A

B

1

P

d

V A L U E I N H E A L T H 1 5 ( 2 0 1 2 ) 2 2 – 3 1

Avai lable onl ine at www.sc iencedirect .com

journal homepage: www.elsevier .com/ locate / jva l

The Cost-Effectiveness of Different Chemotherapy Strategies forPatients with Poor Prognosis Advanced Colorectal Cancer (MRC FOCUS)Andrea Manca, MSc, PhD1,*, Christian Asseburg, PhD2, Yolanda Bravo Vergel, MSc3, Matt T Seymour, MB, BS4,

ngela Meade, PhD5, Richard Stephens, PhD5, Mahesh Parmar, PhD5, Mark J. Sculpher, MSc, PhD1

1Centre for Health Economics, University of York, York, UK; 2Faculty of Health Sciences, University of Eastern Finland, Kuopio, Finland; 3Independent consultant,

arcelona, Spain; 4National Cancer Research Network Coordinating Centre, University of Leeds, Leeds, UK; 5Clinical Trials Unit, Medical Research Council, London, UK

A B S T R A C T

uamtwnsclntuKi

C

Objectives: To assess the value for money of alternative chemother-apy strategies for managing advanced colorectal cancer using irinote-can or oxaliplatin, either in sequence or in combination withfluorouracil. Methods: A cost-effectiveness model was developed us-ing data from the UK fluorouracil, oxaliplatin, and CPT11 (irinotecan) –use and sequencing (FOCUS) trial. The analysis adopted the perspectiveof the UK National Health Service. Input parameters were derived usinga system of risk equations (for probabilities), count data regressionmodels (for resource use), and generalized linear models (for utilities).Parameter estimates were obtained using Markov chain Monte Carlomethods, propagating the simulation values through the state-transi-tion model to characterize appropriately the joint distributions of ex-pected cost, survival and quality-adjusted life years for each treatmentstrategy. An acceptability frontier was used to represent the probabilitythat the optimal option is cost-effective at different values of the cost-

effectiveness threshold. Results: The base-case analysis used drug O

onom

al So

oi:10.1016/j.jval.2011.07.008

nit costs provided by a typical English hospital. First-line doublet ther-py combination therapy fluorouracil (5FU) plus irinotecan was theost cost-effective strategy at standard thresholds, with an incremen-

al cost-effectiveness ratio (ICER) of £14,877 (pound sterling) comparedith first-line 5FU until treatment failure followed by single agent iri-otecan. Other strategies were all subject to extended dominance. Aensitivity analysis using published drug (list) prices found the mostost-effective strategy would be first-line fluorouracil until failure fol-owed by 5FU plus irinotecan (ICER: £19,753). Conclusions: The combi-ation of 5FU and irinotecan (whether used first or second line) appears

o be more cost-effective than the single agent sequential therapiessed in the FOCUS trial, or 5FU plus oxaliplatin.eywords: advanced colorectal cancer, cost-effectiveness, FOCUS trial,

rinotecan, oxaliplatin.

opyright © 2012, International Society for Pharmacoeconomics and

utcomes Research (ISPOR). Published by Elsevier Inc.

Introduction

Colorectal cancer is the third most common cancer in the UnitedKingdom, with almost 30,000 new cases registered in England andWales in 2001, representing over 12% of all new cancer cases [1,2].“Advanced” colorectal cancer (ACRC) is described as a disease thatis either metastatic or too locally advanced for complete surgicalresection of the primary tumor. Approximately 55% of colorectalcancer patients in England and Wales have ACRC, either at thetime of initial presentation or later in the disease course [3,4].

A small but increasing proportion of patients with ACRC aresuitable for treatment with curative intent, usually involving ma-jor liver and/or lung resections and chemotherapy. For the major-ity of ACRC patients, however, the treatment objectives are non-curative: to relieve symptoms and improve quality of life, and tomodestly increase survival duration. Following trial results [5],palliative chemotherapy became the standard of care in ACRC inpatients who were able to tolerate these therapies. Since the mid-1990s the standard treatment for such patients was fluorouracil(FU) with folinic acid, administered in a variety of schedules, thetwo weekly de Gramont regimen (dG) or a modification of it (MdG)being the most popular in the UK. Subsequently, two cytotoxic drugs,

* Address correspondence to: Andrea Manca, Centre for Health EcE-mail: am126@york.ac.uk.

098-3015/$36.00 – see front matter Copyright © 2012, Internation

ublished by Elsevier Inc.

irinotecan (ir) and oxaliplatin (ox), were licensed. These drugs couldbe given either after or in combination with FU, and had good evi-dence of efficacy but with some additional toxicity [6–8]. They in-curred significant extra cost [9,10], although both have come off pat-ent recently, which reduces their acquisition costs and potentiallyincreases their cost-effectiveness. Since 2004, attention has shiftedto newer therapies targeting the epidermal growth factor and vascu-lar endothelial growth factor pathways.

The UK fluorouracil, oxaliplatin, and CPT11 (irinotecan) – use andsequencing (FOCUS) trial was designed to investigate the optimumcombination and sequencing for FU and either irinotecan or oxalip-latin in the UK population by comparing five alternative treatmentstrategies [11]. The published clinical results of this study indicatethat sequential single-agent therapy with FU followed by irinotecanproduces significantly inferior survival to the same two drugs used asa first-line combination “doublet.” In contrast, treatment strategiesinvolving FU alone followed by a second-line doublet were non-infe-rior to first-line doublet therapy [11].

In a fixed budget environment such as the UK National HealthService (NHS), where decision makers are expected to allocateavailable resources efficiently across a wide range of uses, it isessential to assess the extent to which the benefits of a given

ics, University of York, Heslington, YO10 5DD, York, UK.

ciety for Pharmacoeconomics and Outcomes Research (ISPOR).

wt

2btet

psmtu

23V A L U E I N H E A L T H 1 5 ( 2 0 1 2 ) 2 2 – 3 1

investment strategy are worth paying for. Cost-effectiveness anal-ysis (CEA) generates the information decision makers need tocarry out this task [12]. Using individual patient-level clinical andresource use data from the FOCUS trial, we developed an eco-nomic model [13] to estimate the long term cost-effectiveness ofthe five strategies investigated in the FOCUS trial. Reflecting thetrial population in FOCUS, only patients considered fit enough toundergo chemotherapy are relevant to the analysis and, given thatpalliative chemotherapy is now considered standard of care insuch patients, best supportive care was not considered a relevantcomparator.

Methods

Overview

Individual patient data (IPD) on the use and sequencing of thestudy and post-study drugs, non-drug health care resource useas well as patients’ health-related quality of life through theEuroQol five-dimension (EQ-5D) questionnaire instrument [14],

ere collected prospectively at each follow up visit during therial follow-up.

The median follow-up of the survivors in the FOCUS trial was6.5 months [11]. Because the differences in benefits and costsetween the alternative strategies are expected to extend beyondrial follow-up period, a model to estimate the long-term cost-ffectiveness of the five management strategies investigated inhe FOCUS trial was developed [15].

The model included costs from the NHS perspective [16] ex-ressed in UK sterling (2009 prices). Health outcomes were as-essed in terms of quality-adjusted life years (QALYs) based onortality and generic health-related quality of life (HRQoL) using

he EQ-5D data from FOCUS. Costs and QALYs were discountedsing a 3.5% annual discount rate [17].

Treatment strategies

Following the design of FOCUS, five treatment plans involvingthree different strategies for combining and sequencing FU, irino-

Fig. 1 – State tr

tecan, and oxaliplatin were modeled. These are described below:

Strategy A: the standard approach of sequential single-agent MdGregimen using FU until evidence of treatment failure, followed bysingle agent irinotecan;Strategy B-ir: first-line MdG regimen until treatment failure, fol-lowed by doublet therapy with MdG and irinotecan (IrMdG regi-men);Strategy B-ox: first-line MdG regimen until treatment failure, fol-lowed by doublet therapy with MdG and oxaliplatin (OxMdG regi-men);Strategy C-ir: first-line doublet therapy with the IrMdG regimen;andStrategy C-ox: first-line doublet therapy with the OxMdG regimen.

After failing doublet therapy, patients in the C-ir and C-ox armsof the FOCUS trial may have received non-FOCUS chemotherapyand eventually salvage therapy. This was due to a change in cross-over policy half way through the conduct of the FOCUS trial [11].We used salvage to mean any type of chemotherapy initiated afterpatients failed their trial chemotherapy. The costs of any non-FOCUS chemotherapy were included in the analysis. The manage-ment of patients after failing first-line therapy are reported in themain clinical trial publication [11]. All treatment regimens weredetailed in the protocol, with guidance on dose reductions anddelays for toxicity, and criteria to define treatment failure. Forpatients fit and willing to receive further chemotherapy after com-pleting the trial strategy, salvage chemotherapy options were of-fered in the protocol.

Model structure

The structure of the model was designed to reflect the treatmentstrategies investigated in the FOCUS clinical trial. Briefly, at anypoint in time, individuals were assumed to be in one of four mu-tually exclusive states (Fig. 1). Specifically, individuals could beeither alive and on a given chemotherapy plan (A to C-ox) andtreatment phase (prior to second line, prior to salvage, salvage), ordead (for treatment strategies A, C-ox, and C-ir, the transition isfrom first line FOCUS to post-FOCUS palliative/salvage treatment).Transitions between these states were modeled over 3 monthlyintervals and were governed by probabilities estimated directly

tion diagram.

ansi

from the FOCUS data.

24 V A L U E I N H E A L T H 1 5 ( 2 0 1 2 ) 2 2 – 3 1

Model inputs

There were three main categories of input parameter in the model,which included transition probabilities (including mortality); ge-neric health-related quality of life (estimated using EQ-5D data);and health care resource use (and costs). More details on the esti-mation methods used for each of these parameters are reported inthe statistical methods section.

Transition probabilitiesTransition probabilities were used to model the proportions ofindividuals in the different model states at each cycle in themodel. The appendix in Supplemental Materials at: doi:10.1016/j.jval.2011.07.008 details each strategy-specific transition matrix,together with a brief explanation of how each parameter was de-rived.

The trial dataset provided information on the occurrence of theevents being modeled (i.e., start and end treatment dates, death)as represented in Figure 1. Tunnel states, within which modeled“patients” remain for only one cycle, were used to handle timedependency in the transition matrix [18].

Generic health-related quality-of-life weightsEstimated QALYs were derived by weighting individuals’ esti-mated survival duration by their health related quality of life mea-sured in terms of their responses to the EQ-5D questionnaire. Thelatter was completed at baseline, at 6 and 12 weeks and every 12weeks thereafter during the trial follow-up period. Details of themethods used to estimate mean EQ-5D values from FOCUS dataand how these weights were assigned to the corresponding healthstates in the model are provided in the statistical methods sectionbelow. In addition, the appendix reports the values of the modelinput parameters that govern the predictions of mean EQ-5D val-ues associated with each treatment phase.

Health care resource use and costsThe key resource use components included in the analysis werethe drug acquisition and administration of the different chemo-therapy plans; primary care related (general practitioner atten-dance, general practitioner visit, district nurse visit); hospital re-lated (inpatient chemotherapy, general/acute inpatient, day casegeneral/acute, outpatient medical oncology visits, intensive careunit, and high dependency unit); and palliative treatment costs(including salvage chemotherapy, palliative radiotherapy and pal-liative surgery). The main components of the administrationcosts, the mean number of primary care visits, nurse and generalpractitioner home visits, and hospital attendances for chemother-apy administration were estimated directly from FOCUS data.

Briefly, the initial assessment before chemotherapy begins in-cluded abdominal computed tomography scan, chest x-ray, liverfunction test (LFT), full blood count (FBC), and urea and electro-lytes (U&E). The cost for a single-lumen Hickman line was ob-tained from providers. The disposable pump cost included all dis-posables and pharmacist time [19]. The monitoring tests betweenchemotherapy cycles included LFT, FBC, and U&E. The cost of thestaff nurse was calculated as cost per minute based on the recom-mended salary scale for 2009. Other fixed costs related to chemo-therapy administration, such as adjunctive drugs and monitoringtests, were based on the FOCUS protocol.

The appendix provides details of the methods used to estimatethe model inputs parameters attached to each of the above health-care resource use components, together with the values of theseinput parameters. The appendix can be found in supplementalmaterials at: doi:10.1016/j.jval.2011.07.008.

Unit costs used to quantify the total costs associated with theuse of the above health care resources are deterministic inputs in

the model.

In the base-case analysis, study drug unit costs were estimatedusing the figures provided by an (unnamed) NHS Trust (personalcommunication). This choice was dictated by the need to makesure our analysis is relevant to decision makers in the UK by usingcost figures that are as close as possible to the actual costs faced byNHS Trusts because some of the study drugs have recently comeoff patent and their acquisition cost has decreased significantly. Asensitivity analysis was conducted on the key drug costs in thestudy, with the objective of assessing their impact on the cost-effectiveness model results. The sensitivity analysis used BritishNational Formulary (BNF) prices [20] for the study drug costs, andthese figures are reported in brackets under the base-case values.Item-specific unit costs used in the base case and sensitivity anal-yses are reported in Table 1.

Drug acquisition costs were calculated using the most eco-nomic vial size and with allowance for wastage, and chemother-apy cycles are administered as a general medicine day case ormedical oncology outpatient visit as derived from the patient leveldata. As an approximation of the cost of palliative care for ad-vanced colorectal patients in the terminal phase, all patients whoreached the salvage stage were assumed to incur a cost of 2 weeksas hospital at home before death, based on national mean esti-mates of hospital at home/early discharge schemes [21]. Estimatedresource utilization was combined with unit cost information toquantify the average cost per patient associated with each chemo-therapy plan. Average salvage chemotherapy costs were based onthe drugs and dosage recommended in the FOCUS protocol. Pri-mary care and hospital-related costs were estimated based on na-tional databases [21].

Cost-effectiveness analysis

The FOCUS data were analyzed according to the intention-to-treatprinciple, including all randomized patients regardless of theircompliance. The model was run for a period of 10 years, afterwhich the great majority of patients were expected to have died.Therefore, the mean life years and QALYs per patient were calcu-lated for each plan, as well as the mean lifetime costs. The modelreflected the average patient characteristics at randomization inthe FOCUS trial.

The results of the model are presented in two ways. First, meansurvival, lifetime cost, and QALYs for each strategy are reportedand their cost-effectiveness compared, estimating incrementalcost-effectiveness ratios (ICERs) as appropriate, using standarddecision rules [22]. The interpretation of these ICERs required anexternal cost-effectiveness threshold such as the National Insti-tute of Health and Clinical Excellence’s (NICE) £20,000 to £30,000per QALY gained [16].

Second, cost-effectiveness acceptability curves (CEACs) areused to represent the decision uncertainty surrounding adoptiondecisions [23]. These curves plot the probability that a given strat-egy is more cost-effective than the others for a range of alternativethreshold cost-effectiveness values. Because there are five com-peting treatment strategies, an additional curve, namely the cost-effectiveness acceptability frontier (CEAF), is presented here. Thefrontier plots the probability that the optimal option is cost-effec-tive (at different threshold values) [23].

Statistical methods

Input model parameters were estimated from the FOCUS trial in-dividual patient level data as follows. A series of multinomial re-gression models was used to analyze the trial data on treatmentfailure and treatment switch, with the objective of estimating (im-plicitly time dependent) probabilities of transition from a givenstate (as described in Fig. 1) to another. Health care resource useassociated with each state in the model was estimated from the

FOCUS data using a zero-inflated Poisson model [24] for each re-

eral p

25V A L U E I N H E A L T H 1 5 ( 2 0 1 2 ) 2 2 – 3 1

source use category separately. The use of non-chemotherapy sal-vage treatments were analyzed using a Bernoulli model of occur-rence during each Markov cycle.

Finally, the EQ-5D values for specific states in the cost-effec-tiveness model were estimated using a two-part model [24]. First,a logit model was used to quantify the probability of a patient beingin full health (e.g., EQ-5D score equal to 1); and second, a scaledgamma model, was used to estimate the quality of life decrementsfor those individuals in the trial with an EQ-5D score of less than 1.Predictions from the two-part model were used to estimate meanEQ-5D scores for each health state in the model together with

Table 1 – Unit costs.

Item of resource use Base case value (Bprices in brackets

Chemotherapy acquisition drugs5-Fluorouracil (FU) £1.75 (£32)5-Fluorouracil (FU) £3.50 (£64)Irinotecan £13 (£53)Irinotecan £32 (£130)Oxaliplatin £31 (£165)Oxaliplatin £61 (£330)Capecitabine £45Capecitabine £258 (£295)Mitomycin C (£30) (£36)

Adjunctive drugs, devices, andmonitoring tests

Heparin sodium £0.25Normal saline £0.50Dexamethasone £1.73L-folinic acid £16 (£85)Granisetron £0.62 (£26)Dexamethasone £0.68 (£2)Dextrose £0.50Abdominal CT scan £75.21Chest X-ray £21.20Full blood count £2.42Urea and electrolytes £1.12Liver function test £0.61Single-lumen Hickman line £64.86Infusor system £35 (£77)

Primary care and hospital-related costs

GP attendance £23General practitioner visit £70District nurse visit £22Staff nurse, day ward £0.61Day case ward £61Outpatient ward £102Intensive care unit £1,402High-dependency unit £607Inpatient chemotherapy ward £255Inpatient general/acute ward £252Terminal patients – home care £299

Non-focus anticancer treatmentMF – PVI FU � mitomycin £140Oxalicap £496Irinocap £425Radiotherapy £86Palliative surgery £735

Note: Price year 2009 using Hospital and Community Health ServicesDosage cost calculations use by default the more appropriate packagCosts given in pound sterling (£).BNF, British National Formulary; CT, computed tomography; GP, gen

relevant measures of sampling uncertainty.

Probabilistic sensitivity analysis

The model was probabilistic in that all input parameters wereassigned probability distributions to reflect their sample vari-ability, as estimated from the trial data. Estimation of modelinput parameters and evaluation of the model outputs werecarried out simultaneously to fully represent the uncertainty inthe results [25]. The output from the Markov chain Monte Carlosimulation used to estimate the input parameters was used topropagate parameter uncertainty through the model, in order toestimate the joint distribution of costs and effects for each

Unit/dosage Source

25 mg/ml, 100-ml vial NHS Trust and [18]50 mg/mL, 100-ml vial NHS Trust and [18]20 mg/ml, 2-ml vial [18]20 mg/mL, 5-ml vial [18]50-mg vial NHS Trust and [18]100-mg vial NHS Trust and [18]150 mg [18]500 mg NHS Trust and [18]20-mg vial NHS Trust [18]

5 ml [18]250 ml UCLH NHS Trust2 mg, 20 tablets pack [18]17.5-ml vial NHS Trust and [18]3-ml amp NHS Trust and [18]8 mg/mL NHS Trust [18]250 ml, 5% UCLH NHS TrustItem York NHS TrustItem York NHS TrustItem York NHS TrustItem York NHS TrustItem York NHS TrustItem BardItem NHS Trust and [42]

Per visit [42]Per home visit [42]Per home visit [42]Per min patient contact [42]Visit [19, 42]Visit [19, 42]Day [19, 42]Day [19, 42]Day [19, 42]Day [19, 42]Day [19, 42]

Per cycle [18]Per cycle [18]Per cycle [18]Per cycle [19, 42]Per intervention [19, 42]

nd prices index (PSSRU 2010). PSSRU costs all include qualifications.depending on the number of vials required.

ractitioner; NHS, National Health Services.

NF)

pay ae size

treatment strategy [26].

pcalcs

26 V A L U E I N H E A L T H 1 5 ( 2 0 1 2 ) 2 2 – 3 1

The decision model was implemented using the statisticalsoftware package R (Foundation for Statistical Computing, Vi-enna, Austria) [27], whereas transition probabilities, survivalweights for each health states, costs, and health-related qualityof life were estimated using the freely available Bayesian soft-ware WinBUGS (Medical Research Council Biostatistics Unit,Cambridge, England) [28].

Results

A total of 2135 patients were accrued to the FOCUS trial, at 61centers, between May 2001 and December 2003. Their character-istics and full details of the treatment received and clinical out-comes are given in the main clinical report [11].

Mean costs per patient

Figure 2 illustrates the base-case total estimated mean cost peratient by health care resource use component. Hospital-relatedosts were the biggest contributor to the overall expected costs forll treatment plans (between 50% and 55% of total costs). In abso-ute terms, all strategies were very similar; with the main driver ofost differences among chemotherapy strategies being the acqui-ition cost of the chemotherapy drugs.

Table 2 reports the base-case estimated mean cost of eachtreatment strategy and associated estimated life years and QALYsover a lifetime time horizon. Strategy C-ir was the most expensiveoption, costing an average of £18,034 (95% Credibility Interval [CrI]£16,546–£19,548) per patient, as opposed to the standard treat-ment (Plan A), being the cheapest option with an expected cost of

0

2,000

4,000

6,000

8,000

10,000

12,000

UK

ster

ling

(200

9 pr

ices

) 14,000

16,000

18,000

20,000

CA - xo

Treatm

Terminal care Salvage non-chemo Nurse/

Fig. 2 – Mean cost per pa

£15,965 (95% CrI £15,151–£16,732). This difference is explained by

three factors, as can be seen from Figure 2: the higher acquisitioncost of the chemotherapy drugs, the higher hospital costs, and thehigher cost of the terminal care treatment.

Model validation

Figure 3 depicts the results of the quality assessment of the pre-dicted versus observed time to death in each of the five treatmentstrategies, at each weekly follow-up visit.

It can be seen that the mean squared prediction error (MSPE)is always less than 0.005, with the exception of week 96 wherethe MPSE is 0.01 for one of the strategies. The cumulative MPSEfor each of the treatment strategies over the entire time horizonranges between 0.018 (strategy B-ox) and 0.045 (strategy C-ir).

Cost-effectiveness

Consistent with the results of the clinical report [11], strategy C-ir wasfound to provide the longest expected mean survival duration of 1.47years (95% CrI 1.32–1.61), and the highest expected QALYs of 1.07 (95%CrI 0.97–1.17), which was followed by strategies B-ir and B-ox.

Relating QALYs to the cost the NHS has to bear to achieve them(in the base-case analysis), and using the NICE threshold range of£20,000 to £30,000 per QALY gained, leads to the conclusion (Table2) that first-line combination therapy fluorouracil plus irinotecan,strategy C-ir, is the most cost-effective strategy, among those con-sidered in the FOCUS trial, with an incremental cost per QALYgained of £14,877. For threshold values less than £14,877 per QALYgained, single agent FU followed by single agent irinotecan wouldbe the most cost-effective option.

Although ICER results can be used to determine the most cost-

B- Bxo - Cri -ir

strategies

Hospital Salvage chemo FOCUS admin FOCUS drugs

by category (base case).

ent

GP

tient

effective option based on mean costs and outcomes, they do not

Fst

itcmsg

ality-

27V A L U E I N H E A L T H 1 5 ( 2 0 1 2 ) 2 2 – 3 1

incorporate the uncertainty surrounding this decision. Figure 4Apresents the study results in the form of a family of CEACs, whichreveals the probability that the different strategies are the mostcost-effective over a range of potential threshold cost-effective-ness values. At NICE’s £20,000 to £30,000 per QALY gained, theprobability that strategy C-ir is cost-effective ranges between 59%(at a threshold of £20,000/QALY) and 68% (at a threshold of £30,000/QALY).

Figure 4B reports the cost-effectiveness acceptability frontier(CEAF). This curve combines the information from Table 3 andigure 4A, and plots the probability that the optimal option (thetrategy with the highest average cost-effectiveness) is cost-effec-ive (at different threshold cost-effectiveness values).

For threshold values less than £14,877 the optimal interventions strategy A (as we know from Table 2). The probability of beinghe most cost-effective strategy declines as this threshold in-reases. Because C-ir is, on average, more expensive but generatesore QALYs than strategy A, it becomes the optimal strategy as

oon as the system is prepared to pay at least £14,877 per QALYained.

Table 2 – Estimated mean costs and outcomes per patient

Strategy Mean costs

First-line fluorouracil untiltreatment failure, followed bysingle agent irinotecan.(Strategy A)

£15965 (£15151–£16732)

First-line combination therapyfluorouracil plus oxaliplatin.(Strategy C-ox)

£17104 (£15877–£18445)

First-line fluorouracil until failure,then combination therapyfluorouracil plus oxaliplatin.(Strategy B-ox)

£17120 (£15958–£18315)

First-line fluorouracil until failure,then combination therapyfluorouracil plus irinotecan.(Strategy B-ir)

£17163 (£15971–£18373)

First-line combination therapyfluorouracil plus irinotecan.(Strategy C-ir)

£18034 (£16549–£19502)

Note: Mean and 95% credibility interval (CrI). Discounted at 3.5%. CosICER, incremental cost-effectiveness ratio; LYs, life years; QALYs, qu

-0.05

-0.03

-0.01

0.01

0.03

0.05

0 12 24 36 48 60 72 84

Mea

n Sq

uare

d Pr

edic

�on

Err

or

Foll

A B-ir C

Fig. 3 – Predicted versus observed time to death at eac

Sensitivity analysis

The sensitivity analysis used list prices for pharmaceuticals pub-lished in the BNF to reflect the possibility that some NHS trustscould not obtain discounted prices. In the sensitivity analysis pre-sented here the most significant of these undiscounted prices re-late to the cost of FU, irinotecan, oxaliplatin, capecitabine, L-Fo-linic acid, and granisetron.

Under the new set of unit costs, and using the £20,000 to£30,000 per QALY gained threshold as a reference point, strategyC-ir was the most expensive option, costing an average of£23,627 (95% CrI £21,595–£25,676) per patient as opposed to thestandard treatment (plan A), being the cheapest option with anexpected cost of £19,266 (95% CrI £18,239–£20,345).

Under this scenario strategy B-ir is the most cost-effectivestrategy, among those considered in the FOCUS trial, with an in-cremental cost per QALY gained of £19,753. For threshold values ofat least £47,597 per additional QALY, combination therapy C-irwould be the most cost-effective treatment option. At the otherend of the spectrum, for threshold values less than £19,753 per

-case analysis).

Mean LYs Mean QALYs ICER

.29 (1.2–1.38) 0.93 (0.87–0.99) Cheapest (base for ICER)

.33 (1.2–1.45) 0.95 (0.87–1.04) Extendedly dominated(by strategy C-ir)

.35 (1.24–1.47) 0.97 (0.89–1.05) Extendedly dominated(by strategy C-ir)

.36 (1.24–1.48) 1.01 (0.92–1.1) Extendedly dominated(by strategy C-ir)

.47 (1.32–1.62) 1.07 (0.96–1.18) (To strategy A): £14,877

en in pound sterling (£).adjusted life years.

08 120 132 144 156 168 180 192 204 216 228

Up (Weeks)

B-ox B-ir

(base

1

1

1

1

1

ts giv

96 1

ow-

-ir

h weekly follow-up visit, by treatment strategy.

28 V A L U E I N H E A L T H 1 5 ( 2 0 1 2 ) 2 2 – 3 1

A Cost-effectiveness acceptability curves (CEACs) – base case

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0 10000 20000 30000 40000 50000 60000 70000 80000 90000 100000

Prob

abili

ty T

o B

e Co

st-E

ffec

�ve

Ceiling Ra�o (£)

A B-ir C-ir B-ox C-ox

£14,876 (C-ir against A)

B Cost-effectiveness acceptability frontier (CEAF) – base case

This curve plots the optimal strategy for different ceiling ratios

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0 10000 20000 30000 40000 50000 60000 70000 80000 90000 100000

Prob

abili

ty T

o Be

Cos

t-Eff

ec�

ve

Ceiling Ra�o (£)

Strategy A Strategy C-ir

Fig. 4 – (A) Cost-effectiveness acceptability curves (CEACs) – base case. (B) Cost-effectiveness acceptability frontier (CEAF) –

base case. Costs given in pound sterling (£).

osts r

29V A L U E I N H E A L T H 1 5 ( 2 0 1 2 ) 2 2 – 3 1

QALY gained, single agent FU followed by single agent irinotecanwould be the most cost-effective option. At thresholds of £20,000to £30,000 per QALY gained, the probability that strategy B-ir iscost-effective is between 50% and 65%.

Discussion

The objective of the analysis presented here was to estimate thecost-effectiveness of five treatment plans involving three differentstrategies for sequencing FU, irinotecan, and oxaliplatin, for thetreatment of ACRC, reflecting the design of, and using individualpatient level data provided by, the FOCUS clinical trial [11]. FOCUScompared sequential chemotherapy strategies versus combina-tion chemotherapy strategies. As such, the latter were first-linetreatments whereas the sequential chemotherapy strategies wereoffered as first-line, second line, and salvage depending on thetreatment arm. The structure of the cost-effectiveness model re-flects accurately the trial design and the alternative treatmentstrategies being investigated. To our knowledge this is the firsteconomic study to use individual patient level trial data to analyzethe cost-effectiveness of planned sequences of chemotherapiesfor ACRC from the UK NHS perspective for a patient lifetime hori-zon.

Relatively few published trials have compared different treat-ment plans for ACRC in which patients receive the same drugs indifferent sequences. Other than FOCUS [11], the Dutch colorectalcancer group trial [29] and the international LIFE trial [30] bothincluded evaluation of a strategy of initial single-agent fluoropy-rimidine, and a French trial [31] compared different sequences ofdoublet therapies. Two other trials of initial single versus doublettherapy have, as yet, been reported only in abstract form [32,33]. Inall these trials only minimal differences in overall survival, notreaching statistical significance, were observed between the treat-ment arms [34].

Golfinopoulos et al. [35] recently completed a large networkmeta-analysis of randomized trials – comprising 56,677 patientswho were randomly assigned across 580 treatment arms in 242trials published from 1967 to 2007 – comparing systemic treatmentregimens in patients with ACRC. Findings of this meta-analysissuggest considerable survival benefits in recent years due to the

Table 3 – Main costs and clinical results (sensitivity analys

Strategy Mean costs

First-line fluorouracil untiltreatment failure, followed bysingle agent irinotecan.(Strategy A)

£19266 (18239–20345)

First-line fluorouracil until failure,then combination therapyfluorouracil plus irinotecan.(Strategy B-ir)

£20846 (19369–22521)

First-line fluorouracil until failure,then combination therapyfluorouracil plus oxaliplatin.(Strategy B-ox)

£21333 (19813–22943)

First-line combination therapyfluorouracil plus oxaliplatin.(Strategy C-ox)

£23101 (21395–24909)

First-line combination therapyfluorouracil plus irinotecan.(Strategy C-ir)

£23627 (21595–25676)

Note: Mean and 95% confidence intervals (CI). Discounted at 3.5%. CoICER, incremental cost-effectiveness ratio; LYs, life years; QALYs, qu* The sensitivity analysis uses the British National Formulary drug c

introduction of new chemotherapy and immunotherapy drugs,

such as irinotecan, oxaliplatin, bevacizumab, and cetuximab. For apatient with an anticipated 1-year survival on 5FU plus leucovorin,the addition of irinotecan plus bevacizumab prolonged absolutesurvival by 8 months, and the addition of oxaliplatin plus bevaci-zumab, or irinotecan plus oxaliplatin prolonged absolute survivalby 4.7 months. Survival gains were more modest and less defini-tive with the addition of either irinotecan or oxaliplatin. Theseresults are consistent with those reported by the FOCUS trial,which is not surprising since FOCUS was one of the largest trialsincluded in the study by Golfinopoulos et al. [35]. Ideally, futureresearch work in this area should attempt to integrate the effec-tiveness results from this network meta-analysis with relevantresource use and HRQoL data to determine what is the most cost-effective treatment option for patients with ACRC in a given juris-diction.

Following the FOCUS trial, first-line treatment with irinotecanand 5FU is now superseding oxaliplatin and 5FU as the standardtreatment in the UK and has informed the design of the MedicalResearch Council FOCUS3 (NCT00975897) [36] and FOCUS4 (currentlybeing planned). Both trials use first-line combination therapy 5FUplus Irinotecan to be their control arm because this was the mostclinically and cost-effective therapy in FOCUS.

In some countries, though not in the UK, a biological therapytargeting either vascular endothelial growth factor (VEGF) or en-dothelial growth factor receptor (EGFR) is now commonly added tofirst-line chemotherapy for colorectal cancer. Given the high costof targeted therapies, it is therefore necessary to consider whetherthe health economic analysis of FOCUS, which did not includetargeted therapy, is of relevance in these health systems. The anti-VEGF agent bevacizumab has been available for addition to che-motherapy in many health systems since 2004 to 2006, although ithas not been adopted universally because of its high cost andrelatively modest clinical benefits [37–40]. Of relevance to the cur-rent study, there is no clear evidence of interaction between bev-acizumab and the three cytotoxic agents used in FOCUS, either interms of efficacy or toxicity, and what differences have been ob-served have tended to favor combinations with irinotecan/5FU.Therefore, although FOCUS did not include bevacizumab, andclearly does not provide any direct evidence of ICERs of bevaci-zumab-containing combinations, it seems likely that had bevaci-

ean LYs Mean QALYs ICER

(1.2–1.38) 0.93 (0.87–0.99) Cheapest (base for ICER)

(1.24–1.48) 1.01 (0.92–1.1) (to strategy A): £19,753

(1.24–1.47) 0.97 (0.89–1.05) Dominated (by strategy B-ir)

(1.2–1.45) 0.95 (0.87–1.04) Dominated (by strategy B-ir)

(1.32–1.62) 1.07 (0.96–1.18) (to strategy B-ir): £47,597

iven in pounds sterling (£).adjusted life years.eported in brackets in Table 1.

is)*.

M

1.29

1.36

1.35

1.33

1.47

sts gality-

zumab been added to each of the treatment strategies – assuming

clcletTafiiifmcwsHwmb

tHodrtestHct

uECltmertommfi

tfnlts

mpy(ccdtactswprwo

t£iait

ttSLat

30 V A L U E I N H E A L T H 1 5 ( 2 0 1 2 ) 2 2 – 3 1

similar use across strategies – would be expected to result in sim-ilar cost increase in each group leading to similar ICERs to thosereported here. Furthermore, their ranking would have been un-likely to differ. In contrast, EGFR therapies interact with cytotoxicchemotherapy in terms of toxicity and there appear to be complex,and as yet poorly understood, efficacy interactions with specificcytotoxic regimens. Consequently, the results of phase III trials ofEGFR therapies added to first-line chemotherapy are inconsistent,and even after exclusion of patients with KRAS-mutated tumors,several trials have reported no evidence of benefit [41,42]. There-fore, it is not possible to extrapolate ICER data from FOCUS toanti-EGFR combination therapy.

In 2008 Hind et al. [43] described a systematic review and aost-effectiveness model comparing the use of irinotecan, oxalip-atin, and raltitrexed for the treatment of advanced colorectal can-er. Most of the data used in the model came from the publishediterature, and where needed, were supplemented by access toarly FOCUS trial data. The authors concluded that three activeherapies appear to be most clinically effective and cost-effective.here are several methodological differences between our analysisnd that published by Hind et al. [43]. First, our model assesses theve treatment strategies in FOCUS in a direct head-to-head compar-

son, whereas the Hind et al. study compared each strategy includedn their model against first-line fluorouracil until treatment failureollowed by single-agent irinotecan (i.e., our strategy A). Thus, their

odel’s output, ICER, and subsequent conclusions cannot be directlyompared with ours. Furthermore, unlike the model by Hind et al.,e explicitly excluded patients who could benefit from down-staging

urgery, which may have affected our survival predictions. Third,ind et al. valued chemotherapy drugs use at 2004 list prices,hereas our model valued them both using 2009 prices and usingore recent discounted figures. Again, this may limit the compara-

ility of the ICER estimates between the two models.It is important to stress that the systematic literature review of

he existing economic studies in this clinical area conducted byind et al. [43] found that most studies were “. . .subject to a numberf important weaknesses” (see Hind et al., page 87). The authorsiscuss in detail these weaknesses and we invite the interestedeader to refer to their report. One important recommendation ofheir study was that “. . .the most significant improvement to existingconomic evaluations of these therapies would be the analysis of overallurvival, whereby evidence of effectiveness would be drawn from clinicalrials that have evaluated planned sequences of chemotherapies” (seeind et al., page 89). The authors continue, recommending theollection of resource use and utilities data as part of the trial. Allhis is exactly what the FOCUS trial was set up to do [44].

The FOCUS trial measured generic health related quality of lifesing the EQ-5D questionnaire. The availability of patient-levelQ-5D data represents an added value of the present analysis.aution should be exercised with the interpretation of health re-

ated quality of life data reported in different ACRC trials. Theiming of the questionnaire administration in relation to the ad-

inistration of chemotherapy and the time profiles of the toxicffects of the different chemotherapy regimens may limit compa-ability between studies. The problem of non-random censoring inhe context of ACRC trials, and its effect on health related qualityf life estimates, has been already reported [45]. No method ofultiple imputations for missing data, however, was used in ourodel because the overall compliance was very similar among the

ve treatment arms.Our analysis did not formally quantify structural model uncer-

ainty [46]. This is increasingly being recognized as an importanteature of any decision model, especially when modeling tech-iques are used to extrapolate the trial data beyond the trial fol-

ow-up period. In the FOCUS trial, 86% of the patients had died byhe end of the follow-up, thus any methodological uncertainty

urrounding our extrapolation results would only affect the re-

aining 14% of the sample who were alive at the end of the trialeriod. Running the model with a shorter time horizon (e.g., 5ears) did not affect the results of the cost-effectiveness analysisdetailed results not reported). Another element of our model thatould be subject to model uncertainty analysis concerns thehoice of the regression models used to analyze the FOCUS trialata to derive the various input model parameters. In our study,hese models were chosen based on a visual inspection of the datand of fitted residuals, rather than using formal model selectionriteria. Finally, another structural assumption relates to the al-ernative definitions of the health states, so that in exploringtructural uncertainty in the model one might want to assesshether or not structuring the model around health states such asrogression-free, post-progression, and death provides a betterepresentation of the data at hand. Future methodological workill explore the impact of such a structural model uncertainty onur results.

The main conclusion of the present study is that, at conven-ional levels of cost-effectiveness considered by NICE (£20,000 to30,000 per QALY gained), first-line combination therapy 5FU plusrinotecan (strategy C-ir) is the most cost-effective strategy, withn estimated ICER of £14,877 per QALY gained. The other therapiesnvestigated in FOCUS were dominated. These results are sensi-ive to the choice of unit costs for the study drugs.

Source of financial support: Principal trial funding was fromhe UK Medical Research Council (grant number E164/3). Addi-ional support in the form of discounted products was provided byanofi -Synthelabo, Rhone-Poulenc-Rorer (later Aventis), Wyeth-ederle, and Baxter. These companies did not have access to data,nd were not involved in trial design, data collection, the decisiono publish, or the writing of the article.

Supplemental Materials

Supplemental material accompanying this article can be found inthe online version as a hyperlink at doi:10.1016/j.jval.2011.07.008or, if a hard copy of article, at www.valueinhealthjournal.com/issues (select volume, issue, and article).

R E F E R E N C E S

[1] Office for National Statistics. Statistical publication notice - cancerincidence (2005). ISD; 2008. Available from: http://www.wales.nhs.uk/sites3/page.cfm?orgid�242&pid�18140 [Accessed May 6, 2009].

[2] Welsh Cancer Intelligence and Surveillance Unit. Cancer incidence inWales 1992–2001. 2005. Available from: http://www.wales.nhs.uk/sites3/docmetadata.cfm?orgId�242&id�18843 [Accessed August 10,2011].

[3] Centre for Review and Dissemination. The management of colorectalcancer. York, UK: Centre for Review and Dissemination, TheUniversity of York; 1997 [Contract No.: 6].

[4] Boyle P, Langman JS. ABC of colorectal cancer: epidemiology. BMJ 2004;321:805–8.

[5] Ragnhammar P, Hafstrom L, Nygren P, Glimelius B, for the SBU-group1. A systematic overview of chemotherapy effects in colorectalcancer. Acta Oncol 2001;40:282–308.

[6] Saltz LB, Cox, JV, Blanke C, et al. Irinotecan plus fluorouracil andleucovorin for metastatic colorectal cancer. Irinotecan Study Group.N Engl J Med 2000;343:905–14.

[7] Douillard JY, Cunningham D, Roth A, et al. Irinotecan combined withfluorouracil compared with fluorouracil alone as first-line treatmentfor metastatic colorectal cancer: a multi-centre randomised trial.Lancet 2000;355:1041–7.

[8] de Gramont A, Figer A, Seymour M, et al. Leucovorin and fluorouracilwith or without oxaliplatin as first-line treatment in advancedcolorectal cancer. J Clin Oncol 2000;18:2938–47.

[9] National Institute for Clinical Excellence. Capecitabine for thetreatment of advanced gastric cancer. Technology Appraisal Guidance.

2010 (TA191). Available from: http://www.nice.org.uk/nicemedia/live/13049/49875/49875.pdf [Accessed August 10, 2011].

[

[

[

[[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

31V A L U E I N H E A L T H 1 5 ( 2 0 1 2 ) 2 2 – 3 1

[10] Jones M, Hummel S, Bansback N, et al. A rapid and systematic reviewof the evidence for the clinical effectiveness and cost-effectiveness ofirinotecan, oxaliplatin and raltitrexed for the treatment of advancedcolorectal cancer. Health Technol Assess 2001;5:1–145.

11] Seymour M, Maughan TS, Ledermann JA, et al. Different strategies ofsequential and combination chemotherapy for patients with poorprognosis advanced colorectal cancer (MRC FOCUS): a randomisedcontrolled trial. Lancet 2007;370:143–52.

12] Claxton K, Sculpher MJ, Drummond MF. A rational framework fordecision making by the National Institute for Clinical Excellence.Lancet 2002;360:711–5.

13] Briggs AH, Sculpher MJ, Claxton K. Decision modelling for healtheconomic evaluation. Oxford: Oxford University Press, 2006.

14] Brooks R. EuroQol: the current state of play. Health Policy 1996;37:53–72.15] Sculpher MJ, Claxton K, Drummond MF, McCabe C. Whither trial-

based economic evaluation for health care decision making? HealthEcon 2006;15:677–87.

16] National Institute for Health and Clinical Excellence (NICE). Guide tothe methods of technology appraisal. London: National Institute forHealth and Clinical Excellence (NICE) 2008.

17] Barbieri M, Hawkins N, Sculpher M. Who does the numbers? The roleof third-party technology assessment to inform health systems’decision-making about the funding of health technologies. ValueHealth 2008;15:7:677–87.

18] Sonnenberg FA, Beck JR. Markov models in medical decision making: apractical guide. Med Decis Making 1993;13:322–38.

19] Iverson TJ, Hickish T, Schmitt C, et al. Irinotecan in second-linetreatment of metastatic colorectal cancer: improved survival and cost-effect compared with infusional 5-FU. Euro J Cancer 1999;35:1796–804.

20] British National Formulary. British National Formulary. London:British Medical Association and the Royal Pharmaceutical Society ofGreat Britain; 2010.

21] UK Department of Health. Reference cost 2008/2009. London:Department of Health; 2010; Available from: http://www.dh.gov.uk/en/Publicationsandstatistics/Publications/PublicationsPolicyAndGuidance/DH_111591. [Accessed June 2, 2011].

22] Johannensson M, Weinstein, M.C. On the decision rules of cost-effectiveness analysis. J Health Econ 1993;12:459–67.

23] Fenwick E, Claxton K, Sculpher M. Representing uncertainty: the roleof cost-effectiveness acceptability curves. Health Econ 2001;10:779–89.

24] Congdon P. Bayesian Models for Categorical Data. Chicester: JohnWiley and Sons Ltd.; 2005.

25] Spiegelhalter DJ, Best NG. Bayesian approaches to multiple sources ofevidence and uncertainty in complex cost-effectiveness modelling.Stat Med 2003;22:3687–709.

26] Briggs A, Goree R, Blackhouse G, O’Brien B. Probabilistic analysis ofcost-effectiveness models: choosing between treatment strategies forgastroesophageal reflux disease. Med Decis Making 2001;33:290–308.

27] R Development Core Team (2011). R: A language and environment forstatistical computing. R Foundation for Statistical Computing, Vienna,Austria. ISBN 3-900051-07-0, URL. http://www.R-project.org. [AccessedAugust 10, 2011].

28] Lunn DJ, Thomas T, Best N, Spiegelhalter DJ. WinBUGS - a Bayesianmodelling framework: concepts, structure, and extensibility. StatComput 2000;10:325–37.

29] Koopman M, Antonini NF, Douma J, et al. Sequential versuscombination chemotherapy with capecitabine, irinotecan, and

oxaliplatin in advanced colorectal cancer (CAIRO): a phase IIIrandomised controlled trial. Lancet 2007;370:135–42.

30] Cunningham D, Sirohi B, Pluzanska A, et al. Two different first-line5-fluorouracil regimens with or without oxaliplatin in patients withmetastatic colorectal cancer. Ann Oncol 2009;20:244–50.

31] Tournigand C, Andre T, Achille E, et al. FOLFIRI followed by FOLFOX6or the reverse sequence in advanced colorectal cancer: a randomisedGERCOR study. J Clin Oncol 2004;22:229–37.

32] Bouché O, Castaing M, Etienne PL, et al. Randomized strategical trial ofchemotherapy in metastatic colorectal cancer (FFCD 2000-05):preliminary results. J Clin Oncol 2007;25(Suppl.):abstr 4069.

33] Seymour MT, Maughan TS, Wasan HS, et al. Capecitabine andoxaliplatin in elderly and/or frail patients with metastatic colorectalcancer: The FOCUS2 trial. J Clin Oncol 2007;25(Suppl.):abstr 9030.

34] Seymour MT, Punt CJ. Sequential chemotherapy for advancedcolorectal cancer: should we ever start with a single cytotoxic agent?Curr Colorectal Cancer Rep 2008;4:130–8.

35] Golfinopoulos V, Salanti G, Pavlidis N, Ioannidis JP. Survival anddisease-progression benefits with treatment regimens for advancedcolorectal cancer: a meta-analysis. Lancet Oncol 2007;8:898–911.

36] UK Clinical Research Network. A study to determine the feasibility ofmolecular selection of therapy using KRAS, BRAF and topo-1 inpatients with metastatic or locally advanced colorectal cancer[database on the Internet]. Portfolio Database. 2008. Available from:http://public.ukcrn.org.uk/Search/StudyDetail.aspx?StudyID�6445[Accessed August 10, 2011].

37] National Institute for Clinical Excellence. Bevacizumab and cetuximabfor the treatment of metastatic colorectal cancer. TechnologyAppraisal Guidance 2007;33(TA118)

38] National Institute for Clinical Excellence. Bevacizumab and cetuximabfor the treatment of metastatic colorectal cancer (review). TechnologyAppraisal Guidance. 2009;33 (TA118). Available from: http://www.nice.org.uk/nicemedia/pdf/TA118Guidance.pdf [Accessed August 10, 2011].

[39] Saltz L. Current role of bevacizumab in colorectal cancer. Clin AdvHematol Oncol 2009;7:375–6.

[40] Saltz LB, Clarke S, Diaz-Rubio E, et al. Bevacizumab in combinationwith oxaliplatin-based chemotherapy as first-line therapy inmetastatic colorectal cancer: a randomized phase III study. J ClinOncol 2008;26:2013–9.

[41] Adams R, Meade A, Wasan H, et al. Cetuximab therapy in first-linemetastatic colorectal cancer and intermittent palliativechemotherapy: review of the COIN trial. Expert Rev Anticancer Ther2008;8:1237–45.

[42] ClinicalTrial.gov. FLOX � cetuximab (erbitux) for patients withmetastatic colorectal cancer and wild type K-RAS tumor [database onthe Internet]. Available from: http://clinicaltrials.gov/ct2/show/NCT00660582 [Accessed August 10, 2011].

43] Hind D, Tappenden P, Tumur I, et al. The use of irinotecan, oxaliplatinand raltitrexed for the treatment of advanced colorectal cancer:systematic review and economic evaluation. Health Technol Assess2008;12:iii–ix, xi–162.

44] Seymour M. Fluorouracil, oxaliplatin and CPT-11 (irinotecan), use andsequencing (MRC FOCUS): a 2135- patient randomized trial inadvanced colorectal cancer (ACRC). J Clin Oncol 2005;23:3518.

45] Van Cutsem E, Blijham GH. Irinotecan versus infusional 5-fluorouracil:a phase III study in metastatic colorectal cancer following failure offirst-line 5-fluorouracil. Semin Oncol 1999;26:13–20.

46] Jackson CH, Best NG, Richardson S. Bayesian graphical models for

regression on multiple data sets with different variables. Biostatistics2009;10:335–51.