Avai lable onl ine at www.sc iencedirect .com

V A L U E I N H E A L T H 1 6 ( 2 0 1 3 ) 1 0 8 1 – 1 0 9 0

1098-3015/$36.00 –

Published by Elsevie

http://dx.doi.org/10.

E-mail: oriana.ci

� Address correspBuilding, Salmon Po

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

Complete Cytogenetic Response and Major Molecular Response asSurrogate Outcomes for Overall Survival in First-Line Treatment ofChronic Myelogenous Leukemia: A Case Study for Technology Appraisalon the Basis of Surrogate Outcomes Evidence

Ciani Oriana, MSc1,�, Hoyle Martin, PhD1, Pavey Toby, PhD2, Cooper Chris, MA1, Garside Ruth, PhD3, Rudin Claudius, MD4,Taylor Rod, PhD1

1PenTAG, Institute of Health Services Research, University of Exeter Medical School, Exeter, UK; 2School of Human Movement Studies, University of Queensland,Brisbane, Australia; 3European Centre for Environment and Human Health, University of Exeter Medical School, Truro, UK; 4Royal Devon and Exeter NHSFoundation Trust, Exeter, UK

A B S T R A C T

Objectives: In 2012, the National Institute for Health and CareExcellence assessed dasatinib, nilotinib, and standard-dose imatinibas first-line treatment of chronic phase chronic myelogenous leu-kemia (CML). Licensing of these alternative treatments was based onrandomized controlled trials assessing complete cytogeneticresponse (CCyR) and major molecular response (MMR) at 12 monthsas primary end points. We use this case study to illustrate thevalidation of CCyR and MMR as surrogate outcomes for overallsurvival in CML and how this evidence was used to inform NationalInstitute for Health and Care Excellence’s recommendation on thepublic funding of these first-line treatments for CML. Methods: Weundertook a systematic review and meta-analysis to quantify theassociation between CCyR and MMR at 12 months and overallsurvival in patients with chronic phase CML. We estimated lifeexpectancy by extrapolating long-term survival from the weightedoverall survival stratified according to the achievement of CCyRand MMR. Results: Five studies provided data on the observationalassociation between CCyR or MMR and overall survival. Based on the

see front matter Copyright & 2013, International S

r Inc.

1016/j.jval.2013.07.004

ani@pcmd.ac.uk.

ondence to: Ciani Oriana, PenTAG, Institute for Heaol Lane, EX2 4SG Exeter, UK.

pooled association between CCyR and MMR and overall survival, ourmodeling showed comparable predicted mean duration of survival(21–23 years) following first-line treatment with imatinib, dasatinib,or nilotinib. Conclusions: This case study illustrates the consid-eration of surrogate outcome evidence in health technologyassessment. Although it is often recommended that the accept-ance of surrogate outcomes be based on randomized controlledtrial data demonstrating an association between the treatmenteffect on both the surrogate outcome and the final outcome, thiscase study shows that policymakers may be willing to accept alower level of evidence (i.e., observational association).

Keywords: chronic myeloid leukemia, complete cytogenetic response,dasatinib, health technology assessment, HTA, imatinib, intermediateoutcomes, major molecular response, nilotinib, surrogate end points,systematic review, technology appraisal.

Copyright & 2013, International Society for Pharmacoeconomics andOutcomes Research (ISPOR). Published by Elsevier Inc.

Introduction

Chronic myeloid leukemia (CML) is a myeloproliferative neo-plasm of hematopoietic stem cells [1]. CML used to be regardedas a progressive disease whose natural history evolves throughthree phases: the initial chronic phase, during which the diseaseis stable or at the most only slowly progressive, followed after avariable interval by transition through an accelerated phase to arapidly fatal blast crisis [2–4]. Approximately 90% of the peopleaffected by CML are diagnosed during the chronic phase, with amedian age at diagnosis of around 65 years [2]. In the UnitedStates, about 4800 to 5200 new cases are diagnosed every year,which corresponds to an annual incidence of 1.0 to 1.3 per100,000 population [5,6]. Similar annual age-standardized

incidence rates have been published for the United Kingdom(i.e., 1.1 per 100,000 for men and 0.7 per 100,000 for women) [7].Before the introduction of tyrosine kinase inhibitor (TKI) therapy,the median survival time after diagnosis was 6 years [6], with apredicted 5-year overall survival of 42.7% [8] and estimatedprevalence of 25,000 to 30,000 cases in the United States [6] andof 4,000 to 5,000 cases in the United Kingdom [9].

The molecular pathogenesis of CML is well understood, and thedisease presents the Philadelphia chromosome (Ph) as a molecularhallmark [10]. This fusion gene is the result of a reciprocalchromosomal translocation (i.e., t [9;22]), also known as breakpointcluster region-Abelson (BCR-ABL) oncogene, that codes for BCR-ABLtranscripts and fusion proteins with unusual tyrosine-kinaseactivity [11,12]. Diagnosis is confirmed by the identification of

ociety for Pharmacoeconomics and Outcomes Research (ISPOR).

lth Services Research, University of Exeter Medical School, Veysey

V A L U E I N H E A L T H 1 6 ( 2 0 1 3 ) 1 0 8 1 – 1 0 9 01082

either the Ph or the BCR-ABL transcripts, in peripheral blood orbone marrow cells, through cytogenetic [11–14] analysis or reversetranscriptase polymerase chain reaction (PCR), which can be semi-quantitative (real-time PCR or quantitative PCR) [15]. Followingrecognition of the importance of achieving a certain depth ofresponse at different time points for patients with newly diagnosedCML in chronic phase, the European LeukemiaNet has establishedguidelines on therapeutic milestones that should be achieved [15].A complete cytogenetic response (CCyR) is defined as absence ofthe Ph among at least 20 cells in metaphase in a bone marrowaspirate [16], while a major molecular response (MMR) is reached ifthe standardized BCR-ABL:ABL ratio is less than 0.1%, which isequivalent to a 3 log reduction from the 100% baseline foruntreated patients [17,18].

Knowledge of the molecular basis of this neoplastic diseasehas led to a new generation of drugs, the TKIs, radically changingthe previous standard of care based on interferon-alpha (IFN-α)for patients with CML [13,14]. Imatinib, the first rationally devel-oped selective TKI to be approved for the treatment of a cancer[19] by the European Medicines Agency in 2001, was rapidlyadopted as first-line medical treatment for CML in chronic phasein the National Health Service in the United Kingdom [20]. Theefficacy of imatinib in comparison with older treatments hasbeen confirmed in a single randomized controlled trial (RCT), theInternational Randomized Study of Interferon and STI571 (IRIS)trial [14], a prospective, multicenter, open-label, phase 3 RCTcomparing imatinib 400 mg/day with IFN-α plus cytarabine. Inearly 2012, two newer TKIs—dasatinib [17,21] and nilotinib [22–24]—initially promoted for the treatment of patients resistant orintolerant to imatinib [15,25], have been assessed by the NationalInstitute for Health and Care Excellence (NICE) as alternative first-line treatments to imatinib in England and Wales [26]. Theevidence of the relative effectiveness of these three alternativetreatment options was based on two comparative RCTs, one thatcompared dasatinib with imatinib [21] (Dasatinib vs. Imatinib inPatients With Newly Diagnosed Chronic Phase CML, the DASI-SION trial) and the other comparing nilotinib with imatinib [24](Evaluating Nilotinib Efficacy and Safety in clinical Trials-newlydiagnosed patients, the ENESTnd trial). In both trials, the primaryend points were biomarkers, that is, confirmed CCyR by 12months in DASISION and MMR at 12 months in ENESTnd.Although average survival from diagnosis can reach 15 years[25] among this population, these two trials provide only imma-ture data on overall survival with a maximum follow-up of 2years at the time of the assessment.

Central to this coverage decision, therefore, was considerationof CCyR and MMR as valid surrogate outcomes (i.e., biomarkersintended to substitute and predict for a final patient-relevantoutcome [27]) for long-term overall survival in first-line TKItherapy for chronic phase CML to determine estimates of life-years gained across alternative treatments [28].

The dual aims of this study were 1) to assess the evidencebase for the use of CCyR and MMR as surrogates for overallsurvival in patients with chronic phase CML treated with TKI (i.e.,dasatinib, nilotinib, and imatinib) and 2) to describe how thisevidence was used to predict long-term survival in the relatedcost-effectiveness model. The policy implications of the valida-tion and use of surrogate outcomes in coverage decisions will bediscussed.

Methods

This study consisted of two distinct methodological steps: 2) asystematic review and meta-analysis of the evidence base toquantify the association between CCyR and MMR as surrogatesfor overall survival in chronic phase patients with CML treated

with first-line TKIs and 2) modeling of the observed CCyR andMMR at 12 months to predict long term (412 months) patientsurvival in first- and second-generation TKI therapies.

Systematic Review and Meta-Analysis

Our systematic review and meta-analysis was conducted andreported in accord with the Preferred Reporting Items for System-atic Reviews and Meta-Analyses statement [29].

Search strategyWe initially identified studies from a previous systematic reviewof imatinib for first-line treatment of CML in chronic phase [30].The following bibliographic databases were searched: Medline,Medline in Process, Embase, PsycINFO (all via OVID), TheCochrane Library, Web of Science (via ISI), and Econlit (via CSA)from October 2002 (the end date of the previous systematicreview [30]) to May 2012. The searches were limited to the Englishlanguage. The Medline search strategy is reported in the Supple-mental Material found at http://dx.doi.org/10.1016/j.jval.2013.07.004.

Inclusion criteriaStudies were included if they considered 1) adults (median age 418 years) with chronic phase CML based on cytogenetic and/orfluorescence in situ hybridization and/or reverse transcriptasepolymerase chain reaction results; 2) patients treated withdasatinib, nilotinib, or imatinib; 3) patients naive to previousinterferon or TKI treatment; and if they 4) reported the associa-tion between CCyR or MMR at 12 months and overall survival. Weexcluded studies published only as conference abstracts, narra-tive reviews, editorials, opinion pieces, and individual casestudies, or studies whose findings were not judged to be general-izable to the CML population in the United Kingdom or Westerncountries. Where a study had been reported in several publica-tions, we considered the article with the longest follow-up. Titles,abstracts, and full text of any potentially relevant studies wereindependently screened by two reviewers (O.C. and T.P.) with anydiscrepancies resolved by discussion, with the involvement of athird reviewer if necessary.

Data extraction and quality assessmentThe methodological quality of included studies was assessedaccording to a modified list of criteria specified by the NationalHealth Service Centre of Reviews and Dissemination [31]. Studycharacteristics and data were extracted by one reviewer (O.C.) byusing a standardized data extraction form and independentlychecked by a second reviewer (T.P. or R.T.). To judge the reliabilityof CCyR and MMR at 12 months as surrogate measures for long-term overall survival, we referred to the following surrogatevalidation criteria: 1) evidence from RCTs demonstrating treat-ment effects on the surrogate correspond to treatment effects onthe patient-relevant outcome, 2) evidence from observationalstudies demonstrating consistent association between surrogateoutcome and final patient-relevant outcome, and 3) evidence ofbiological plausibility of relationship between the surrogate out-come and the final patient-relevant outcome [32].

Data analysesFor each study, overall survival was extracted at each yearfollowing trial recruitment (or randomization) up to the latestfollow-up point reported, separately according to whether a CCyRor an MMR was achieved at 12 months. In all studies, overallsurvival data were estimated by the Kaplan-Meier method by usinglandmark analysis to evaluate differences in the final patient-relevant outcomes between responders and nonresponders.

V A L U E I N H E A L T H 1 6 ( 2 0 1 3 ) 1 0 8 1 – 1 0 9 0 1083

The landmark method determines each patient’s response at afixed time point, with survival calculated from that time pointonward and associated statistical tests being conditional onpatients’ landmark responses [33]. Data digitalization software(WinDIG Version 2.5) was used to extract data from Kaplan-Meiersurvival curves.

For consistency, we selected 12 months after the start of first-line therapy as the landmark for our analysis, as the DASISION[21] and ENESTnd [24] trials consider, respectively, the rate ofconfirmed CCyR and MMR at 12 months postrandomization asprimary end points. Average overall survival rates at yearlyintervals were estimated for both responders and nonresponders,weighted by the initial number of patients in the two groups foreach trial. Wilson 95% confidence intervals (CIs) were derived foreach point estimate assuming binomial distributed variables andno censoring of data [34]. Analyses were undertaken by usingSTATA v.11.2 (StataCorp, TX).

Modeling and Data Extrapolation

The above systematic review and meta-analysis provided aliterature-based estimation of overall survival, according towhether patients with CML achieved either a CCyR and MMRresponse or not. A four-step analytical approach was then under-taken to estimate long-term overall survival separately forimatinib, dasatinib, and nilotinib treatment.

–

Step 1: CCyR and MMR response rates at 12 months fordasatinib from ENESTnd [24] and for nilotinib from DASISION[21] were derived by using a WinBUGS mixed-treatmentcomparison analysis [35,36]. The appropriateness of the indi-rect comparison was assessed by checking that the baselinecharacteristics of the two trials were similar. A fixed-effectspairwise meta-analysis [37] was then undertaken to obtain anoverall estimate of the proportion of patients achieving aCCyR and separately an MMR for each treatment [38].

–

Step 2: Estimation of CML-related mortality from historicaltrial data [38]. Mortality was assumed to occur because ofCML-related causes and non-CML causes. Given limited andimmature historical data on CML-related death, the proba-bility of CML-related death was assumed constant over time(as this was deemed most parsimonious), and to depend onwhether a CCyR was achieved. In a separate exercise, theprobability of CML-related death was assumed to depend onwhether an MMR was achieved. Non-CML mortality was takenfrom UK life tables [39], and the age at diagnosis wasestimated as the average age at diagnosis across all historicaltrials, weighted by the number of responders or nonrespond-ers in each trial, as appropriate. The constant probability ofCML-related death was estimated to minimize the sum ofsquares of differences between the actual historical overallsurvival and modeled overall survival at each year.

–

Step 3: Estimation of overall survival separately for respondersand nonresponders given a cohort of patients starting first-line treatment at age 57 years (i.e., the mean age at diagnosisin the United Kingdom) [7]. Overall survival was estimated byapplying mortality from the general population with startingage 57 years and the appropriate estimate of CML-relatedmortality from step 2.

–

Step 4: Estimation of overall survival for each treatment arm(i.e., imatinib, dasatinib, and nilotinib) by averaging theresponder and nonresponder overall survival, estimated instep 3, weighted by the proportion of patients who did and didnot achieve a response to first-line treatment at 12 months.

We compared our estimates of expected overall survival with theactual 24- and 36-month overall survival from the ENESTnd trial

[40,41], the 24-month overall survival from the DASISION trial[42], and the longer term (i.e., 7 years) imatinib survival data fromthe IRIS trial [43]. In addition, two sensitivity analyses wereperformed on the pool of articles that contributed historical databy 1) excluding IRIS trial reports [43,44] and 2) including theunique dasatinib- and nilotinib-treated patient cohort identifiedin the Jabbour et al. study [45]. Given that the IRIS trial has beenused for validation, the former was performed to test theinfluence of IRIS data on the estimated surrogate relationship,while the latter was performed to check whether the samerelationship might be specific to type of TKIs.

Modeling analyses were carried out by using WinBUGS (MRCBiostatistics Unit, Cambridge, UK) and the Excel “Solver” function(Excel 2012 Microsoft Corporation, Redmond, WA).

Results

Study Identification

The process of study selection is summarized in Figure 1. Sixpublications met the inclusion criteria, reporting on five separatestudies—two RCTs and three cohort studies (Table 1).

Study and population characteristicsWe were able to include five studies in the quantitative analysis.One study, performed in India, was judged to be unlikely toreflect the clinical management of patients with CML in devel-oped economies and therefore excluded [49]. Only one arm in acohort study reported patients with CML who were treated bydasatinib or nilotinib [45], with all the others considering imatinibtreatment. We therefore decided to include only the imatinibtreatment arm from the same study in our base-case analysisand to contrast the overall results with those reported for thedasatinib and nilotinib arm in Jabbour et al. [45]. As for the twoRCTs, only the arms receiving standard-dose imatinib as first-linetherapy were considered. This choice was taken because 1) theIRIS trial was inadequate to demonstrate a survival benefit forimatinib relative to IFN-α therapy in newly diagnosed Phþchronic-phase CML in the light of the high rate of crossover(65% at 72-month follow-up) from IFN-α plus cytarabine toimatinib [50], and 2) Hehlmann et al. [48] compared the 400 mg/day imatinib with the high-dose therapy (i.e., 800 mg/day) orcombined therapy with interferon, which were not among thetreatment options under comparison in our analysis.

Study quality and hierarchy of surrogate evidenceThe included studies consistently showed moderate to goodinternal validity (see Table 1 in Supplemental Material found athttp://dx.doi.org/10.1016/j.jval.2013.07.004) and were therefore allconsidered in the base-case analysis. In the two RCTs [43,44,48],the association between CCyR and MMR and overall survival wasexamined as a stratified comparison of overall survival in MMRand CCyR responders versus nonresponders for the imatinib 400mg/day arm only. Thus, the level of surrogate outcome evidenceidentified by this review was entirely observational, that is,“level 2” evidence according to the three-level surrogacy evaluation scheme proposed by Elston and Taylor [32].

Association between CCyR or MMR at 12 months and overallsurvivalTable 2 shows the weighted pooled mean overall survival (and 95%CI) at yearly intervals, up to 7 years after the initiation of imatinibtreatment, according to achievement of CCyR and MMR (or not) at 12months [43–48]. For imatinib-treated patients with CML, the impactof achieving a CCyR at 12 months progressively translate into a

Fig. 1 – Flow diagram of the study inclusion process.

V A L U E I N H E A L T H 1 6 ( 2 0 1 3 ) 1 0 8 1 – 1 0 9 01084

survival benefit compared with patients who do not achieve suchresponse. The advantage of achieving an MMR at 12 months interms of overall survival rates, however, is less clear. Overall survivalrate estimates for imatinib-treated patients are slightly lower thanthe level of overall survival seen in the cohort of patients treatedwith nilotinib or dasatinib who achieve a CCyR at 12 months aftertreatment initiation in Jabbour et al. [45], although without reachingstatistical significance (logrank test: P ¼ 0.80).

Prediction of long-term overall survivalThe estimated long-term overall survival for responders andnonresponders compared with our literature-based synthesizedhistorical data and survival for the general population in Englandand Wales is shown in Figure 2. After considering CCyR andMMR response rates by 12 months for each of the three treatmentarms under comparison (see Table 2 in Supplemental Materialfound at http://dx.doi.org/10.1016/j.jval.2013.07.004), we were ableto stratify long-term extrapolated overall survival by therapy(Fig. 3).

The extrapolated long-term overall survival curves for ima-tinib, dasatinib, and nilotinib treatment appear very compara-ble. The estimated mean overall survival across the threetherapies (see Table 3 in Supplemental Material found athttp://dx.doi.org/10.1016/j.jval.2013.07.004) ranges from 2.7 to4.1 years less than the life expectancy of the general populationof England and Wales. Taking into account the potentialuncertainties in the estimation of trial-based response rates,according to our model, people treated with first-line imatinibare expected to survive 21.3 years on average, 1.4 years lessthan people treated with nilotinib and dasatinib, using theCCyR data. They are expected to survive 22.0 years, 0.6 yearsless than patients treated with dasatinib and nilotinib, whenusing the MMR data. Results from the sensitivity analysesshowed little or no impact on estimated mean overall survivalwhen data from the IRIS trial were excluded [43,44] or whendata from the dasatinib and nilotinib cohort [45] were included,thus supporting the consistency of the surrogate to final outcome relationship across the interventions.

We checked the accuracy of our estimates by comparing mod-eled overall survival and trial-based overall survival from the twoRCTs of first-line dasatinib [21] and nilotinib [24] and from theimatinib arm of the IRIS trial [43]. It appears that themodeled overallsurvival is consistent with these data:

–

at 2-year follow-up, dasatinib overall survival observed inDASISION was 95% compared with 97% estimated by ourmodel [42];

–

at 2-year follow-up, nilotinib overall survival observed inENESTnd was 97% [40] compared with 97% in the model; at3-year follow-up, the overall survival observed in the trial was95% [41] compared with 95% estimated by the model;

–

imatinib observed overall survival was 95% and 96% inDASISION and ENESTnd, respectively, compared with 96% inthe model based on the CCyR surrogate relationship and 97%based on the MMR surrogate relationship [40,42].

In addition, the estimated overall survival for the imatinib armclosely predicts the actual overall survival in the imatinib arm ofthe IRIS RCT (see Fig. 1 in Supplemental Material found at http://dx.doi.org/10.1016/j.jval.2013.07.004). This is not a completelyindependent verification of overall survival through this method,because some of the data on overall survival for imatinibresponders and nonresponders from the IRIS RCT were also usedto estimate the overall survival surrogate relationships. Nonetheless, it serves as useful calibration of the model’s survival outputsbecause other historical data also heavily influenced the surro-gate overall survival estimates.

Discussion

The molecular biology of CML supports the adoption of both CCyRand MMR as potential markers for monitoring of disease pro-gression [13,51]. It has also been shown, however, that TKIs canhave potential unexpected off-target effects (i.e., stem cellschromosomal instability, inhibition of proinflammatory func-tions [52–55]) that may call into question the ability of these

Table 1 – General characteristics of studies used to estimate surrogate relationships between response and survival.

de Lavalladeet al. [46]

Kantarjian et al.[47]

Jabbour et al.�

[45]Hehlmann et

al. [48]Hughes et al. [43]

(IRIS)Roy et al.[44] (IRIS)

Country UK US US US Germany International InternationalYear published 2008 2008 2011 2011 2011 2010 2006Study type Cohort Cohort Cohort Cohort RCT RCT RCT (retrospective

comparison)N patients

(TKI arm)204 276 73 154 324 476 551

Median age (y)(range)

46 (18–79) 48 (15–84) 48 (15–78) 47 (18–85) 54 (16–88) 50 (20–69) 50 (18–70)

Intervention Imatinib400 mg/d

Imatinib400 mg/d or800 mg/d

Imatinib400 mg/d

Dasatinib ornilotinib

Imatinib400 mg/d

Imatinib400 mg/d

Imatinib 400mg/d

Comparator None None None None Imatinib400 mg/d

combined withIFN Imatinib800 mg/d

IFN-α pluscytarabine

IFN-α pluscytarabine

Median follow-up(mo)

38 48 110 28 43 77 42

Comparisonbetweenresponders vs.nonresponders

CCyR vs. no CCyR CCyR vs.minor CyR†

MMR vs. no MMR

CCyR vs. noCCyR

CCyR vs. noCCyR

MMR vs. noMMR

MMR vs. noMMR

CCyR vs. noCCyR

Male sex, n (%) 116 (57) NA NA NA 194 (60) 296 (62) 341 (62)Median time

from diagnosisto treatment (mo)

1.7 1.1 1 1 0.6 (within 6 mobefore study

entry)

(within 6 mobefore study

entry)Sokal risk

score‡, n (%)Low: 59 (29) Low: 179 (65) Low: 50 (68) Low: 118 (77) EuroSCORE Low: 175 (37) Low: 201 (37)

Low: 113 (35) Intermediate: 99(21)

Intermediate: 111(20)

Intermediate: 86(42)

Intermediate: 76(27)

Intermediate:22 (30)

Intermediate: 27(18)

Intermediate: 172(53)

High: 63 (13) High: 71 (13)

High: 59 (29) High: 21 (8) High: 1 (2) High: 9 (5) High: 39 (12) Not known: 139(29)

Not known: 168(30)

Population Phþ CML in CP Phþ CML in CP CML in CP§ CML in CP§ CML in CP Phþ CML in CP Phþ CML in CP

CcyR, complete cytogenetic response, CP, chronic phase; EuroSCORE, European System for Cardiac Operative Risk Evaluation; IFN-α, interferon-alpha; IRIS, International Randomized Study ofInterferon and STI571; MMR, major molecular response; Phþ, Philadelphia chromosome positive; RCT, randomized controlled trial; TKI, tyrosine kinase inhibitor.� Jabbour et al. [45] studied three cohorts of patients treated, respectively, with imatinib 400 mg daily, imatinib 800 mg daily, and second-generation TKIs (i.e., nilotinib and dasatinib). We excludedthe imatinib 800 mg daily arm for the purpose of this analysis as nonlicensed dose. Data from imatinib 400 mg daily cohort were included in our base-case analysis; data from the second-generation TKIs cohort were used in a sensitivity analysis.

† The group of people achieving a minor cytogenetic response at 12 mo after the first-line treatment initiation (N ¼ 5) in Kantarjian et al. [47] study report was excluded from the pooled overallsurvival average estimate.

‡ Sokal score is a prognostic classification system for patients with chronic myeloid leukemia that is designed to identify patients with low, intermediate, or high risk of poor outcome.§ More than 85% of the patients are Phþ.

VALUE

INH

EALTH

16

(2013)1081–1090

1085

Table 2 – Pooled weighted mean overall survival (and 95% CI) by CCyR and MMR response (or not) at 12 mo afterthe starting of imatinib therapy.

Time (mo) OS % (95% CI)�

CCyR No CCyR P† MMR No MMR P†

12 100 (99.4– 100) 100 (98.1–100) 1.00 100 (99.1– 100) 100 (99.4– 100) 1.0024 98.3 (96.8– 100) 94 (89.7–96.5) 0.15 100 (99.1– 100) 96.7 (95.0– 97.9) 0.4236 97.7 (96.2–98.6) 88.9 (83.9– 92.2) 0.03 99.2 (98.0– 99.8) 95.7 (93.8– 97.1) 0.3548 98.3 (96.1– 99.4) No data Not calculable 96.6 (94.4– 97.9) 93.3 (91.0– 95.0) 0.2060 97.4 (95.2– 98.6) 73.6 (62.4– 82.4) 0.02 96.9 (95.1– 98.1) 89.4 (86.9– 91.5) 0.1172 98 (88.4–99.6) No data Not calculable 96.0 (93.2– 97.5) 90 (87.0– 92.3) o0.0184 94 (82.1– 97.7) No data Not calculable 92.5 (87.6– 95.9) 89.2 (83.5– 93.4) NS

CcyR, complete cytogenetic response; CI, confidence interval; MMR, major molecular response; NS, nonsignificant; OS, overall survival.� Wilson 95% CIs.† P value for difference in the OS rate between responders and nonresponders. Where more than one studies are available, P values derivedfrom t-statistic meta-regression analysis. Where only one study is available, P values derived from the study publication.

V A L U E I N H E A L T H 1 6 ( 2 0 1 3 ) 1 0 8 1 – 1 0 9 01086

markers to fully capture the efficacy and safety pathways of theTKI therapy. In this article, we assessed the empirical evidencefor the use of CCyR and MMR at 12 months as surrogate outcomesfor overall survival in patients with chronic phase CML receivingfirst-line TKI treatment.

The results confirm the current adoption of CCyR at 12months as a gold standard for a good response, whereas MMRprovides a measure of success rather than a measure of failure(i.e., not achieving high levels of molecular response does notconstitute treatment failure in patients with CML) [25]. Oursystematic review identified three cohort studies [45–47] andtwo RCTs [43,44,48] examining the association between thesetwo biomarkers and overall survival in patients with chronicphase CML receiving first-line imatinib. While these studiesshowed a consistent association between CCyR and MMR andlong-term (i.e., 1–7 years) overall survival, this was based onobservational analyses comparing responders versus nonrespond-ers. Based on the pooled observational association between CCyRand MMR and overall survival, our modeling showed comparablepredicted mean duration of survival (21–23 years) following first-line treatment with imatinib, dasatinib, or nilotinib.

Although a plausible biological rationale constitutes a basicstep toward the identification of a surrogate outcome [32,56,57], itis not sufficient by itself to prove that the treatment effect on thesubstitute end point may predict the treatment effect on the finalpatient-relevant outcome. Empirical evidence of an associationbetween these end points and final patient-relevant outcome, aswell as between treatment effects on them is also needed. Ideally,this evidence should be in the form of multiple RCTs that assessthe effects of the treatment on both the end point marker andfinal patient-relevant outcome at relevant follow-up time [58].

Findings of Previous Studies

To our knowledge, this is the first systematic review and meta-analysis to examine the scientific basis of the validation of CCyRand MMR as surrogates for long-term overall survival in patientswith chronic phase CML treated with TKIs. Two previous studieshave examined this question in patients treated with IFN-α.

Anstrom et al. [59] fitted a proportional hazards model toestimate long-term survival by conditioning on CCyR at 2 yearslandmark time point. Their data sources were the IRIS trial at 19-month follow-up [14] and four clinical trials [60–63] assessingpatients treated with IFN-α plus low-dose cytarabine. Theypredicted a residual life expectancy after CCyR at 2 years of 16.7years and of 5.8 years for non-CCyR cohorts. In comparison, ourestimates were 24.5 and 14.3 years, respectively. The life

expectancy estimates of Anstrom et al. [59] for first-line imatinibwere 15.3 years compared with our estimate of 21.3 years usingCCyR as a surrogate. The differences in the estimates can beexplained first by the different landmark time considered for theCCyR (i.e., 1 year vs. 2 years), second by the choice of the baselinesurvival functions (i.e., life-table estimates for the general U.S.population weighted according to baseline age and sex distribu-tions of the IRIS population vs. UK life-table estimates weightedby the number of responders or nonresponders in each trial, asappropriate), and third by the assumption that long-term survivalestimates for imatinib-treated patients are similar to thosederived from two cohorts of IFN-α–based regimens with [61] orwithout [60] CCyR at 2 years.

Schrover et al. [64] assumed that prolonged survival afterattaining a major cytogenetic response (i.e., 0%–35% Phþ cellsamong at least 20 cells in metaphase in a bone marrow aspirate)may be independent of treatment and developed a survival modelfor patients with chronic phase CML using a logistic regression topredict survival according to major cytogenetic response rate. Theyestimated an average difference in survival between respondersand nonresponders at 2 and 4 years after landmark of 15.0% and25.8%, respectively, and predicted a proportion of patients alive at 5year of 70%. Our model predicts more favorable outcomes for thepatients (i.e., fitted 5-year survival in patients without CCyR is 77%,98% in patients with CCyR); however, our results cannot be easilycompared with those reported by Schrover et al. [64], given thatthey considered a different surrogate end point and data derivedfrom IFN-α– based RCTs.

Recently, a systematic review and meta-analysis has evaluatedthe efficacy and safety of second-generation TKIs (including bosu-tinib) versus imatinib [65]. The inclusion criteria for this study wereslightly different from those in our systematic review: Gurion et al.[65] considered only RCTs, also published as conference abstracts,with no restriction on adult population. Although the objective wasnot that of validating CCyR and MMR at 12 months as surrogatesfor overall survival, they observed no statistically significant differ-ence between first- and second-generation TKIs groups in all-causemortality rates at 12 months (relative risk [RR] 0.76; 95% CI 0.42–1.37) despite a general improvement in the CCyR rate at 12 months(RR 1.16; 95% CI 1.09–1.23) and MMR at 12 months (RR 1.68; 95% CI1.48–1.91) in patients allocated to the second-generation TKIs armas compared to patients allocated to the imatinib arm.

Strengths and Limitations

In a situation in which only evidence about clinical effectivenessmeasured in terms of cytogenetic or molecular response and

Fig. 2 – Observed vs. fitted overall survival for patients (a) with and without a CCyR (upper panel) and (b) with and without aMMR (lower panel) at 12 months. CCyR, complete cytogenetic response at 12 months; MMR, major molecular response at 12months. Observed overall survival rates derived from Table 2.

V A L U E I N H E A L T H 1 6 ( 2 0 1 3 ) 1 0 8 1 – 1 0 9 0 1087

immature data about overall survival were available, we systema-tically looked for evidence supporting the adoption of both CCyRand MMR at 12 months as reliable predictors of overall survival bylooking at TKIs-treated patients data, naive to previous pharma-cological therapies for CML. In fact, although initial marketing-authorization for imatinib was granted on the primary efficacyend point of the proportion of patients achieving major cytoge-netic response, based on the strong association between cytoge-netic response and survival observed with IFN-α, theaccompanied scientific discussion document specified that the

mechanisms of action of imatinib and IFN-α were different andthe association between survival improvement and achievementof cytogenetic response needed to be confirmed for imatinib andfollowing TKI drugs [66].

Nonetheless, we acknowledge some potential limitations in ouranalysis. First, we were able to examine the validity of the twobiomarkers as surrogate end points only on the basis of aggregatedata. Access to individual patient data has the advantage of stand-ardizing the methods of statistical analysis not only across but alsowithin studies [67]. Second, we relied on a landmark analysis, which

Fig. 3 – Overall survival for each treatment arm estimated by surrogate relationship based on CCyR and MMR separately. CCyR,complete cytogenetic response at 12 months; MMR, major molecular response at 12 months.

V A L U E I N H E A L T H 1 6 ( 2 0 1 3 ) 1 0 8 1 – 1 0 9 01088

determines each patient’s response at a fixed time point, withsurvival calculated from that time point onward and has thedisadvantage of excluding those patients who die before the land-mark from the analysis. This proportion, however, is around 1%according to DASISION and ENESTnd data [21,24]. Nevertheless, thelandmark method is one of the suggested approaches to comparingsurvival by response category [33]. Third, as in a previous study [64],we used baseline numbers of patients in responder and nonres-ponder groups to weight pooled averages of overall survival,assuming no censoring. As a further potential weakness, the modeldoes not take into account the speed of achieving the surrogateresponse, its depth, or duration [68]. This might have inducedunderestimation of long-term effectiveness of dasatinib and niloti-nib, given that they are believed to be superior to imatinib in allthese respects and considering that the historical surrogate data isall based on overall survival for patients taking imatinib. The extentof this bias is however unquantifiable. Also, and for the samereason, the model suits only those situations in which no subse-quent TKI therapy is implemented following first-line treatment.

In spite of these potential limitations, our predicted estimatesof overall survival were well aligned with observed survival fromthe DASISION [42], ENESTnd [40,41], and IRIS [43] trials. As moretrial-based data, in particular patient-level data derived from RCTs,on the relationship between cytogenetic and molecular responseand long-term outcomes, is collected and reported, it will bepossible to refine our surrogate-to-final outcome relationships.

Implications for Policy and Practice

In March 2012, NICE issued guidance recommending nilotinib andstandard-dose imatinib for the first-line treatment of Phþ CML[26]. Although the Appraisal Committee concluded that dasatiniband nilotinib provided superior clinical benefit over standard-doseimatinib, as measured by surrogate outcomes, the cost per quality-adjusted life-year (QALY) gained for nilotinib compared withstandard-dose imatinib was estimated to fall below NICE’swillingness-to-pay threshold of €20,000 to €30,000/QALY, whiledasatinib was either dominated by nilotinib (i.e., dasatinib was less

effective and more costly) or in comparison to imatinib was judgedto have a cost per QALY in excess of €200,000. Based on NICE’srecommendation, it is estimated that about 509 new patients eachyear in the United Kingdom will receive first-line therapy withimatinib and nilotinib [26].

The use of surrogate validation evidence in this case by NICEhas policy implications for cancer therapy beyond the manage-ment of CML. With pressure for faster patient access to innovativetherapies, surrogate or intermediate outcomes (such as tumorresponse, event-free survival) are increasingly being used asprimary outcomes in licensing trials of new cancer therapies.Policymakers are also facing reimbursement [28,69] decisions onthese new treatments on the basis of evidence on impact onintermediate outcomes with little or no definitive data on theimpact of therapy on overall survival. In response, health technol-ogy assessment groups and national or regional agencies respon-sible for drug coverage are beginning to introduce accessrestrictions on the basis of surrogate outcome data [70,71]. Basi-cally, new guidelines for technology appraisals state that it is nolonger sufficient for new therapies to claim effectiveness onsurrogates accepted on the basis of the biological plausibility frompathophysiological studies or the understanding of the diseaseprocess. Instead, such claims need to be grounded on “validated”surrogate outcomes, that is, outcomes that have proven associa-tion and predictive capacity to the final patient-relevant outcome.Although the highest level of evidence should come from a meta-analysis of RCTs demonstrating consistent association betweentreatment effects on the surrogate outcomes and treatment effectson the patient-relevant outcomes, the recent NICE evaluation offirst-line therapies for chronic phase CML demonstrates not onlythat evidence of surrogate outcome validation can be central to apositive listing but also that observational-level evidence maysuffice for the new drug to be included in public formularies.

Ongoing and future RCTs of TKI first-line therapies in chronicstage CML should continue to follow up patients to provide thenecessary data to examine the strength and consistency of therelationship between treatment changes in CCyR and MMR andlong-term overall survival.

V A L U E I N H E A L T H 1 6 ( 2 0 1 3 ) 1 0 8 1 – 1 0 9 0 1089

Acknowledgments

Source of financial support: Ciani Oriana is currently receiving aPeninsula College of Medicine and Dentistry PhD studentship.This work was supported by the National Institute for HealthResearch (NIHR) health technology assessment program (projectnumber 08/226/01). The authors thank Rob Anderson and ObiohaUkoumunne for their valuable contribution to the study. R.G. ispartially supported by the NIHR Collaboration for Leadership inApplied Health Research and Care (CLAHRC) for the South WestPeninsula and by the European Regional Development Fund andthe European Social Fund Convergence Programme for Cornwalland the Isles of Scilly. The views expressed in this publication arethose of the authors and not necessarily those of the NationalHealth Service, the NIHR, the Department of Health in England, orthe European Union.

Supplemental Materials

Supplemental material accompanying this article can be found inthe online version as a hyperlink at http://dx.doi.org/10.1016/j.jval.2013.07.004 or, 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] Sawyers CL. Chronic myeloid leukemia. N Engl J Med 1999;340:1330–40.[2] Baccarani M, Dreyling M. Chronic myeloid leukaemia: ESMO Clinical

Practice Guidelines for diagnosis, treatment and follow-up. Ann Oncol2010;21(Suppl. 5):v165–7.

[3] Calabretta B, Perrotti D. The biology of CML blast crisis. Blood2004;103:4010–22.

[4] Goldman JM. Chronic myeloid leukemia: reversing the chronic phase.J Clin Oncol 2010;28:363–5.

[5] Piccaluga PP, Paolini S, Bertuzzi C, et al. First-line treatment of chronicmyeloid leukemia with nilotinib: critical evaluation. J Blood Med2012;3:151–6.

[6] Huang X, Cortes J, Kantarjian H. Estimations of the increasingprevalence and plateau prevalence of chronic myeloid leukemia in theera of tyrosine kinase inhibitor therapy. Cancer 2012;118:3123–7.

[7] Haematological Malignancy Research Network. Chronic myelogenousleukaemia. Available from: http://www.hmrn.org/Statistics/Cancer_Info.aspx?id=1. [Accessed July 23, 2012].

[8] Rachet B, Maringe C, Nur U, et al. Population-based cancer survivaltrends in England and Wales up to 2007: an assessment of the NHScancer plan for England. Lancet Oncol 2009;10:351–69.

[9] Cancer, Research, UK. Leukaemia incidence statistics. 2011. Availablefrom: http://www.cancerresearchuk.org/cancer-info/cancerstats/types/leukaemia/incidence/uk-leukaemia-incidence-statistics#Prevalence.[Accessed February 26, 2013].

[10] Nowell P. A minute chromosome in human chronic granulocyticleukemia. Science 1960;132:1497.

[11] Hehlmann R, Hochhaus A, Baccarani M. Chronic myeloid leukaemia.Lancet 2007;370:342–50.

[12] Ren R. Mechanisms of BCR-ABL in the pathogenesis of chronicmyelogenous leukaemia. Nat Rev Cancer 2005;5:172–83.

[13] Druker BJ. Translation of the Philadelphia chromosome into therapy forCML. Blood 2008;112:4808–17.

[14] O’Brien SG, Guilhot F, Larson RA, et al. Imatinib compared withinterferon and low-dose cytarabine for newly diagnosed chronic-phasechronic myeloid leukemia. N Engl J Med 2003;348:994–1004.

[15] Baccarani M, Cortes J, Pane F, et al. Chronic myeloid leukemia: anupdate of concepts and management recommendations of EuropeanLeukemiaNet. J Clin Oncol 2009;27:6041–51.

[16] Hochhaus A, Kantarjian HM, Baccarani M, et al. Dasatinib inducesnotable hematologic and cytogenetic responses in chronic-phasechronic myeloid leukemia after failure of imatinib therapy. Blood2007;109:2303–9.

[17] Cortes J, Rousselot P, Kim DW, et al. Dasatinib induces completehematologic and cytogenetic responses in patients with imatinib-resistant or -intolerant chronic myeloid leukemia in blast crisis. Blood2007;109:3207–13.

[18] Hughes TP, Kaeda J, Branford S, et al. Frequency of major molecularresponses to imatinib or interferon alfa plus cytarabine in newlydiagnosed chronic myeloid leukemia. N Engl J Med 2003;349:1423–32.

[19] Capdeville R, Buchdunger E, Zimmermann J, et al. Glivec (STI571,imatinib), a rationally developed, targeted anticancer drug. Nat RevDrug Discov 2002;1:493–502.

[20] National Institute for Health and Care Excellence. Leukaemia (ChronicMyeloid)—Imatinib (TA70). 2003. Available from: http://www.nice.org.uk/ta70. [Accessed July 23, 2012].

[21] Kantarjian H, Shah NP, Hochhaus A, et al. Dasatinib versus imatinib innewly diagnosed chronic-phase chronic myeloid leukemia. N Engl JMed 2010;362:2260–70.

[22] Cortes JE, Jones D, O’Brien S, et al. Nilotinib as front-line treatment forpatients with chronic myeloid leukemia in early chronic phase. J ClinOncol 2010;28:392–7.

[23] Rosti G, Palandri F, Castagnetti F, et al. Nilotinib for the frontlinetreatment of Ph(þ) chronic myeloid leukemia. Blood 2009;114:4933–8.

[24] Saglio G, Kim DW, Issaragrisil S, et al. Nilotinib versus imatinib fornewly diagnosed chronic myeloid leukemia. N Engl J Med2010;362:2251–9.

[25] Cortes J, Kantarjian H. How I treat newly diagnosed chronic phase CML.Blood 2012;120(7):1390–7.

[26] National Institute for Health and Care Excellence. NICE recommendsnilotinib and standard dose imatinib for first line chronic myeloidleukaemia. 2012. Available from: http://www.nice.org.uk/newsroom/pressreleases/NilotinibAndImatinibForCMLDraftGuidance.jsp.[Accessed July 23, 2012].

[27] Biomarkers Definition Working Group. Biomarkers and surrogateendpoints: preferred definitions and conceptual framework. ClinPharmacol Ther 2001;69:7.

[28] National Institute for Health and Care Excellence. Leukaemia (chronicmyeloid, first line)—dasatinib, nilotinib and standard-dose imatinib:final appraisal determination. 2012. Available from: http://guidance.nice.org.uk/TA/Wave24/15/FAD. [Accessed July 23, 2012].

[29] Moher D, Liberati A, Tetzlaff J, et al. Preferred reporting items forsystematic reviews and meta-analyses: the PRISMA statement. BMJ2009;339:b2535.

[30] Dalziel K, Round A, Stein K, et al. Effectiveness and cost-effectivenessof imatinib for first-line treatment of chronic myeloid leukaemia inchronic phase: a systematic review and economic analysis. HealthTechnol Assess (Winchester, England) 2004;8(iii):1–120.

[31] Centre for Reviews and Dissemination. Undertaking SystematicReviews of Research on Effectiveness: CRD’s Guidance for Carrying Outor Commissioning Reviews (2nd ed.). CRD Report 4. York: NHS Centrefor Reviews and Dissemination, 2001.

[32] Elston J, Taylor RS. Use of surrogate outcomes in cost-effectivenessmodels: a review of United Kingdom health technology assessmentreports. Int J Technol Assess Health Care 2009;25:6–13.

[33] Anderson JR, Cain KC, Gelber RD. Analysis of survival by tumorresponse and other comparisons of time-to-event by outcomevariables. J Clin Oncol 2008;26:3913–5.

[34] Wilson E. Probable inference, the law of succession, and statisticalinference. J Am Stat Assoc 1927;22:4.

[35] Lu G, Ades AE. Assessing evidence inconsistency in mixed treatmentcomparisons. J Am Stat Assoc 2006;101:22.

[36] Spiegelhalter DTA, Best N, Lunn D. WinBUGS User Manual: Version 1.4.Cambridge: MRC Biostatistics Unit, 2003.

[37] Sutton A, Abrams K, Jones D, et al. Methods for Meta-Analysis inMedical Research. Chichester: Wiley, 2000.

[38] Pavey T, Hoyle M, Ciani O, et al. Dasatinib, nilotinib and standard-doseimatinib for the first-line treatment of chronic myeloid leukaemia:systematic reviews and economic analyses. Health Technol Assess(Winchester, England) 2012;16(iii–iv):1–277.

[39] GAD. Life tables. Available from: http://www.gad.gov.uk/Demography%20Data/Life%20Tables/. [Accessed March 13, 2011].

[40] Kantarjian HM, Hochhaus A, Saglio G, et al. Nilotinib versus imatinibfor the treatment of patients with newly diagnosed chronic phase,Philadelphia chromosome-positive, chronic myeloid leukaemia: 24-month minimum follow-up of the phase 3 randomised ENESTnd trial.Lancet Oncol 2011;12:841–51.

[41] Larson RA, Hochhaus A, Hughes TP, et al. Nilotinib vs imatinib inpatients with newly diagnosed Philadelphia chromosome-positivechronic myeloid leukemia in chronic phase: ENESTnd 3-year follow-up.Leukemia 2012;26:2197–203.

[42] Kantarjian HM, Shah NP, Cortes JE, et al. Dasatinib or imatinib innewly diagnosed chronic-phase chronic myeloid leukemia: 2-yearfollow-up from a randomized phase 3 trial (DASISION). Blood2012;119:1123–9.

V A L U E I N H E A L T H 1 6 ( 2 0 1 3 ) 1 0 8 1 – 1 0 9 01090

[43] Hughes TP, Hochhaus A, Branford S, et al. Long-term prognosticsignificance of early molecular response to imatinib in newlydiagnosed chronic myeloid leukemia: an analysis from theInternational Randomized Study of Interferon and STI571 (IRIS). Blood2010;116:3758–65.

[44] Roy L, Guilhot J, Krahnke T, et al. Survival advantage from imatinibcompared with the combination interferon-alpha plus cytarabine inchronic-phase chronic myelogenous leukemia: historical comparisonbetween two phase 3 trials. Blood 2006;108:1478–84.

[45] Jabbour E, Kantarjian H, O’Brien S, et al. The achievement of an earlycomplete cytogenetic response is a major determinant for outcome inpatients with early chronic phase chronic myeloid leukemia treatedwith tyrosine kinase inhibitors. Blood 2011;118:4541–6: (quiz 759).

[46] de Lavallade H, Apperley JF, Khorashad JS, et al. Imatinib for newlydiagnosed patients with chronic myeloid leukemia: incidence ofsustained responses in an intention-to-treat analysis. J Clin Oncol2008;26:3358–63.

[47] Kantarjian H, O’Brien S, Shan J, et al. Cytogenetic and molecularresponses and outcome in chronic myelogenous leukemia: need fornew response definitions? Cancer 2008;112:837–45.

[48] Hehlmann R, Lauseker M, Jung-Munkwitz S, et al. Tolerability-adaptedimatinib 800 mg/d versus 400 mg/d versus 400 mg/d plus interferon-alpha in newly diagnosed chronic myeloid leukemia. J Clin Oncol2011;29(12):1634–42.

[49] Rajappa S, Varadpande L, Paul T, et al. Imatinib mesylate in earlychronic phase chronic myeloid leukemia: experience from a developingcountry. Leukemia Lymphoma 2008;49:554–8.

[50] Hochhaus A, O’Brien SG, Guilhot F, et al. Six-year follow-up of patientsreceiving imatinib for the first-line treatment of chronic myeloidleukemia. Leukemia 2009;23:1054–61.

[51] Deininger MW, Goldman JM, Melo JV. The molecular biology of chronicmyeloid leukemia. Blood 2000;96:3343–56.

[52] Chakraborty S, Stark JM, Sun CL, et al. Chronic myelogenous leukemiastem and progenitor cells demonstrate chromosomal instability relatedto repeated breakage-fusion-bridge cycles mediated by increasednonhomologous end joining. Blood 2012;119:6187–97.

[53] Carroll M. BCR/ABL and chromosomal instability: debate resolved.Blood 2012;119:6180–1.

[54] Futosi K, Nemeth T, Pick R, et al. Dasatinib inhibits proinflammatoryfunctions of mature human neutrophils. Blood 2012;119:4981–91.

[55] Zarbock A. The shady side of dasatinib. Blood 2012;119:4817–8.[56] Bucher HC, Guyatt GH, Cook DJ, et al. Users’ guides to the medical

literature, XIX: applying clinical trial results. A. How to use an articlemeasuring the effect of an intervention on surrogate end points.Evidence-Based Medicine Working Group. JAMA 1999;282:771–8.

[57] Lassere MN. The Biomarker-Surrogacy Evaluation Schema: a review ofthe biomarker-surrogate literature and a proposal for a criterion-based,quantitative, multidimensional hierarchical levels of evidence schemafor evaluating the status of biomarkers as surrogate endpoints. StatMethods Med Res 2008;17:303–40.

[58] Burzykowski T, Molenberghs G, Buyse M. The Evaluation of SurrogateEndpoints. US: Springer Science þ Business Media, Inc., 2005.

[59] Anstrom KJ, Reed SD, Allen AS, et al. Long-term survival estimates forimatinib versus interferon-alpha plus low-dose cytarabine for patientswith newly diagnosed chronic-phase chronic myeloid leukemia.Cancer 2004;101:2584–92.

[60] Baccarani M, Rosti G, de Vivo A, et al. A randomized study ofinterferon-alpha versus interferon-alpha and low-dose arabinosylcytosine in chronic myeloid leukemia. Blood 2002;99:1527–35.

[61] Bonifazi F, de Vivo A, Rosti G, et al. Chronic myeloid leukemia andinterferon-alpha: a study of complete cytogenetic responders. Blood2001;98:3074–81.

[62] Guilhot F, Chastang C, Michallet M, et al. Interferon alfa-2b combinedwith cytarabine versus interferon alone in chronic myelogenousleukemia. French Chronic Myeloid Leukemia Study Group. N Engl J Med1997;337:223–9.

[63] Kantarjian HM, O’Brien S, Cortes JE, et al. Complete cytogenetic andmolecular responses to interferon-alpha-based therapy for chronicmyelogenous leukemia are associated with excellent long-termprognosis. Cancer 2003;97:1033–41.

[64] Schrover RJ, Adena MA, De Abreu Lourenco R, et al. Development of apredictive population survival model according to the cytogeneticresponse rate for patients with chronic myeloid leukemia in thechronic phase. Leukemia Lymphoma 2006;47:1069–81.

[65] Gurion R, Gafter-Gvili A, Vidal L, et al. Has the time for first-linetreatment with second generation tyrosinekinase inhibitors in patientswith chronic myelogenous leukemia already come? Systematic reviewand meta-analysis. Haematologica 2013;98(1):95–102.

[66] European Medicines Agency. Glivec: EPAR—scientific discussion. 2005.Available from: http://www.ema.europa.eu/ema/index.jsp?curl=pages/medicines/human/medicines/000406/human_med_000808.jsp&mid=WC0b01ac058001d124. [Accessed July 23, 2012].

[67] Riley RD, Lambert PC, Abo-Zaid G. Meta-analysis of individualparticipant data: rationale, conduct, and reporting. BMJ 2010;340:c221.

[68] Cortes JE. Not only response but early response to tyrosine kinaseinhibitors in chronic myeloid leukemia. J Clin Oncol 2012;30:223–4.

[69] Clement FM, Harris A, Li JJ, et al. Using effectiveness and cost-effectiveness to make drug coverage decisions: a comparison of Britain,Australia, and Canada. JAMA 2009;302:1437–43.

[70] Institute for Quality and Efficiency in Healthcare. Validity of surrogateendpoints in oncology. 2011. Available from: https://www.iqwig.de/a10-05-validity-of-surrogate-endpoints-in.986.en.html?tid=1325&random=f6799f. [Accessed July 23, 2012].

[71] National Institute for Health and Care Excellence. Review of the Guideto the Methods of Technology Appraisal. 2012. Available from: http://www.nice.org.uk/aboutnice/howwework/devnicetech/TAMethodsGuideReview.jsp?domedia=1&mid=CB13DD0D-19B9-E0B5-D4AB0011AEE2B0E7.[Accessed July 23, 2012].