5
; NHS reference
costs 2015-2016
22

24
Progressive-disease state £502 £402 £602 (£404 - £601) Reck M, et al, 2016
5
; NHS reference
costs 2015-2016
22

Adverse event cost
Pembrolizumab 0 0 0 Reck M, et al, 2016
5
; Brown T, et al,
2013
23
Chemotherapy £169 £135 £203 (£136 - £202) Brown et al, 2013
23
One-time cost
PD-L1 testing cost** £358 - - (£288 - £428) NICE TA428
25
; NICE TA 447
9
Subsequent treatment
cost***

Pembrolizumab £494 - - - eMIT
16
Chemotherapy £535 - - - eMIT
16

Terminal care cost £4261 - - (£3426 - £5096) NHS reference costs 2015-2016
22
*A weighted average of chemotherapy regimens administered in KEYNOTE-024.
**Pembrolizumab group only.
***Cost of second-line pembrolizumab is considered same as cost of docetaxel.

25
Table 3. Base Case Results.
Pembrolizumab Chemotherapy Incremental
Life-years 2.45 1.13 1.32
Quality-adjust life-years 1.54 0.71 0.83
Total costs £92,833 £20,368 £72,465
ICER £86,913 per QALY

26
Figure 1. Treatment Assignment by Line of Therapy.

27
Figure 2. Tornado Diagram.

£60,000 £70,000 £80,000 £90,000 £100,000 £110,000 £120,000 £130,000
Adverse events cost
Administration cost - chemotherapy
Administration cost - pembrolizumab
Drug acquistiion cost - chemotherapy
Discount rate
Utility - Progression-free, chemotherapy
Utility - Progressive disease, second-line docetaxel
Utility - Progression-free, pembrolizumab
Median progression-free survival - pembrolizumab
Drug acquisition cost - pembrolizumab
Median overall survival - chemotherapy
Median overall survival - pembrolizumab

28
Figure 3. Cost-Effectiveness Acceptability Curve.

0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
£20 £30 £40 £50 £60 £70 £80 £90 £100 £110 £120 £130 £140 £150 £160 £170 £180 £190 £200 £210 £220 £230 £240 £250 £260 £270 £280 £290 £300
Probabilityof
being cost-effective
Pembrolizumab Chemotherapy
(in1,000s)

29
References:
1. Chatterjee M, Turner DC, Felip E, et al. Systematic evaluation of pembrolizumab dosing in
patients with advanced nonsmall-cell lung cancer. Ann Oncol 2016;27: 1291-8.
2. Garon EB, Rizvi NA, Hui R, et al. Pembrolizumab for the treatment of non–small-cell lung
cancer. N Engl J Med 2015;372: 2018-28.
3. Herbst RS, Baas P, Kim DW, et al. Pembrolizumab versus docetaxel for previously treated,
PD-L1-positive, advanced nonsmall-cell lung cancer (KEYNOTE-010): a randomised
controlled trial. Lancet 2016; 387: 1540-50.
4. European Medicines Agency.
http://www.ema.europa.eu/ema/index.jsp?curl=pages/medicines/human/medicines/003820/h
uman_med_001886.jsp&mid=WC0b01ac058001d124. Accessed Jan 2018.
5. Reck M, Rodríguez-Abreu D, Robinson AG, et al. Pembrolizumab versus Chemotherapy for
PD-L1-Positive Non-Small-Cell Lung Cancer. N Engl J Med. 2016 Nov 10;375(19):1823-
1833.
6. Brahmer J, Rodriguez-Abreu D, Robinson A, et al. Updated Analysis of KEYNOTE-024:
Pembrolizumab vs Platinum-Based Chemotherapy for Advanced NSCLC With PD-L1 TPS
>= 50%. Journal of Thoracic Oncology. Vol. 12 No. 11S2.
7. Howard DH, Bach PB, Berndt ER, et al. Pricing in the Market for Anticancer Drugs. J Econ
Perspect. 2015;29(1):139-62.
8. British Medical Association and Royal Pharmaceutical Society of Great Britain. British
national Formulary. Accessed October 2017.

30
9. National Institute for Health and Care Excellence (2017). Final appraisal determination –
Pembrolizumab for untreated PD-L1-positive metastatic non-small-cell lung cancer.
Technology appraisal guidance [TA447]. Published date: 28 June 2017.
10. Beck JR, Kassirer JP, Pauker SG. A convenient approximation of life expectancy (the
"DEALE"). I. Validation of the method. Am J Med. 1982 Dec;73(6):883-8.
11. National Institute for Health and Care Excellence (2011). Lung cancer: diagnosis and
management. Clinical guideline [CG121]. Published date: April 2011.
12. Brown J, Cook K, Adamski K, et al. Utility values associated with advanced or metastatic
non-small cell lung cancer: data needs for economic modeling. Expert Rev Pharmacoecon
Outcomes Res. 2017 Apr;17(2):153-164.
13. Chouaid C, Agulnik J, Goker E, et al. Health-related quality of life and utility in patients with
advanced non-small-cell lung cancer: a prospective cross-sectional patient survey in a real-
world setting. J Thorac Oncol. 2013 Aug;8(8):997-1003.
14. Nafees B, Stafford M, Garviel S, et al. Health state utilities for non small cell lung cancer.
Health Qual Life Outcomes. 2008; 6: 84.
15. National Institute for Health and Care Excellence (2016). Crizotinib for untreated anaplastic
lymphoma kinase-positive advanced non-small-cell lung cancer. Technology appraisal
guidance [TA406]. Published 28 September 2016.
16. Electronic market information tool (eMIT), Commercial Medicines Unit, Department of
Health. Electronic market information tool.
https://www.gov.uk/government/publications/drugs-and-pharmaceutical-electronic-market-
information-emit. Accessed October 2017.

31
17. Sacco JJ, Botten J, Macbeth F, et al. The average body surface area of adult cancer patients
in the UK: a multicentre retrospective study. PLoS One. 2010 Jan 28;5(1):e8933.
18. Bach PB, Conti RM, Muller RJ, Schnorr GC, Saltz LB. Overspending driven by oversized
single dose vials of cancer drugs. BMJ. 2016 Feb 29;352:i788.
19. National Institute for Health and Care Excellence (2009). Pemetrexed for the first-line
treatment of non-small-cell lung cancer. Technology appraisal guidance [TA181]. Published
date: 23 September 2009.
20. National Institute for Health and Care Excellence (2010). Gefitinib for the first-line treatment
of locally advanced for metastatic non-small-cell lung cancer. Technology appraisal guidance
[TA192]. Published date: 28 July 2010.
21. National Institute for Health and Care Excellence (2015). Pembrolizumab for treating
advanced melanoma after disease progression with ipilimumab. Technology appraisal
guidance [TA357]. Published date: 07 October 2015.
22. Department of Health. NHS reference costs 2015–2016.
https://www.gov.uk/government/publications/nhs-reference-costs-2015-to-2016. Accessed
August 2017.
23. Brown T, Pilkington G, Bagust A, Boland A, Oyee J, Tudur Smith C, et al. Clinical
effectiveness and cost-effectiveness of first-line chemotherapy for adult patients with locally
advanced or metastatic non-small cell lung cancer: a systematic review and economic
evaluation. Health Technol Assess 2013;17(31).
24. Personal Social Services Research Unit. Unit Costs of Health and Social Care 2016.
http://www.pssru.ac.uk/project-pages/unit-costs/unit-costs-2016/. Accessed October 2017.

32
25. National Institute for Health and Care Excellence (2017). Pembrolizumab for treating PD-
L1-positive non-small-cell lung cancer after chemotherapy. Technology appraisal guidance
[TA428]. Published date: 11 January 2017.
26. Huang M, Lou Y, Pellissier J, et al. Cost Effectiveness of Pembrolizumab vs. Standard-of-
Care Chemotherapy as First-Line Treatment for Metastatic NSCLC that Expresses High
Levels of PD-L1 in the United States. PharmacoEconomics (2017) 35:831–844.
27. Neumann PJ, Cohen JT, Weinstein MC. Updating cost-effectiveness–the curious resilience
of the 50,000-per-QALY threshold. N Engl J Med. 2014;371(9):796–7.
28. Bottomley A, Flechtner H, Efficace F, Vanvoorden V, Coens C, Therasse P, et al. Health
related quality of life outcomes in cancer clinical trials. Eur J Cancer. 2005;41:1697–1709.

33

Chapter II. Trends in the Price per Median and Mean Life-Year Gained
Among Newly Approved Cancer Therapies 1995 to 2017

Note: This Chapter has been accepted for publication (currently in press).

34
Abstract
Background: The prices of newly approved cancer drugs have risen over the past decades. A key
policy question is whether the clinical gains offered by these drugs in treating specific cancer
indications justify the price increases.
Objectives: To evaluate the price per median and mean life year gained among newly approved
cancer therapies from 1995 to 2017.
Methods: We collected data on the price (in 2017 USD) per life-year gained among cancer drug-
indication pairs approved by the US Food and Drug Administration (FDA) between 1995 and
2017. We modeled trends using fractional polynomial and linear spline regression models that
controlled for route of administration and cancer type fixed effects.
Results: We found that between 1995 and 2012, price increases outstripped median survival
gains, a finding consistent with previous literature. Nevertheless, price per mean life-year gained
increased at a considerably slower rate, suggesting that new drugs have been more effective in
achieving longer-term survival. Between 2013 and 2017, price increases reflected equally large
gains in median and mean survival, resulting in a flat profile for benefit-adjusted launch prices in
recent years.
Conclusions: Although drug costs have been rising more rapidly than median survival gains, they
have been rising at about the same rate as mean survival gains. This suggests that when
accounting for longer-term survival gains, the benefits of new drugs are roughly keeping pace
with their costs, despite rapid cost growth.

35
Introduction
High and rising launch prices of new cancer drugs have raised American public stakeholder and
policy concern.
1-3
For example, Provenge (sipuleucel-T), an immunotherapy for prostate cancer,
launched in 2010 at $93,000 for a complete course of treatment.
4
A year later in 2011, Yervoy®
(ipilimumab) for metastatic melanoma was introduced at $120,000 per year of treatment, and in
2014, Blincyto® (blinatumomab) for lymphoblastic leukemia was priced at $178,000 per year of
treatment.
5,6

While prices are high and appear to have grown over time, these drugs may also result in better
health as measured by incremental survival or other health gains for those taking them.
7-9

One previous study examined trends in the price of health associated with new cancer drugs.
Howard and coauthors considered trends in the price per health gained among all cancer drugs
approved in the US between 1997 and 2013.
10
They found that the price per life year gained
increased by approximately $8,500 per year.
As the measure of health gained, Howard and coauthors used median survival reported in the
clinical trials accepted by the FDA for approval. Median survival is a clinical trial endpoint that
is met when 50% of the trial sample has died. It is often considered the most reliable cancer
endpoint because it is precise and unambiguously documented by the date of death.
11
However,
median survival estimates may mask gains in longer-term survival, thereby representing an
underestimate of the true survival gain.
12

Unlike median survival, mean survival estimates capture the presence of longer survival gains by
taking into account the survival-curve distribution after 50% survival has been achieved, known
as the right-tail of survival. Capturing longer-term benefits of survival can be particularly

36
valuable as existing evidence suggests that for decisions involving poor odds of survival, each
additional day of survival is worth more than the last.
13,14
Moreover, the importance of these
longer-term gains has been acknowledged by several institutions, including the American
Society of Clinical Oncology (ASCO), the Institute for Clinical and Economic Review (ICER),
and the UK’s National Institute for Health and Care Excellence (NICE). In their scoring of net
health benefits, ASCO awards bonus points to cancer drugs that generate meaningful survival
gains at a time point that is twice the median.
15
Endorsed by the International Society for
Phamacoeconomics and Outcomes Research (ISPOR), ICER’s current value framework assesses
potential clinical gains over the lifetime of patients eligible for treatment with new drugs and
includes consideration for the presence of right-tailed survival gains reported in supporting
studies.
16,17
The presence of potential right-tailed survival gains was an important feature in the
institute’s assessment of the effectiveness and value of Poly ADP-Ribose Polymerase (PARP)
Inhibitors.
18
In 80% of cancer drug assessments, NICE incorporated mean survival estimates into
their reimbursement decisions, whereas median survival was incorporated in only 36% of
cases.
12

Estimating mean survival requires information on the survival times for all individuals, but those
times are generally unobserved. As a result, statistical models are used to approximate the
survival distribution and extrapolate findings to the point at which all patients die.
19,20
Mean
estimates are calculated by fitting the existing data to parametric models that can differ by
assumed distribution (e.g., exponential, log-normal, etc.).
12,21
While different models are
appropriate in different contexts, model assumptions can affect survival estimates.
12

Consequently, guidelines issued by ISPOR and NICE suggest that publications should provide
justification for the specific extrapolation approach used and demonstrate that the modeling

37
choice meets accepted best fit criteria.
17,22
Quality of mean survival estimates can also differ
depending on the underlying data used. We focused on mean survival estimates which have been
extrapolated from clinical trials data since they are considered more rigorous than extrapolations
based solely on patient-level claims data.
22-24

In this paper, we examine trends observed between 1995 and 2017 in the price of health
associated with new cancer drug-indication pairs. Given the differing benefits between median
and mean survival estimates, we employed both measures and test for differences. Unlike
previous papers, we also included in our sample first and follow-on cancer indications and test
for model robustness for inclusion of the latter. We believe this is a useful exercise since cancer
drug prices are generally uniform across indications, accounting for follow-on approvals may
offer a more complete understanding of drug prices relative to survival gains.
25,26

We used a fractional polynomial model to flexibly estimate trends over time and identify
possible breaks in trend over the 24 year study period. Based on detected changes in trend
around 2013, we estimated separate trends from 1995 to 2012 and 2013 to 2017. The latter time
period has not yet been analyzed by existing studies. It is also an interesting focus given
significant stakeholder scrutiny on the high prices of recently launched cancer drugs.
10,14,15,22,27-38

Study Data and Methods
Data. We constructed a dataset of approved cancer drugs and their approved indications by
month-year, route of administration, median and mean survival gains, and prices from 1995 to
2017. We briefly describe the data—which is at the drug-indication level—here and document
additional details in the Supplementary Appendix Section A. Figure 1 provides a flow chart that
illustrates the data collection process.

38
First, we used Centerwatch—a source cited by several studies on drug innovation—to construct a
list of cancer drugs approved between January 1995 and May 2017.
10,39,40
We excluded drugs
treating only cancer-related symptoms or treatment side effects (e.g. diarrhea, anemia, etc.). For
each drug, we used FDA approved drug labels to identify the relevant cancer indications, defined
as a combination of cancer type and line of therapy (e.g., first line treatment of advanced breast
cancer). Approval dates for each drug-indication pair were checked against those listed in the
FDA Orange Book and the National Cancer Institute. The route of administration for each cancer
drug-indication pair was collected from the FDA Orange Book.
We examine efficacy using measures of life years gained (LYG). Median data were obtained
from published trials; these trials were referenced by the FDA drug label in its “Full Prescribing
Information” section. We calculated survival gains by differencing median overall survival
between treatment and comparator arms.
When overall median survival was not reported, we used progression-free median survival,
defined as the period of time the cancer is under control. Our use of this alternative metric is
consistent with that in previously published studies.
10
These progression-free survival estimates
are used by the FDA as a basis for approval, and they provide a useful signal of quality to the
manufacturer who must set prices for a new drug in the absence of data on overall survival.
Moreover, previously published studies suggest that progression-free survival is highly
correlated with overall survival especially in cancers with short survival post-progression.
41-43

When neither overall nor progression-free survival was available—for instance, when only a
symptom-based endpoint was reached (e.g., hazard ratios, complete or partial response rates,
cytogenetic responses rates) or when drug-indications were approved on the basis of single-arm
trials, we obtained estimates of median LYG by assuming an exponentially distributed survival

39
curve—which has been well-validated by the literature—and dividing mean survival estimates
by ln(2).
10,44-46

Mean LYG were obtained from a search of previously published economic evaluations reported
in the Tufts Cost Effectiveness Analysis (CEA) Registry.
47,48
Continuously updated since 1976,
the Tufts CEA registry provides a comprehensive database of 5,655 CEAs across various
diseases, and the data has been used in over 50 peer-reviewed publications.
49
All mean LYG
estimates were obtained from extrapolations of the corresponding clinical trials used for drug
approval, and the results of these extrapolations were published in peer-reviewed journals. To
further ensure a consistent comparison between median and mean LYG estimates, we selected
mean LYG estimates from models that compared the drug therapy to the same comparator as
those used in our median survival data. We also utilized progression-free-survival mean LYG
estimates when a progression-free-survival measure was used in the median LYG data. If no
economic modeling could be identified, the drug-indication pair was dropped.
Finally, we calculated a “treatment-episode price” for each indication-drug pair in the sample by
multiplying the drug’s monthly price by the typical duration of treatment for the specific
indication.
22,50,51
Detailed in the Supplemental Materials and Exhibit A2, we followed the
literature by calculating each drug’s monthly price, at the time of approval of earliest price
observed following approval, to the Medicare program (i.e., Medicare reimbursements).
52,53
All
prices were inflation-adjusted and expressed in 2017 USD. These prices represent the actual
dollar amounts that Medicare, the largest public insurer, pays for drugs. In most cases, these
prices will be greater than the prices hospitals, physicians, and pharmacies pay to acquire these
drugs.
10

40
Statistical Analysis. We computed the price per median and mean LYG for each drug-indication
and analyzed trends in these variables using two approaches. We first estimated the trends using
a fractional polynomial model, which allowed us to flexibly parameterize trends without
imposing a pre-determined specification.
54
We considered 44 different degree-2 polynomials and
selected the model with the lowest deviance. All models additionally controlled for route of
administration and cancer type fixed effects (16 categories).
55
Details regarding all models tested
are provided in the Supplementary Appendix Section B.
Then, informed by the results of the fractional polynomial model, we then estimated a linear
spline regression model with a knot (i.e., a point of separation in the piecewise regression
system) at 2013. The knot allowed slopes to differ between 1995 to 2012 and 2013 to 2017. The
linear spline models for drug-indication d in year t is of the form:
𝑌
"#
= 𝛽
&
+𝛽
(
min (𝑇𝑖𝑚𝑒,18)+𝛽
6
min(0,𝑇𝑖𝑚𝑒−18)+𝛼1(𝑂𝑟𝑎l)
>?
+𝜂
A(")
+𝜖
"#

𝑌
"#
measures either the price per median LYG or price per mean LYG. 𝑇𝑖𝑚𝑒 equals the approval
date minus 1995, so times 0, 18, and 22 correspond to January 1
st
in 1995, 2013, and 2017,
respectively. Again, we additionally controlled for route of administration (1(𝑂𝑟𝑎l)
>?
) and 16
cancer type fixed effects (𝜂
A(")
). Thus, 𝛽
(
identifies the average annual change in price per
median or mean LYG from 1995 to 2013, and 𝛽
6
identifies the average annual change in price
per median or mean LYG from 2013 to 2017.
To identify whether changes in trend were driven by prices changes or changes in survival gain,
we estimated the linear spline regression model with price, median LYG, and mean LYG as
separate dependent variables (𝑌
"#
).

41
We tested the sensitivity of our estimates by examining several alternative price-per-LYG
metrics. In addition to dropping outliers and data points approximated by dividing mean survival
by ln(2), we considered specifications using logged price per LYG and monthly Medicare price
per LYG (which ignored the duration of treatment episodes). To assess whether changes in trend
from 1995 to 2012 and 2013 to 2017 were associated with the 2012 budget sequester which
reduced Part B reimbursements starting April 2013 from average sales price (ASP) plus 6% to
ASP plus 4.3%, we additionally examined a specification that kept the part B pricing formula
constant (e.g. ASP plus 6%). If results persisted in that specification, then they could not be
driven by the change in pricing policy.
56
Finally, abstracting from fluctuations in Medicare
pricing policies, we estimated our results using the average wholesale price (AWP) per LYG,
with AWP data from the earliest post-approval date observed in IBM’s Micromedex Red Bok.
57

All analyses were performed using the statistical software Stata, version 15.1, and estimates with
p<0.05 using two-tailed tests were considered significant.

Study Results
Sample Characteristics. During the study period, the FDA approved 91 distinct cancer drugs, of
which 38 (41.8%) were approved for a single indication and 53 (58.2%) for multiple indications.
In total, our sample included 181 cancer drug-indication approved pairs. Table 1 shows each
drug-indication included and provides characteristics of our analytic sample. Annual counts of
drug-indication pairs are shown in Supplementary Appendix Figure A1.
The average treatment-episode price ranged from $25,274 (standard deviation: $28,828) for
pancreatic cancer to $171,448 (standard deviation: $73,835) for thyroid cancer. Pancreatic cancer
also had the lowest monthly price at $4,222 (standard deviation: $3,439), but due to differences

42
in treatment duration, the indication with the highest monthly price was the “other” category,
which includes indications with few approvals over the study period. Within the “other”
category, high prices were driven by dinutuximab for the treatment of neuroblastomas (with a
monthly price of $84,561) and polifeprosan with carmustine implant for the treatment of brain
cancer (with a monthly price of $51,621).
Across all drug-indications, the average treatment-episode price was $82,260 (standard
deviation: $100,174), and average median and mean LYG estimates were 0.56 and 0.91 years
(standard deviations: 0.63 and 0.98), respectively. After dividing the treatment-episode price by
the mean and median LYG for each drug-indication pair, we found that the average price per
median and mean LYG were $195,355 (standard deviation: 196,748) and $131,889 (standard
deviation: $149,259), respectively.
Overall Trends. Estimates from the fractional polynomial indicated that the real price per
median LYG, adjusted for drug indication and route, had been increasing until approximately
2013, when the trend changed (Figure 2). Specifically, from 1995 to 2013, the adjusted real
price per median LYG rose from $81,648 per LYG to almost $176,867 per LYG. But from 2013
to 2017, the adjusted price per median LYG fell to $126,950 per LYG. The observed break in
trend in price per median LYG is apparent when examining both first and follow-on indications
(Supplementary Appendix Figure A2).
The adjusted price per mean LYG exhibited a slower rise. The price per mean LYG initially
declined, perhaps due to the small sample sizes in earlier years. From 2002 to 2012, the adjusted
price per mean LYG increased from $108,250 per LYG to $129,798 per LYG, and it continued
to increase to $143,891 per LYG in 2017. While there is no clear break in trend in the price per

43
mean LYG across all drug-indications, Supplementary Appendix Figure A1 suggests a break
in trend in 2013 did occur among the price per mean LYG for follow-on indications.
The fractional polynomial estimates suggested that changes in trend can be approximated by a
linear spline regression model that allowed for a change in trend around 2013 (Figure 2). The
adjusted price per median LYG of a drug increased by $6,129 (p = 0.14) each year from 1995 to
2012 (Table 2, Panel A), whereas the adjusted price per mean LYG increased at a markedly
slower rate of $476 (p=0.88) per year.
The positive trend, in particularly the price per median LYG, slowed in recent years. From 2013
to 2017, neither the adjusted price per median LYG ($10,488 reduction per year, p=0.44) nor the
adjusted price per mean LYG ($412 increase per year, p=0.97) increased significantly.
These findings remained robust to several alternative price metrics (Table 3). Our results were
not driven by outliers (i.e., dropping two observations that, as Figure 3 indicates, are clear
outliers), and they persist even when we consider only trials that report overall or progression-
free survival. The natural logarithm specification indicated a 10% (p<0.001) annual increase in
price per median LYG from 1995 to 2012, and a slower price per mean LYG growth at 8.8%
(p=0.001) a year. In the recent period from 2013 to 2017, price per median LYG fell by 16%
(p=0.04) a year, and price per mean LYG fell by 12% (p=0.17) a year.
The recent halt in price-per-LYG growth was also not due to the 2012 budget sequester: our
results persisted in the specification that ignored the budget sequester. Moreover, estimates in the
recent period continued to be negative and statistically insignificant when using monthly price
per LYG and Red Book derived AWP per LYG measures.

44
While market dynamics surrounding first indications can differ from follow-on indications, our
findings are robust to focusing only on first-indications: the growth in price per median and
mean LYG between 2013 and 2017 was negative and statistically insignificant when examining
only first indications (Table 2, Panel B).
Finally, we show that our results are robust to adding additional knots in the spline regression. In
Appendix Table A3, we show results from adding a knot in 2003 to account for possible
changes in the price per mean LYG in the earlier period. We also consider knots for each 5-year
time period (e.g., knots in 2000, 2005, 2010, and 2015). While some imprecision in results are
introduced, estimates in the recent period are persistently negative and statistically insignificant.
Table 2 also shows the trends in price and LYG separately. During the recent period, launch
prices—that have not been benefit-adjusted—rose at a rate of 18.5% per year ($15,274 per year
from a mean of $82,260, p=0.016). In contrast, gains in median survival increased at a faster rate
of 22% per year (0.12 LYG from a mean of 0.55, p=0.001). These findings suggest that on
average, median survival gains have outstripped price increases from 2013 to 2017. Mean
survival increased at 16% per year (0.15 LYG from a mean of 0.91, p=0.017). These
observations again hold when examining only first indications. From 2013 to 2017, the median
survival gain increased by 41% per year (0.24 LYG from a mean of 0.59, p=0.004), whereas the
price increased by a statistically insignificant 13% per year ($12,845 from a mean of $82,261,
p=0.31).

45
Discussion
This study is the first to report trends in both price per median and mean LYG among newly
approved cancer therapies launched in the U.S. between 1995 and 2017. We find the price per
median LYG increased considerably more than the price per mean LYG among these drugs
during this study period. These findings suggest different measures of a drug’s clinical benefit
may generate different implications for assessing value.
Our results on trends in the price per median LYG prior to 2013 are consistent with David
Howard and coauthors, who report an annual increase of $8,500 (in 2013 USD) in the price per
median LYG.
10
Among first-approved indications, we find that the price per median LYG
increased at $8,942 (in 2017 USD) per year, and our estimate of a 10% annual increase in price
per median LYG is consistent with their findings. Our study is the first to extend the Howard
analysis to later years. We find in recent years the growth in price per median LYG has halted,
and this slowdown appears to be driven by larger gains in median survival relative to treatment-
episode prices. Consequently, our results suggest that on average, cancer drug prices have
become more aligned with their clinical benefits.
The more stable trends in launch prices we estimate for 2013 onwards may in part reflect the
increasing prevalence of oral cancer drug launches.
55
From a pharmaceutical manufacturer’s
perspective, pricing incentives operate differently across formulations and payer benefit
designs.
58
Non-oral infused and injected cancer drugs are typically reimbursed under a payer’s
medical benefit (‘Part B’ in fee for service Medicare), while oral cancer drugs are typically
reimbursed under a payer’s pharmacy benefit (‘Part D’ in fee for service Medicare).
59
Drugs
covered under each benefit are subject to automatic 340B drug discounts if administered or
dispensed by qualified providers, and if not approved with an orphan designation.
60
Drug use

46
qualifying for these discounts have increased over the time period of our study.
61
In addition,
branded pharmacy benefit covered oral drugs are subject to rebates, which provide funds paid
from manufacturers to Pharmacy Benefit Managers (PBMs) and payers as a reward for high
volume use and more favored formulary tier placement.
58
Unlike 340B discounts, the presence of
rebates feeds back to the setting of Medicare reimbursement for Part D covered drugs
10
and the
ability of PBMs to extract rebates increases as more drugs become available to treat a specific
cancer. In our analysis, sequential entry in oral cancer drugs was observed in numerous cancer
indications and therapeutic classes, suggesting more innovation may have acted to help flatten
out the actual prices Medicare paid for these therapies. The incremental impact of branded
competition within therapeutic class and by disease indication on Medicare’s reimbursement for
Part D covered drugs is an important area of future empirical study.
We also found cancer drugs launched after 2012 showed marked increases in both median and
mean LYG, compared to those launched previously. It is possible that this finding is related to
the regulatory changes that Congress enacted, coincident with this time period. Since 2012,
recommendations from the Institute of Medicine have emphasized the need to develop “high-
value” drugs which significantly extend survival and/or substantially improve quality of life.
30,31

Starting with the US Congress’s July 2012 enactment of the FDA Safety and Innovation Act
(FDASIA), the FDA has expedited the approval of drugs with evidence of substantial
improvements.
32
FDASIA allowed FDA to approve drugs based on surrogate or intermediate
clinical endpoints that were clinically meaningful, defined as being “reasonably likely to predict
clinical benefit.” It also established a new drug review pathway for those drugs meeting specific
endpoints in the Breakthrough Therapy Program.
32
43% of the cancer drugs in our sample
approved between 2012 and 2017 received the ‘breakthrough therapy’ designation and they had

47
survival gains that were more than double that of non-breakthrough drugs (although the
difference had limited statistical significance, p=0.11).
62
Future research should empirically
examine whether FDASIA’s implementation contributed to survival gains among cancer drug
approvals in recent years.
33-37,63,64

Limitations. Our analysis has a number of limitations. First, we used Medicare prices, as
opposed to provider acquisition prices, to identify drug costs. Acquisition costs are not publicly
reported in the U.S. health care system and nor do they accurately reflect the actual prices
patients and their insurers pay for them.
Second, we were unable to capture mean LYG estimates for all drug-indication approvals. Since
data availability is dependent on having a sufficient follow-up study post-FDA approval, it is not
surprising we were missing mean LYG data for a larger share of drug-indications in the more
recent years (15 of the 26 dropped observations occurred in the 2013 to 2017 period).
Third, we do not assess the validity of extrapolations among mean LYG estimates, and it is
possible that the accuracy of estimates improved over time as alternative survival modeling
methods were adopted. Mean LYG estimates can be sensitive to model choice, and the fit of
alternative models, sensitivity analyses, and levels of parameter uncertainty should be assessed.
12

While such metrics exist in the majority of studies that we follow, the lack of standardization
across metrics precludes us from analyzing how potential error in modeling affects our estimates.
Nevertheless, it is important to note that our results regarding the slow-down in price per median
LYG persist and are robust to the inclusion of all drug-indication pairs for which we have
median LYG data.

48
Fourth, we focused solely on the absolute survival gain and did not account for patient quality of
life, work gains or other elements of potential value entailed by these innovations. Many of these
measures are not routinely collected nor reported in clinical trials accepted by the FDA for new
drug approval. Standardizing these elements’ measurement and collection across cancer clinical
trials and real world outcome studies is an important future research goal furthering many
stakeholder aims including improved value assessments.
27,65-67

Conclusion
From 1995 to 2012, the price per mean LYG increased at a considerably slower rate than price
per median LYG, suggesting that new cancer drugs have been more effective in increasing long-
term survival. In recent years, benefit-adjusted launch prices have remained relatively stable, due
to price increases reflecting equally large gains in median and mean survival. In fact, from 2013
to 2017, the price per median LYG and price per mean LYG both fell, though the drop was only
statistically significant for the price per median LYG.

49
Table 1. Treatment-Episode Price and Life Year Gained by Cancer Indication, 1995-2017
Indication Treatment-
Episode Price
Monthly
Price
Median
LYG
Mean
LYG
Drug Names
Breast $54,255 $4,762 0.47 0.55 Ado-trastuzumab, anastrozole, capecitabine,
docetaxel, eribulin mesylate, everolimus,
exemestane, fulvestrant, gemcitabine,
ixabepilone, lapatinib, letrozole, palbociclib,
pertuzumab, ribociclib, trastuzumab
Cervical $29,368 $6,125 0.22 0.27 Bevacizumab, topotecan hydrochloride
Colorectal $36,007 $7,412 0.23 0.36 Bevacizumab, capecitabine, cetuximab,
irinotecan, oxaliplatin, panitumumab,
ramucirumab, regorafenib, trifluridine and
tipiracil, ziv-aflibercept
Gastric $90,956 $8,183 0.36 0.81 Docetaxel, everolimus, imatinib mesylate,
ramucirumab, regorafenib, sunitinib,
trastuzumab
Head and neck $70,561 $9,967 0.61 0.89 Cetuximab, docetaxel, nivolumab
Kidney $78,531 $10,638 0.33 0.562 Axitinib, bevacizumab, cabozantinib,
everolimus, lenvatinib, nivolumab, sorafenib,
sunitinib, temsirolimus
Leukemia $149,247 $13,974 1.30 1.82 Alemtuzumab, Bendamustine hydrochloride,
dasatinib, ibrutinib, idelalisib, imatinib
mesylate, midostaurin, nelarabine, nilotinib,
obinutuzumab, ofatumumab, rituximab
Lung $74,005 $9,514 0.31 0.52 Afatinib, alectinib, atezolizumab,
bevacizumab, ceritinib, docetaxel, erlotinib,
gefitinib, gemcitabine, nectinumumab,
nivolumab, osimetrinib, pembrolizumab,
pemetrexed, ramucirumab, topotecan
hydrochloride
Lymphoma $95,025 $9,074 1.02 1.4 Bendamustine hydrochloride, bevacizumab,
bortezomib, ibritumomab, ibrutinib, idelalisib,
nivolumab, rituximab, tositumomab
Melanoma $71,440 $13,400 0.35 1.20 Cobimetinib, dabrafenib, ipilimumab,
nivolumab, pembrolizumab, trametinib,
vemurafenib
Myeloma $127,361 $11,517 0.55 1.36 Bortezomib, daratumumab, elotuzumab,
ixazomib, panobinostat, pomalidomide
Other $69,291 $18,174 0.58 0.99 Avelumab, dinutuximab, nivolumab,
olaratumab, pembrolizumab, pemetrexed,
polifeprosan w/carmustine implant, sorafenib,
temozolomide, trabectedin
Ovarian $144,227 $12,009 0.45 1.17 Bevacizumab, gemcitabine, niraparib, olaparib
Pancreatic $25,274 $4,222 0.25 0.58 Erlotinib, everolimus, gemcitabine, irinotecan,
sunitinib
Prostate $57,870 $7,633 0.35 0.41 Abiraterone acetate, cabazitaxel, docetaxel,
enzalutamide, radium ra 223 dichloride,
sipuleucel-t
Thyroid $171,448 $11,214 0.86 1.78 Lenvatinib, sorafenib, vandetanib
All $82,260 $9,902 0.56 0.91

Notes: Values are averaged across listed drugs. The “Other” category includes cancer types for which we have only
a few drug approvals: brain, liver, Merkel cell carcinoma, mesothelioma, neuroblastoma, soft tissue carcinoma, and
urothelial carcinoma.

50
Table 2. Trends in Median LYG, Mean LYG, and Price

Price per
Median
LYG
($1,000s)
Price per
Mean LYG
($1,000s)
Median
LYG
Mean
LYG
Price
($1,000s)
(1) (2) (3) (4) (5)
A. All Indications
Time:
1995-2012 6.13 0.48 -0.0047 -0.00072 2.82
(4.15) (3.25) (0.011) (0.019) (1.94)
2013-2017 -10.49 0.41 0.12*** 0.15** 15.27**
(13.45) (10.52) (0.037) (0.060) (6.29)
Mean D.V. 195.35 131.89 0.55 0.91 82.26
No. Obs 181 181 181 181 181

Time: B. First Indications Only

1995-2012 8.94 -2.86 -0.021 0.0067 5.60*
(8.27) (6.28) (0.020) (0.034) (3.15)
2013-2017 -32.87 -9.17 0.24*** 0.15 12.85
(33.02) (25.09) (0.078) (0.14) (12.60)
Mean D.V. 236.67 154.98 0.59 1.04 95.56
No. Obs 181 181 181 181 181

Notes: Each column displays results from a separate linear spline regression, adjusted for indication and route of
administration, with a knot at year 2013. Standard errors in parentheses ***p<0.01, ** p<0.05, * p<0.1. Columns
(1), (2), and (5) show changes in the price per year (in thousands) and columns (3) and (4) show changes in life
years gained (LYG). The mean of the dependent variables (D.V.) are shown.

51
Table 3. Robustness Checks with Varying Measures of Price

A. Dropping
Outliers, Price per
B. Only Trials with OS
or PFS
C. Ignoring 2012
Sequester
Time: Median
LYG
(1)
Mean
LYG
(2)
Median
LYG
(3)
Mean
LYG
(4)
Median
LYG
(5)
Median
LYG
(6)
1995-2012 9.780*** 3.043 9.863*** 1.906 6.187 0.518
(2.961) (2.564) (3.110) (3.013) (4.161) (3.254)
2013-2017 -5.438 3.876 -11.512 -0.328 -10.16 0.611
(9.523) (8.246) (10.293) (9.972) (13.48) (10.55)
No. of Obs 179 179 138 138 181 181

D. Log Price per E. Monthly Price per F. AWP per
Median
LYG
(3)
Mean
LYG
(4)
Median
LYG
(7)
Mean
LYG
(8)
Median
LYG
(9)
Mean
LYG
(10)
1995-2012 0.104*** 0.0884*** 0.285 -1.428* 10.35** 3.611
(0.0239) (0.0260) (0.824) (0.800) (4.520) (3.451)
2013-2017 -0.160** -0.116 -4.600* -0.600 -13.64 0.0977
(0.0775) (0.0842) (2.672) (2.593) (14.65) (11.18)
No. of Obs 181 181 181 181 181 181

Notes: Each column and panel displays results from a separate linear spline regression, adjusted for indication and
route of administration, with a knot at year 2013. Standard errors in parentheses ***p<0.01, ** p<0.05, *p<0.1. In
panel A, we dropped two outliers in the data with price per LYG grained than $800,000 per life year gained. The
outliers corresponded to nilotinib for its treatment of leukemia and temozolomide for its treatment of brain cancer.
In panel B, we considered only trials with overall or progression-free survival. In panel C, we examined prices as if
the 2012 sequestration did not occur so that Medicare prices for IV drugs from April 2013 onward were still
reimbursed at 106% average sales price. In panel D, we considered the logged price per LYG. In panel E, we
reported the monthly price (as opposed to the treatment episode price) per mean or median LYG. In panel F, we
measured prices using the average wholesale price in the date closest to approval.

52
Figure 1. Data Collection Flow Chart

207 new approvals
(cancer drug-indication pairs) between
1995 and 2017
Source: CenterWatch, FDA labels,
National Cancer Institute
26 observations excluded
due to missing mean
survival data
181 observations
with price per life year gained
Source: Medicare Parts B and D,
Redbookf
Median survival gains using overall
survival (n=84); if not available, then
progression free survival (n=67); if
neither, then mean*ln(2)
Source: Clinical Trials
Mean survival gain, estimated against
same comparators as used in the median
survival clinical trial
Source: Tufts CEA

53
Figure 2. Fractional Polynomial Regression Estimate

Notes: Predictions are estimated from a fractional polynomial model that adjusts for indication and route
of administration. We show trends in the predicted price per median LYG in thousands (navy) and the
predicted price per mean LYG in thousands (maroon).

54
Figure 3. Trends in Drug Prices per Median vs. Mean Life Year Gained

Notes: Linear spline regressions are adjusted for indication and route of administration, with a knot at
2013. Slope estimates are shown in Exhibit 4.

55
Supplemental Materials
Medicare Price Data
Much of our treatment episode price calculations followed Bach (2009).
52,53
Medicare prices per
unit were calculated following Exhibit A1. We identified dosing recommendations from the drug
label and assumed that a typical treatment lasted 12 weeks for an average adult having body
surface area of 1.7 m
2
and weighing 70 kg. Then, we divided by 2.77 (the average number of
months in 12 weeks) to arrive at a monthly price. Note that the use of a monthly price takes into
account differences in treatment durations across drug-indications.
The Thomson Healthcare’s Redbook provided estimates of the average wholesale price (AWP)
for each drug-indication prior to 2005 and additionally during 2005 for oral drugs not covered
under Part B. For each drug-indication, we used the price at the time of approval (if available) or
the earliest price observed following approval. In 52% of estimates, there was a lag, which lasted
an average of 2.3 years.
Our Part D estimates from 2006 to 2017 rely on a combination of datasets: the 20% sample of
Medicare Part D claims data, Medicare Part D Spending Dashboard, and PlanFinder. Although
the claims data—which provides information on drug name, total price, and days’ supply of
medication for each filled prescription—are the most accurate representation of monthly
Medicare prices, our access to claims data do not extend beyond 2013.
In the absence of extended claims data, we used the Medicare Part D Spending Dashboard for
data from 2014-2016. From 2012 to 2013, we had overlapping data from the claims data and the
Medicare Spending Dashboard. The difference in monthly cost estimates from the two sources
was on average 3.1%.

56
Because the Dashboard does not extend beyond 2017, we used PlanFinder for data in 2017 (the
source used in Bach 2009).
52,53
In 2016, we had overlapping data from the Spending Dashboard
and PlanFinder. The difference in monthly estimates was 33%, with the PlanFinder prices
significantly higher than the Spending Dashboard. Thus, measurement error from the PlanFinder
will bias the price trend to be overstated, and our estimates will represent upper bounds.
Approval dates
In a small number of cases, approval dates for follow-on indications were missing, and we used
news or manufacturer press releases to identify the precise approval date.
Median life years gained (LYG)
When possible, we used measures overall median survival. If data on that clinical endpoint was
not collected, we used measures of progression free survival. In a small subset of our data (52
drug-indication pairs), we observed both median overall and progression-free-survival. The
absolute difference was on average 0.16 months.
Mean LYG
If multiple economic evaluations were available, we averaged the mean LYG estimates across
studies. Relying on either the maximum or minimum mean LYG estimate did not affect trend
estimates. When studies reported only quality-adjusted life-years (QALYs) gained, we used
study-specific utility values to convert QALYs gained to LYG. One study (on radium ra 223) did
not report associated utility estimates, and we relied on the literature to identify the average
indication-specific utility.
68
In about 10% of observations, mean LYG estimates were missing in
the Tufts CEA database, and we supplemented the data from the UK’s National Institute for
Health and Care Excellence Final Appraisal Determination (18 observations), Scottish Medicines

57
Consortium (2 observations) and All Wales Medicine Strategy Group (1 observations). All
results are robust to dropping these observations.
Across the entire Tufts CEA sample, 49% of samples received some industry funding. Authors
were predominately affiliated with academic institutions (85%), though several authors had
primary or additional affiliation with pharmaceutical or biotech companies (35%). It is possible
that industry and non-industry affiliated studies differ in their extrapolation methods, and it is an
interesting question for future research.

A. Methodological Details
For the fractional polynomial model, we considered 44 different degree-2 polynomials of the
form:
𝑌
"#
= 𝛽
&
+𝛽
(
𝑥
"#
(D
E
)
+𝛽
6
𝑥
"#
(D
F
)
+𝛼1(Oral)
>?
+𝜂
A(")
+𝜖
"#

The dependent variable 𝑌
"#
is the price per mean or median LYG. The main independent variable
𝑥
"#
refers to the year of the drug-indication approval. We additionally control cancer type fixed
effects 𝜂
A(")
and the route of administration 1(Oral)

with an indicator equal to one if the drug is
an oral drug.
The powers {𝑝
(
,𝑝
6
} are drawn from the set 𝑃 = {−2,−1,−0.5,0,0.5,1,2,3}, where 𝑥
&
should be
interpreted as ln(𝑥). Models with repeated powers sequentially multiply the dependent variable
by ln(x), so the model with {𝑝
(
,𝑝
6
} = {0,0} was:
𝑌
"#
= 𝛽
&
+𝛽
(
ln (𝑥)+𝛽
6
{ln(𝑥)}
6
+𝛼1(Oral)
>?
+𝜂
A(")
+𝜖
"#

58
Post estimation, we calculated the partial F-test across all 44 models and selected the model with
the lowest deviance. The resulting models shown in the Figure 2 are from the following
equations:
(Price/Mean LYG)
"#
= 𝛽
&
+𝛽
(
𝑥
>?
Z(
+𝛽
6
𝑥
>?
Z(
∗ln(𝑥)+𝛼1(Oral)
>?
+𝜂
A(")
+𝜖
"#

(Price/Median LYG)
"#
= 𝛽
&
+𝛽
(
𝑥
>?
]
+𝛽
6
𝑥
>?
]
∗ln(𝑥)+𝛼1(Oral)
>?
+𝜂
A(")
+𝜖
"#

59
Figure A1. Number of Drug-Indications per Year

Notes: This plot shows the number of drug-indication approvals per year, where an indication is classified
as a cancer type and line of therapy.

60
Figure A2. Fractional Polynomial, by First and Follow-On Indications

(a) Predicted Price per Median LYG

(b) Predicted Price per Mean LYG

Notes: Estimates come from a fractional polynomial that adjusts for indication and route of
administration. We omit three observations from with negative predicted price per median LYG among
follow-on indications.

61
Table A1: Medicare Price Calculation and Data Source
Year of Approval Medicare Price Data Source
Panel A. Part B physician administered and covered oral drugs
1995-1996 100% of AWP
Redbook historical AWP 1997-2003 95% of AWP
2004 85% of AWP
2005-2012 106% of ASP Medicare Part B ASP
Drug Pricing Files 2013-current 104.3% of ASP*
Panel B. Part D oral drugs
1995-2005 Follow above method** Redbook historical AWP
2006-2013 Total cost Medicare Part D 20% claims data
2014-2016 Average spending per claim
Medicare Part D
Drug Spending Dashboard
2017 Full cost of drug
PlanFinder for Humana PDP Enhanced
Plan for beneficiary in zip 90089

Notes: We follow Medicare reimbursement rules based on average wholesale price (AWP) or average
sales price (ASP). * In a sensitivity test (Column C of Exhibit A2), we keep this as 106% of ASP. ** For
oral drugs in 2005, we calculated Medicare prices using 95% of AWP, which is Medicare’s payment
policy if the drug’s ASP is not available

62
Table A2: Trends in Median LYG, Mean LYG, and Price, With Additional Spline Knots
Price per
Median LYG
($1,000s)
Price per
Mean LYG
($1,000s)
Median
LYG
Mean
LYG
Price
($1,000s)
Time: (1) (2) (3) (4) (5)
A. Knots at 2003 and 2013
1995-2002 -1.649 -17.09 0.0220 0.0263 5.713
(13.75) (10.67) (0.0374) (0.0617) (6.434)
2003-2012 8.823 6.559 -0.0140 -0.0101 1.821
(6.156) (4.777) (0.0168) (0.0276) (2.881)
2013-2017 -13.53 -6.452 0.133*** 0.156** 16.40**
(14.42) (11.19) (0.0392) (0.0647) (6.747)
Obs. 181 181 181 181 181

B. Knots at 2000, 2005, 2010, and 2015
1995-1999 7.498 -43.13 0.0618 0.156 -12.00
(33.82) (26.13) (0.0915) (0.152) (15.69)
2000-2004 0.221 7.817 -0.0133 -0.0632 12.83
(17.56) (13.56) (0.0475) (0.0787) (8.142)
2005-2009 10.81 0.371 -0.0129 0.000254 2.305
(13.79) (10.66) (0.0373) (0.0618) (6.398)
2010-2014 3.050 9.139 0.0177 0.0469 0.108
(12.96) (10.02) (0.0351) (0.0581) (6.012)
2015-2017 -32.64 -28.93 0.251*** 0.214 37.64**
(32.08) (24.78) (0.0868) (0.144) (14.88)
Obs. 181 181 181 181 181

Notes: Each column displays results from a separate linear spline regression, adjusted for indication and
route of administration. Standard errors in parentheses ***p<0.01, ** p<0.05, * p<0.1.

63

References:
1. Burki TK. Rising cancer drug costs in the USA. The Lancet Oncology. 2017;18(11):e652.
2. Promoting Value, Affordability, and Innovation in Cancer Drug Treatment. Bethesda,
MD: President's Cancer Panel; 2018.
3. Gordon N, Stemmer SM, Greenberg D, Goldstein DA. Trajectories of Injectable Cancer
Drug Costs After Launch in the United States. Journal of Clinical Oncology.
2017;36(4):319-325.
4. Dendreon: Provenge to cost $93K for full course of treatment. 2010;
https://www.fiercebiotech.com/biotech/dendreon-provenge-to-cost-93k-for-full-course-
of-treatment. Accessed July 23, 2018.
5. Helfand C. Yervoy- Bristol-Myers Squibb. FiercePharma 2011;
https://www.fiercepharma.com/special-report/yervoy-bristol-myers-squibb. Accessed
July 23, 2018.
6. Staton T. Amgen slaps record-breaking $178K price on rare leukemia drug Blincyto.
2014; https://www.fiercepharma.com/marketing/amgen-slaps-record-breaking-178k-
price-on-rare-leukemia-drug-blincyto. Accessed July 23, 2018.
7. The Council of Economic Advisors. Reforming Biopharmaceutical Pricing at Home and
Abroad. 2018; https://www.whitehouse.gov/wp-content/uploads/2017/11/CEA-Rx-
White-Paper-Final2.pdf. Accessed March 1, 2018.
8. Mailankody S, Prasad V. Five years of cancer drug approvals: Innovation, efficacy, and
costs. JAMA Oncology. 2015;1(4):539-540.
9. Rupp T, Zuckerman D. Quality of life, overall survival, and costs of cancer drugs
approved based on surrogate endpoints. JAMA Internal Medicine. 2017;177(2):276-277.

64
10. Howard DH, Bach PB, Berndt ER, Conti RM. Pricing in the Market for Anticancer
Drugs. Journal of Economic Perspectives. 2015;29(1):139-162.
11. US Department of Health and Human Services and Food and Drug Administration.
Clinical Trial Endpoints for the Approval of Cancer Drugs and Biologics. 2007;
https://www.fda.gov/downloads/Drugs/Guidances/ucm071590.pdf. Accessed May 5,
2018.
12. Latimer NR. NICE DSU Technical Support Document 14: Undertaking Survival
Analysis for Economic Evaluations Alongside Clinical Trials - Extrapolation with
Patient-Level Data. 2011; http://nicedsu.org.uk/wp-content/uploads/2016/03/NICE-DSU-
TSD-Survival-analysis.updated-March-2013.v2.pdf. Accessed Nov 28, 2018.
13. Doctor JN, Huesch MD, Meeker D. Rethinking the value of survival: clinical trials
should measure patient preferences for survival on entry to trials. Journal of Clinical
Epidemiology. 2016;77:137-138.
14. Lakdawalla DN, Romley JA, Sanchez Y, Maclean JR, Penrod JR, Philipson T. How
Cancer Patients Value Hope And The Implications For Cost-Effectiveness Assessments
Of High-Cost Cancer Therapies. Health Affairs. 2012;31(4):676-682.
15. Schnipper LE, Davidson NE, Wollins DS, et al. Updating the American Society of
Clinical Oncology Value Framework: Revisions and Reflections in Response to
Comments Received. Journal of Clinical Oncology. 2016;34(24):2925-2934.
16. Neumann PJ, Cohen JT. ICER’s Revised Value Assessment Framework for 2017–2019:
A Critique. PharmacoEconomics. 2017;35(10):977-980.

65
17. Institute for Clinical and Economic Review. Overview of the ICER Value Framework
and Proposals for an Update for 2017-2019. https://icer-review.org/wp-
content/uploads/2018/03/ICER-value-assessment-framework-update-FINAL-062217.pdf.
Accessed Dec 1, 2018.
18. Institute for Clinical and Economic Review. Poly ADP-Ribose Polymerase (PARP)
Inhibitors for Ovarian Cancer Effectiveness and Value. 2017; https://icer-review.org/wp-
content/uploads/2017/02/MWCEPAC_OVARIAN_EVIDENCE_REPORT_08302017.pd
f. Accessed Dec 1, 2018.
19. Latimer NR. Survival Analysis for Economic Evaluations Alongside Clinical Trials—
Extrapolation with Patient-Level Data: Inconsistencies, Limitations, and a Practical
Guide. Medical Decision Making. 2013;33(6):743-754.
20. Guyot P, Ades AE, Beasley M, Lueza B, Pignon J-P, Welton NJ. Extrapolation of
Survival Curves from Cancer Trials Using External Information. Medical Decision
Making. 2016;37(4):353-366.
21. Williams C, Lewsey JD, Mackay DF, Briggs AH. Estimation of Survival Probabilities for
Use in Cost-effectiveness Analyses: A Comparison of a Multi-state Modeling Survival
Analysis Approach with Partitioned Survival and Markov Decision-Analytic Modeling.
Medical Decision Making. 2016;37(4):427-439.
22. Howard DH, Chernew ME, Abdelgawad T, Smith GL, Sollano J, Grabowski DC. New
Anticancer Drugs Associated With Large Increases In Costs And Life Expectancy.
Health Affairs (Millwood). 2016;35(9):1581-1587.

66
23. Woodward RM, Brown ML, Stewart ST, Cronin KA, Cutler DM. The value of medical
interventions for lung cancer in the elderly: results from SEER-CMHSF. Cancer.
2007;110(11):2511-2518.
24. Lin PJ, Winn AN, Parsons SK, Neumann PJ, Weiss ES, Cohen JT. Linking Costs and
Survival in the Treatment of Older Adults With Chronic Myeloid Leukemia: An Analysis
of SEER-Medicare Data From 1995 to 2007. Medical Care. 2016;54(4):380-385.
25. DiMasi JA, Faden LB. Competitiveness in follow-on drug R&D: a race or imitation?
Nature Reviews Drug Discovery. 2011;10(1):23-27.
26. Cohen J, Cabanilla L, Sosnov J. Role of follow-on drugs and indications on the WHO
Essential Drug List. Journal of Clinical Pharmacy and Therapeutics. 2006;31(6):585-
592.
27. Seidman J, Anderson M, Masi D, Atkins M, Japha M. Measuring Value Based on What
Matters to Patients: A New Value Assessment Framework. Health Affairs Blog 2017;
https://www.healthaffairs.org/do/10.1377/hblog20170523.060220/full/. Accessed Dec 1,
2018.
28. Salas-Vega S, Mossialos E. Cancer Drugs Provide Positive Value In Nine Countries, But
The United States Lags In Health Gains Per Dollar Spent. Health Affairs (Millwood).
2016;35(5):813-823.
29. Prasad V, De Jesús K, Mailankody S. The high price of anticancer drugs: origins,
implications, barriers, solutions. Nature Reviews Clinical Oncology. 2017;14:381.
30. Institute of Medicine. Delivering High-Quality Cancer Care: Charting a New Course for
a System in Crisis. Washington, DC: The National Academies Press;2013.

67
31. The President's Cancer Panel. Promoting Value, Affordability, and Innovation in Cancer
Drug Treatment. Bethesda, MD 2018.
32. Food and Drug Administiration Safety and Innovation Act. In. Vol 126 Stat. 112th
Congress: Public Law 112-114; 2012:993-1130.
33. Darrow JJ, Avorn J, Kesselheim AS. New FDA Breakthrough-Drug Category —
Implications for Patients. New England Journal of Medicine. 2014;370(13):1252-1258.
34. Shea M, Ostermann L, Hohman R, et al. Impact of breakthrough therapy designation on
cancer drug development. Nature Reviews Drug Discovery. 2016;15:152.
35. Aggarwal S. A survey of breakthrough therapy designations. Nature Biotechnology.
2014;32:323.
36. Kesselheim AS, Woloshin S, Eddings W, Franklin JM, Ross KM, Schwartz LM.
Physicians’ knowledge about fda approval standards and perceptions of the
“breakthrough therapy” designation. JAMA. 2016;315(14):1516-1518.
37. Darrow JJ, Avorn J, Kesselheim AS. The FDA Breakthrough-Drug Designation — Four
Years of Experience. New England Journal of Medicine. 2018;378(15):1444-1453.
38. Innovation and Value Initiative. Open-Source Value Project- Sequential Treatment
Strategies for Patients with Metastatic EGFR + Non-Small Cell Lung Cancer. 2018;
https://www.thevalueinitiative.org/wp-content/uploads/2018/08/IVI120-OSVP-EGFR-
pos-NSCLC-modeling-protocol-v2.pdf. Accessed Dec 1, 2018.
39. CenterWatch. FDA Approved Drugs for Oncology; https://www.centerwatch.com/drug-
information/fda-approved-drugs/therapeutic-area/12/oncology. Accessed August, 2017.

68
40. Pérez-Herrero E, Fernández-Medarde A. Advanced targeted therapies in cancer: Drug
nanocarriers, the future of chemotherapy. European Journal of Pharmaceutics and
Biopharmaceutics. 2015;93:52-79.
41. Amir E, Seruga B, Kwong R, Tannock IF, Ocana A. Poor correlation between
progression-free and overall survival in modern clinical trials: are composite endpoints
the answer? European Journal of Pharmaceutics and Biopharmaceutics. 2012;48(3):385-
388.
42. Beauchemin C, Johnston JB, Lapierre M, Aissa F, Lachaine J. Relationship between
progression-free survival and overall survival in chronic lymphocytic leukemia: a
literature-based analysis. Current Oncology. 2015;22(3):e148-e156.
43. Broglio KR, Berry DA. Detecting an Overall Survival Benefit that Is Derived From
Progression-Free Survival. Journal of the National Cancer Institute. 2009;101(23):1642-
1649.
44. Han G, Schell MJ, Zhang H, et al. Testing violations of the exponential assumption in
cancer clinical trials with survival endpoints. Biometrics. 2017;73(2):687-695.
45. Henze N, Meintanis SG. Recent and classical tests for exponentiality: a partial review
with comparisons. Metrika. 2005;61(1):29-45.
46. Beck JR, Kassirer JP, Pauker SG. A convenient approximation of life expectancy (the
“DEALE”): I. validation of the method. The American journal of medicine.
1982;73(6):883-888.
47. National Comprehensive Cancer Network. https://www.nccn.org/. Accessed April, 2018.

69
48. Tufts Cost Effectiveness Analysis Registry.
http://healtheconomics.tuftsmedicalcenter.org/cear4/SearchingtheCEARegistry/Searchthe
CEARegistry.aspx. Accessed May, 2017.
49. Tufts Medical Center. Publications Using Registry Data.
http://healtheconomics.tuftsmedicalcenter.org/cear4/Resources/Publications.aspx.
Accessed July 23, 2018.
50. Busch SH, Berndt ER, Frank RG. Creating Price Indexes for Measuring Producitivy in
Mental Health Care. In: Garber AM, ed. Frontiers in Health Policy Research. Vol 4.
Cambridge, MA: MIT Press for the National Bureau of Economic Research; 2001.
51. Berndt ER, Cutler DM, Frank RG, Griliches Z, Newhouse JP, Triplett JE. Medical Care
Prices and Output. In: Cuyler AC, Newhouse JP, eds. Handbook of Health Economics.
Vol 1A. Amsterdam: Elsevier Science; 2000.
52. Bach PB. Limits on Medicare's Ability to Control Rising Spending on Cancer Drugs.
New England Journal of Medicine. 2009;360(6):626-633.
53. Memorial Sloan Kettering Cancer Center. Methods for Drug Price Calculations. 2017;
https://www.mskcc.org/sites/default/files/node/25097/documents/methods-for-drug-
price-calculations-081517.pdf. Accessed July 27, 2018.
54. Royston P, Altman DG. Regression Using Fractional Polynomials of Continuous
Covariates: Parsimonious Parametric Modelling. Journal of the Royal Statistical Society
Series C (Applied Statistics). 1994;43(3):429-467.
55. Conti RM, Fein AJ, Bhatta SS. National Trends In Spending On And Use Of Oral
Oncologics, First Quarter 2006 Through Third Quarter 2011. Health Affairs (Millwood).
2014;33(10):1721-1727.

70
56. Polite B, Conti RM, Ward JC. Reform of the Buy-and-Bill System for Outpatient
Chemotherapy Care Is Inevitable: Perspectives from an Economist, a Realpolitik, and an
Oncologist. American Society of Clinical Oncology Educational Book. 2015:e75-80.
57. IBM Watson Health. Micromedex Redbook.
http://truvenhealth.com/Products/Micromedex/Product-Suites/Clinical-Knowledge/RED-
BOOK. Accessed June 18, 2018.
58. Aitken M, Kleinrock M, Kumar S. Global Oncology Trends 2017: Advances,
Complexity, and Cost. 2017; https://www.iqvia.com/-/media/iqvia/pdfs/institute-
reports/global-oncology-trends-2017.pdf. Accessed Dec 1, 2018.
59. Faiman B. Oral cancer therapy: policy implications for the uninsured and underinsured
populations. Journal of the Advanced Practitioner in Oncology. 2013;4(5):354-360.
60. Drug Channels Institute. The 340B program reached $19.3 billion in 2017--as hospitals'
charity care has dropped. 2018; https://www.drugchannels.net/2018/05/exclusive-340b-
program-reached-193.html. . Accessed Dec 1, 2018.
61. Conti RM, Bach PB. Cost Consequences of the 340B Drug Discount Program. JAMA.
2013;309(19):1995-1996.
62. Hwang TJ, Franklin JM, Chen CT, et al. Efficacy, Safety, and Regulatory Approval of
Food and Drug Administration–Designated Breakthrough and Nonbreakthrough Cancer
Medicines. Journal of Clinical Oncology. 2018;36(18):1805-1812.
63. Kumar H, Fojo T, Mailankody S. An Appraisal of Clinically Meaningful Outcomes
Guidelines for Oncology Clinical Trials. JAMA Oncology. 2016;2(9):1238-1240.

71
64. Beaver JA, Howie LJ, Pelosof L, et al. A 25-year experience of us food and drug
administration accelerated approval of malignant hematology and oncology drugs and
biologics: A review. JAMA Oncology. 2018;4(6):849-856.
65. Seabury SA, Goldman DP, Maclean JR, Penrod JR, Lakdawalla DN. Patients Value
Metastatic Cancer Therapy More Highly Than Is Typically Shown Through Traditional
Estimates. Health Affairs. 2012;31(4):691-699.
66. Devlin NJ, Lorgelly PK. QALYs as a measure of value in cancer. Journal of Cancer
Policy. 2017;11:19-25.
67. Whitehead SJ, Ali S. Health outcomes in economic evaluation: the QALY and utilities.
British Medical Bulletin. 2010;96(1):5-21.
68. Lloyd AJ, Kerr C, Penton J, Knerer G. Health-Related Quality of Life and Health
Utilities in Metastatic Castrate-Resistant Prostate Cancer: A Survey Capturing
Experiences from a Diverse Sample of UK Patients. Value in Health. 2015;18(8):1152-
1157.

72

Chapter III. Variation in oncology spending at the practice-level and its
association with patient survival in a privately insured population with
metastatic cancer

73
Abstract
Oncology spending is projected to continue to rise in the next decade. Variations observed in
spending at different regional levels and the lack of an association between higher spending and
better patient outcomes have been cited as evidence for waste in health care system. This
analysis aimed to examine the variation in oncology spending at the practice level and assess its
impact on patient survival in a privately insured population with metastatic cancer.
A commercial claims database containing Medicare Advantage population was used for this
analysis. Patients newly diagnosed with metastatic cancer during 2008-2016 were identified and
total oncology expenditures incurred in 6 months following the metastatic cancer diagnosis were
included. At the patient level, all costs were attributed to the patient’s primary practice. Risk-
adjusted spending was estimated to account for impact from the differences in patient
characteristics and risk-adjusted practice-level spending was the average across all patients that
were assigned to it. Using this measure, the practices were ranked and assigned to quintiles. An
ordinary least square model was fit to identify potential drivers for increased practice oncology
expenditures. The association between practice-level oncology spending ang patient survival was
assessed using generalized linear models and Cox models.
Substantial variation in spending still existed after risk adjustment using patient demographic and
clinical variables. Practice level factors identified that might explain the residual variation
included number of oncologist visit per patient and the proportion of patients using targeted
therapy. Results from the generalized linear models and the Cox model showed that patients who
were treated at practices in the quintile of highest risk-adjusted practice-level spending
experienced significantly lower two-year and five-year all-cause mortality rates as well as
significantly longer survival compared to those treated at practices in the lowest risk-adjusted

74
practice-level spending quintile. This association was not observed in other risk-adjusted
spending quintiles, however.
Wide variation in oncology spending at the clinical practice level was demonstrated after
accounting for patient characteristics in the private market, highlighting the potential for care
efficiency improvement at the practice level. Improved patient survival was observed in some
high-spending practices, which suggested that those patients might have benefited from the
specific practicing style or treatment choices.

75
Introduction
Scientific advances in the past few years have revolutionized the way cancer is diagnosed and
treated, which leads to increased overall demand for oncology care due to prolonged patient
survival, thus placing a serious financial burden on the US health care system. The overall costs
of cancer were estimated at approximately $125 billion in 2010 and was projected to reach $158
billion by 2020 in the US (2010 USD).
1
For over decades, the geographic variations in health
care spending have been documented and better patient outcomes were not always achieved in
those areas where more services were delivered.
2-9
A 2013 Institute of Medicine (IOM) report
demonstrated the persistent variation in health care spending observed across all geography
levels as well as the inconclusive association between higher spending and better outcomes, and
stressed the need to improve care quality to avoid unsustainable health spending.
10

In oncology specifically, a recent review also reported mixed findings on the relationship
between spending and cancer outcomes.
11
They found that studies conducted within the US often
reported negative association or no association, while a positive association was almost always
observed in international comparisons of country-level spending and outcomes. In this review,
most of the US studies identified in the review were based on data from the Medicare fee-for-
service population at a regional level (i.e. state-level, health referral region or health service
area). Geographic units are typically not where most health care decisions are made and thus
IOM recommends the reforms and policy efforts targeting the hospital or group practice level to
maximize efficiency.
10
In 2015, a Medicare analysis demonstrated substantial practice-level
variation in oncology spending after accounting for differences in patient demographic and
clinical variables.
12
Their results indicated that the unexplained variation in spending might be

76
attributable to practice factors and physician behaviors and the use of high-value oncology
services could potentially be promoted through alternative payment models.
In the US, people with private health insurance coverage account for approximately 67% of the
total population with health care coverage.
13
Previous research reported unexplained and
persistent variation in health care spending in the private sector after controlling for the influence
of differential price at the regional level but not at the practice level.
10,14
To my knowledge, no
study has examined the variation in oncology spending at the practice-level and its association
with patient outcomes in a privately insured population. Thus, findings from this analysis could
add to the body of evidence about the current quality of oncology care across the nation and
highlight the opportunity for future efforts in improving oncology care quality.

Methods
Data
A commercial claims database containing Medicare Advantage population with over 47.8
million unique lives across a wide geography of the US was used for this analysis. This database
contains information on member demographics, medical and pharmacy claims, and inpatient
stays as well as provider and facility characteristics. Data from 2007–2017 were extracted for
this analysis. Specifically, data from 2008-2016 were used for analysis to allow for at least one
year of data for diagnosis screening and one year for follow-up.
Study sample
The study cohort consisted of patients with metastatic cancer. Cancer types included in the
analysis were prostate, kidney, colorectal, liver, lung pancreatic, gastric and ovarian cancer.

77
These cancer types were chosen as these were the most prevalent solid cancers in the US.
Besides the primary cancer diagnosis, patients with solid cancer will receive a diagnosis of
secondary cancer once the tumor cells metastasize. Thus, the secondary cancer diagnosis served
as a proxy for metastatic cancer diagnosis in patient identification in absence of cancer staging
information in this claims database. Specifically, metastatic cancer patients were defined as those
who were newly diagnosed with one of the cancers in 2008-2016 and also had a diagnosis of
secondary cancer in year of the primary cancer diagnosis.
15
For both cancer diagnoses, one
diagnosis in the inpatient setting or two diagnoses that were at least 30 days apart in the
outpatient setting were required. In addition, patients were required to be continuously enrolled
in the plan for the 12 months prior to the secondary cancer diagnosis as well as for at least 6
months following the metastatic cancer diagnosis.
Study outcomes
To summarize the relevant spending incurred by this group of patients, total cancer-related
spending over a 6-month period following the secondary cancer diagnosis was included. A 6-
month interval was selected as most decisions on the critical components of cancer care, such as
initial cancer treatment, were typically made during this period. If a patient died within 6 months
following secondary cancer diagnosis, then the spending incurred from date of metastatic cancer
diagnosis and date of death was included. It is worthwhile to note that the cost variable in this
database was standardized to remove the differences in pricing across geographies, health plans
and provider contracts. Therefore, the variation demonstrated in this analysis likely reflect
variations in service utilization, not prices. All costs were adjusted to 2017 US dollars in this
analysis.

78
Metastatic cancer patients could visit facilities other than the practice of their primary oncologist
to receive different types of services of their cancer management, such as advanced imaging or
inpatient stay. Therefore, all costs of a patient were attributed to his/her primary practice, which
was defined as the practice that had the greatest number of Evaluation and Management (E&M)
visits.
12
Specifically, costs were first summed up at the patient level to get the total patient
spending. Patients were attributed to each provider using the unique provider ID on each claim,
and then the providers were attributed to practices using unique group practice ID or hospital
affiliation ID associated with each provider. At each practice, the total spending was divided by
the number of patients assigned to that practice to get the mean practice spending. To avoid
unstable estimate of practice-level spending, the practices with less than 20 patients during the
entire study period were dropped.
To investigate the impact of different practice-level spending on cancer patients, three patient-
level outcomes were defined – the two-year all-cause mortality, five-year all-cause mortality and
survival duration. Survival duration was defined as the time between date of metastatic cancer
diagnosis and date of death. The source of the date of death in this database was the Death
Master File maintained by the Social Security Office. The date was set at the 15
th
of the month as
only the year and month of death was provided. Patients who did not die during the entire study
period or missed death information were censored at end of study period (Dec 31, 2017).
Statistical Analysis
To examine the variation in oncology spending at the practice level, a risk-adjusted spending
measure was first calculated to remove the variation caused by different patient characteristics.
The risk adjustment included three steps. First, a generalized linear model with a log link,
gamma distribution and robust standard error clustered at the practice level was fit, from which

79
predicted patient total spending was estimated. Patient-level covariates included age, sex, year of
secondary cancer diagnosis, cancer type, year of first cancer diagnosis and Charlson Comorbidity
Index (CCI), which was calculated based on the diagnoses that patient received in the year prior
to the metastatic cancer diagnosis. Then the predicted practice-level spending is the sum of
predicted patient total spending divided by number of patients at each practice. Second, an
observed-to-predicted ratio was calculated for each practice as the ratio of observed practice-
level spending and predicted practice-level spending. Finally, the risk-adjusted mean practice
spending equaled the observed-to-predicted ratio multiplied by the national median practice
mean payment.
12
Practices were then ranked according to the risk-adjusted mean practice
spending from low to high and assigned to quintiles.
A number of practice-level factors that might potentially explain the variation observed in
oncology spending at the practice-level were summarized and explored, including the practice
size, defined by number of oncologists per practice; patient volume, defined by number of
patients per oncologist; number of oncologist visits per patient and primary cancer treatment
decision, defined by the proportion of patients receiving chemotherapy, targeted therapy,
immunotherapy, radiation or surgery at a practice. An ordinary least square model was fit to
evaluate and quantify the impact of these factors on risk-adjusted practice-level spending.
To assess the association between practice spending level and patient outcomes, three models
were fit. First, two generalized linear models with a Poisson distribution, log link and robust
standard error were fit with the dependent variable being if died in two years or if died in five
years, respectively and the key explanatory variable being quintile of risk-adjusted practice-level
spending. To provide a more comprehensive picture on the recent advances in cancer treatment
innovation, duration of survival was also included as it would mainly be affected by the primary

80
cancer treatment that patients with metastatic cancer received. A vector of patient-level
covariates including age category, sex, year of metastatic cancer diagnosis, CCI, cancer type,
year of first cancer diagnosis was also adjusted for in all three models.
The statistical analyses were performed using SAS 9.4 and Stata 15.0.

Results
Sample characteristics
The practice and patient characteristics by quintile of risk-adjusted practice-level spending are
summarized in Tables 1 and 2, respectively. There were about 164 practices in each quintile and
the middle spending quintiles appeared to be consisted of larger practices with greater number of
patients and oncologists. The risk-adjusted practice-level spending was approximately $117,000
in Q1 (the lowest risk-adjusted spending quintile) and $198,000 in Q5 (the highest risk-adjusted
spending quintile). Practices in the higher risk-adjusted practice-level spending quintiles
appeared to have more use of targeted therapy and immunotherapy despite similar cancer
distribution. The ratio between the risk-adjusted practice-level spending in Q1 and Q5 was
approximately 1.7, indicating substantial practice-level variation in oncology spending after
accounting for the differences in patient characteristics.
Practice-level drivers of spending
To further study the potential sources of residual variation in spending at the practice level, an
OLS model including a number of practice-level factors was fit. Table 3 shows results from the
OLS model assessing the impact of practice-level factors on practice-level spending. The number
of oncologist visit per patient and use of targeted therapy appeared to be the two factors with a

81
significant impact on risk-adjusted practice-level spending. One more oncologist visit per patient
is associated $1,170 reduction in practice-level spending and 1 percentage point increase in the
use of targeted therapy at a practice is associated with $539 increase in practice-level spending.
The effect size of use of immunotherapy suggests a larger impact on practice-level spending,
however it is insignificant due to the very small proportion of patients ever on immunotherapy
during the study period in our data sample.

Association between practice-level spending and patient outcomes
Results from the two generalized linear models on two-year and five-year all-cause mortality
rates were shown in Figures 1 and 2, respectively. Patients treated at practices in Q5, the highest
risk-adjusted practice-level spending quintile, experienced significantly lower two-year all-cause
mortality rates compared to those treated at the practices in Q1, the lowest risk-adjusted practice-
level spending quintile, with a risk ratio of 0.85. Similar results were observed for five-year all-
cause mortality, with patients treated at practices in Q5 had a significantly lower risk ratio of
0.86 compared to those treated at practices in Q1. Patients treated at practices in other spending
quintiles had similar mortality outcomes compared to those treated at practice Q1, however.
To provide more insights into the magnitude of the effect of the improvements in cancer
management and treatment innovation, patient overall survival was also evaluated and was
compared across quintiles. Table 2 shows results from the crude and multivariate Cox models.
After accounting for patient-level characteristics, patients treated at practices in Q5 had
significantly longer survival compared to those treated at practices in Q1 with a hazard ratio of
death of 0.82 (95% CI, 0.78-0.87), consistent with the findings from the generalized linear
models on mortality risks.

82
Since treating at practices in the highest risk-adjusted spending quintile was found to be
associated with significantly better survival outcomes, a natural question arises: if the additional
survival benefit was worth the excess costs of cancer care. To project the life expectancy of
patients, an exponential distribution was assumed for the survival function, which is a widely
used assumption in modeling patient survival in late-stage cancers.
16
Given the estimated hazard
ratio of death from the Cox model (0.82, 95% CI: 0.78-0.87), the life expectancy difference was
calculated to be approximately 0.33 years between patients treated at practices in Q5 and Q1.
With a cost difference of $81,000 (the risk-adjusted practice-level spending was $116,773 in Q1
and $197,658 in Q5, respectively), treating the same patients at practices in Q5 in contrast to
treating them at practices Q1 was associated with a cost effectiveness ratio of approximately
$245,000 per life-year saved.

Discussion
The association between health care spending and patient outcomes have been previously studied
at different regional levels and the mixed findings have been frequently cited as evidence of
waste in care delivery in the high-spending areas.
9,17-21
Using data from a privately insured
population with metastatic cancer, this study demonstrated substantial variation in practice-level
oncology spending across a broad range of cancers. Use of targeted therapy at the practice level
was revealed to be a significant driver of increased oncology spending. In addition, improved
survival outcomes were observed in patients treated at practices in the highest risk-adjusted
spending quintile than those treated at practices in the lowest risk-adjusted spending quintile.
Taken together, findings from this analysis indicate that some patients may have benefited from

83
being treated at the high-spending practices by receiving better treatments or other practice
styles.
The magnitude of the variation in spending reduced after risk-adjustment using patient
demographic and clinical characteristics. This is consistent with previous findings that
differences in patient-level characteristics are one of the sources for variation in health care
spending.
22-24
The large amount of residual variation suggested impact from factors at the
practice level, such as practice characteristics or physician behaviors. While evidence is limited
as to the full spectrum of potential contributing practice-level factors, the use of targeted therapy
was identified as one significant driver of higher oncology expenditures. Given the ongoing
discussion on the overuse or misuse of new, innovative oncology drugs, the relationship between
spending and cancer patient survival is an important topic for investigation. As improved
survival outcomes were observed in patients treated at practices in highest-spending quintile, the
survival differences might be partly explained by the differences in the cancer treatment choice.
One major strength of this analysis is that the variation in oncology spending and its association
with patient survival was studied at the practice-level. A large body of research has demonstrated
variation in health care spending at different geographical levels, such as state, hospital referral
region or hospital services area level, and assessed the impact of regional differences in spending
on patient outcomes.
3,4, 9,17-21
As the geographical region is not a level of actual decision-making,
findings based on these studies may not provide direct supporting evidence for policies or
reforms in an area where no single delivery system dominates. By focusing on the practice level
oncology spending, findings from this analysis highlight the room for improvement in cancer
care efficiency at the practice level, which is a suitable unit for interventions such as payment
reforms.
10
Secondly, by using data from a privately insured population, the results filled the gap

84
in the knowledge of variations in oncology expenditure in the private market. Previous studies
have mostly focused on the variation in health care spending in the public market by employing
Medicare or Medicaid data. Although Medicare is the single largest payer in the US, spending in
the private sector accounts for approximately 34% of total spending on medical care.
27

Moreover, evidence suggests that high-spending areas in the public market and high-spending
areas in the private market are only weakly correlated.
10
In other words, factors contributing to
the variation in health care spending in the public market could be different from those
contributing to the variation in spending in private market. For example, one known factor is that
variation in prices of medical services are substantially larger in the private market than that in
the public market where the base prices were set by the CMS.
There are several limitations worth noting. Firstly, the data used for this analysis was from a
claims database, of which the purpose was to provide billing information and thus granular
patient clinical characteristics were lacking. As a result, residual confounding might exist as
certain relevant clinical information might not have been fully captured. In this analysis, patient
health status was characterized by using CCI, an index based on diagnoses observed in one year
before the metastatic cancer diagnosis. In addition, as this is a cohort of patients with metastatic
cancer, of whom the underlying health status are generally considered very poor, the health
heterogeneity across patients could be considered minimal. Secondly, due to data availability,
impact from a limited number of practice-level factors were explored in this analysis. Other
practice factors, such as if being an academic oncology center or adoption of an alternative
payment model, might also affect the practice-level spending. Future research with detailed data
on practice characteristics are needed to better identify specific practice-level drivers for
oncology spending. Finally, although mortality rate and survival duration are objective outcomes

85
directly affected by the quality of cancer care, other patient outcomes, such as quality of life, are
also important components to consider when assessing the quality of care.
In summary, this analysis demonstrated that there was wide variation in oncology spending at the
clinical practice level after accounting for patient characteristics in the private market,
highlighting the potential for care efficiency improvement at the practice level. Improved patient
survival was observed in some high-spending practices, which suggested that those patients
might have benefited from the specific practicing style or treatment choices. To shed further light
on practice-level policy efforts or reforms, future research is still needed to identify the specific
drivers for the additional costs at the practice level.

86
Table 1. Practice characteristics by quintile of risk-adjusted practice-level spending

Q1 Q2 Q3 Q4 Q5 Q5/Q1
ratio
Risk-adjusted mean
practice-level spending
$ 116,773 $ 133,865 $ 143,964 $ 156,804 $ 197,658 1.69
No. of practices 164 164 164 164 162

Total no. of patients 6741 9926 9531 10403 7912

Mean no. of patients 41 61 58 63 49

Median no. of patients 30 40 36 38 34

Total no. of oncologists 1111 1872 2449 2468 1619

Mean no. of oncologists 12 21 23 28 18

Median no. of oncologists 11 17 19 24 14

Mean no. of patients per
oncologist
8.8 10.1 10.7 10.3 10.6

Cancer treatment choice

Traditional chemotherapy 25.3% 30.3% 37.9% 39.2% 35.5% 1.40
Targeted therapy 11.0% 13.2% 16.3% 18.2% 19.4% 1.76
Immunotherapy 0.9% 0.9% 1.2% 1.3% 1.7% 1.89
Radiation 13.1% 14.5% 16.1% 17.0% 14.5% 1.11
Surgery 3.0% 3.7% 3.7% 3.5% 3.6% 1.20

87
Table 2. Patient characteristics by quintile of risk-adjusted practice-level spending

Q1 Q2 Q3 Q4 Q5
Total no. of patients 6741 9926 9531 10403 7912
Mean no. of patients 41 61 58 63 49
Median no. of patients 30 40 36 38 34

Male (%) 59.8% 54.9% 55.7% 56.1% 56.1%
Age (years) 66.1 66.0 66.7 67.1 68.0
Charlson Comorbidity
Index (mean)
8.3 8.3 8.4 8.4 8.2
Cancer distribution

lung 25.9% 27.2% 27.6% 29.9% 26.5%
colorectal 25.0% 26.1% 25.5% 27.0% 24.5%
prostate 18.5% 15.5% 16.4% 15.2% 18.9%
kidney 5.8% 6.7% 6.1% 5.7% 5.9%
pancreas 5.3% 6.4% 6.5% 6.2% 6.2%
bladder 4.4% 3.6% 4.0% 3.8% 4.2%
gastric 2.9% 3.6% 3.2% 3.4% 3.2%
liver 2.3% 2.4% 2.1% 2.1% 2.2%
ovarian 9.9% 8.5% 8.6% 6.7% 8.4%

88
Table 3. Practice-level factors contributing to the variation in oncology spending
Practice-level Factors Coefficient p-value 95% CI
No. of oncologist visit per patient -$1,170 0.001 (-1824, 515)
No. of patients per oncologist $3 0.989 (-371, 376)
Total no. of oncologists $99 0.097 (-18, 215)
Average treatment decision*

Traditional chemotherapy $141 0.194 (-72, 353)
Targeted therapy $539 0.001 (235, 843)
Immunotherapy $767 0.237 (-506, 2041)
Radiation -$73 0.525 (-297, 152)
Surgery $218 0.552 (-500, 936)
*Average treatment decision is the proportion of patients receiving a specific cancer treatment (e.g. surgery) at a
practice; 1 unit increase is 1 percentage point increase.

89
Table 4. Hazard ratio of death by quintile of risk-adjusted practice-level spending

Unadjusted Cox Adjusted Cox
Hazard ratio 95% CI Hazard ratio 95% CI
Risk-adjusted practice-
level spending quintile

Q1 Reference - Reference -
Q2 1.04 (0.99, 1.10) 1.03 (0.98, 1.09)
Q3 1.23 (1.17, 1.30) 1.01 (0.95. 1.07)
Q4 1.03 (0.98, 1.09) 0.95 (0.90, 1.00)
Q5 0.96 (0.91, 1.02) 0.82 (0.78, 0.87)
Age category

18-50 - - Reference -
51-60 - - 1.07 (0.99, 1.14)
61-70 - - 1.13 (1.06, 1.21)
71-80 - - 1.35 (1.27, 1.45)
>80 - - 1.64 (1.52, 1.76)
Male - - 1.14 (1.10, 1.19)
Year of metastatic cancer
diagnosis
- - 0.79 (0.77, 0.80)
Year of first cancer
diagnosis
- - 0.95 (0.94, 0.97)
Charlson comorbidity
index
- - 1.07 (1.06, 1.08)

90
Cancer type* - -

colorectal - - 0.48 (0.42, 0.53)
lung - - 0.87 (0.79, 0.98)
prostate - - 0.52 (0.46, 0.58)
ovarian - - 0.54 (0.48, 0.61)
kidney - - 0.57 (0.50, 0.64)
pancreas - - 1.00 (0.89, 1.13)
bladder - - 0.66

gastric - - 0.91 (0.80, 1.04)
*Reference group: liver cancer.

91
Figure 1. Two-year all-cause mortality risk ratios

Reference group: Q1, lowest risk-adjusted practice-level spending quintile

92

Figure 2. Five-year all-cause mortality risk ratios

Reference group: Q1, lowest risk-adjusted practice-level spending quintile

93
References:
1. Marriotto AB, Yabroff KR, Shao Y, et al. Projections of the Cost of Cancer Care in the
United States: 2010–2020. J Natl Cancer Inst. 2011 Jan 19; 103(2): 117–128.
2. Baicker, K., and A. Chandra. 2004. Medicare spending, the physician workforce, and
beneficiaries’ quality of care. Health Affairs Supplemental Web Exclusives W4-184-97.
3. Fisher Elliott S, Wennberg David E, Stukel Therese A, Gottlieb Daniel J, Lucas FL, Pinder
Etoile L. The Implications of Regional Variations in Medicare Spending: the Content,
Quality, and Accessibility of Care. Part 1. Annals of Internal Medicine. 2003a;138(4):273–
287.
4. Fisher Elliott S, Wennberg David E, Stukel Therese A, Gottlieb Daniel J, Lucas FL, Pinder
Etoile L. The Implications of Regional Variations in Medicare Spending: the Content,
Quality, and Accessibility of Care. Part 2. Annals of Internal Medicine. 2003b;138(4):288–
298
5. MedPAC. 2009. Measuring regional variation in service use. In Report to the Congress:
December 2009. Washington, DC: MedPAC.
6. MedPAC. 2011. Regional variation in Medicare service use. In Report to the Congress:
January 2011. Washington, DC: MedPAC.
7. Romley JA, Jena AB, Goldman DP. Hospital spending and inpatient mortality: evidence
from California: an observational study. Annals of Internal Medicine. 2011;154(3):160–7.
Epub 2011/02/02.
8. Zhang, Y., K. Baicker, and J. P. Newhouse. 2010. Geographic variation in the quality of
prescribing. New England Journal of Medicine 363(21):1985-1988.

94
9. Baumgardner J, et al. Greater Spending Associated with Improved Survival for Some
Cancers in OCM-Defined Episodes. J Manag Care Spec Pharm, 2018 Jun;24(6):504-513.
10. Institute of Medicine. 2013. Variation in health care spending: Target decision making, not
geography. Washington, DC: The National Academies Press.
11. Li M, Lakdawalla D and Goldman D.
12. Clough JD, Patel K, Riley GF, et al. Wide variation in payments for Medicare beneficiary
oncology services suggests room for practice-level improvement. Health Aff (Millwood).
2015 Apr;34(4):601-8.
13. Berchick ER, Hood E, and Barnett JC. Current Population Reports, P60-264, Health
Insurance Coverage in the United States: 2017, U.S. Government Printing Office,
Washington, DC, 2018
14. Ellis P, Sandy LG, Larson AJ, et al. Wide variation in episode costs within a commercially
insured population highlights potential to improve the efficiency of care. Health Aff
(Millwood). 2012 Sep;31(9):2084-93.
15. Jacobson M, et al. Geographic Variation in Physicians' Responses to a Reimbursement
Change. N Engl J Med 2011; 365:2049-2052
16. Beck JR, et al. A convenient approximation of life expectancy (the "DEALE"). II. Use in
medical decision-making. Am J Med. 1982 Dec;73(6):889-97.
17. Brooks GA, Li L, Sharma DB, et al: Regional variation in spending and survival for older
adults with advanced cancer. J Natl Cancer Inst 105:634–642, 2013
18. Divi V, Tao L, Whittemore A, et al: Geographic variation in Medicare treatment costs and
outcomes for advanced head and neck cancer. Oral Oncol 61:83–88, 2016

95
19. Hassett MJ, Neville BA, Weeks JC: The relationship between quality, spending and
outcomes among women with breast cancer. J Natl Cancer Inst 106, 2014
20. Skolarus TA, Ye Z, Zhang S, et al: Regional differences in early stage bladder cancer care
and outcomes. Urology 76:391–396, 2010
21. Tannenbaum S, Soulos PR, Herrin J, et al: Regional Medicare Expenditures and Survival
Among Older Women With Localized Breast Cancer. Med Care 55:1030–1038, 2017
22. American Hospital Association. Report of the Task Force on Variation in Health Care
Spending. Washington, DC: American Hospital Association; 2011.
23. Bernstein J, Reschovsky JD, White C. Geographic variation in health care: Changing policy
directions. Washington, DC: National Institute for Health Care Reform; 2011.
24. Finkelstein A, Gentzkow M, Williams H. Sources of Geographic Variation in Health Care:
Evidence from Patient Migration. Q J Econ. 2016 Nov;131(4):1681-1726.
25. Phillipson TJ, Seabury SA, Lockwood LM, et al. Geographic Variation in Health Care: The
Role of Private Markets
26. Bradley EH, Canavan M, Rogan E, et al: Variation In Health Outcomes: The Role Of
Spending On Social Services, Public Health, And Health Care, 2000-09. Health Aff
(Millwood) 35:760–768, 2016
27. MedPAC. Health care spending and the Medicare program. Washington, DC: MedPAC;
2012.

Abstract (if available)

Abstract This dissertation evaluates the value of oncology services in the current oncology lexicon from different perspectives. As we see tremendous resources dedicated to treating cancer and the growth trajectory is projected to continue, the value of the oncology care delivered has become a crucial part of decision-making. Recent debates have focused on the value of new, innovative oncology treatments as there has been a striking increase in cancer drug prices. The definition of value may vary depending on the stakeholder involved, but nearly all shares the concept of clinical improvement achieved relative to cost. Therefore, correctly measuring the clinical benefits and costs associated with the oncology services is critical in recognizing the true value in oncology care

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Value in health in the era of vertical integration

PDF

Assessing value defects in limb preservation care

PDF

Health economics and outcomes research for informed decision making in rapidly evolving therapeutic areas

PDF

Three essays on estimating the effects of government programs and policies on health care among disadvantaged population

PDF

Economic, clinical, and behavioral outcomes from medical and pharmaceutical treatments

PDF

Essays in health economics

PDF

Characterization of health outcomes in patients with hemophilia A and B: Findings from psychometric and health economic analyses

PDF

Disease modifying treatments for Alzeimer's disease: modeling, value assessment and eligible patients

PDF

Long-term impacts of childhood adversity on health and human capital

PDF

Health care utilization and spending of the U.S. aging population

PDF

Essays in opioid use and abuse

PDF

Delivering better care for children with special health care needs: analyses of patient-centered medical home and types of insurance

PDF

Racial/ethnic differences in patient experiences with health care in association with the Gleason’s score level at prostate cancer diagnosis: findings from the SEER-CAHPS data

PDF

Thesis on Medicare Part D

PDF

Value based purchasing: decision-making processes underlying hospital acquisitions of orthopedic devices

PDF

Essays in pharmaceutical and health economics

PDF

The price and clinical impact of specialty drug management in the US healthcare system

PDF

Primary care physicians' experiences working within the patient-portal to improve the quality of patient care

PDF

Essays in health economics and provider behavior

PDF

Impact of pharmacy-based transitional care on healthcare utilization and costs

Asset Metadata

Creator Hu, Xiaohan (author)

Core Title Value in oncology care and opportunities for improvement

School School of Pharmacy

Degree Doctor of Philosophy

Degree Program Health Economics

Publication Date 12/05/2019

Defense Date 10/09/2019

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag economic value,innovative oncology drugs,OAI-PMH Harvest,oncology spending,quality of oncology care,Survival

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Goldman, Dana (committee chair), Lakdawalla, Darius (committee member), Romley, John (committee member)

Creator Email huxiaoha@usc.edu,staceeeey811@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c89-240703

Unique identifier UC11673835

Identifier etd-HuXiaohan-7995.pdf (filename),usctheses-c89-240703 (legacy record id)

Legacy Identifier etd-HuXiaohan-7995.pdf

Dmrecord 240703

Document Type Dissertation

Rights Hu, Xiaohan

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA