Thalidomide versus bortezomib based regimens as first-line therapyfor patients with multiple myeloma: a systematic reviewAmbuj Kumar,1,2* Iztok Hozo,3 Keath Wheatley,4 and Benjamin Djulbegovic1,2,5

Thalidomide (T) or bortezomib (B) in combination with melphalan plus prednisone (MP) is superior to MP asfirst line therapy for previously untreated myeloma. However, direct head-to-head comparison of Melphalan,Prednisone plus Bortezomib (MPB) versus Melphalan, Prednisone plus Thalidomide (MPT) is lacking. We per-formed an indirect meta-analysis to assess the treatment effects of MPB versus MPT via common comparatorMP using the systematic review and meta-analytical techniques. A comprehensive literature search (MEDLINEand gray literature) was undertaken. Systematic review was performed as per the Cochrane Collaboration rec-ommendations. Initial search yielded 1,013 citations, of which six randomized controlled trials (RCTs) enroll-ing 2,798 patients met the inclusion criteria. Comparison of MPT versus MP (five RCTs) showed no survivaldifference [hazard ratio (HR) 0.82, 95% confidence interval (CI) 0.64–1.05] but a statistically significant differ-ence in event-free survival favoring MPT (HR 0.66, 95% CI 0.56–0.77) without excessive treatment-related mor-tality [risk ratio (RR) 1.11, 95% CI 0.64–1.92]. Comparison of MPB vs. MP (one RCT) showed a statistically sig-nificant benefit for survival (HR 0.65, 95% CI 0.51–0.84) and event-free survival (HR 0.48, 95% CI 0.37–0.63)without difference in treatment-related mortality (RR 0.42, 95% CI 0.11–1.63) with MPB. The indirect compari-son of MPB versus MPT showed no difference between MPB versus MPT for all outcomes but a significantbenefit for complete response (RR 2.34, 95% CI 1.12–4.90), and grade III/IV adverse events (RR 0.53, 95% CI0.38–0.73) favoring MPB. There is an uncertainty about definitive superiority of one type of regimen over theother. Therefore, direct head-to-head comparison between these competing regimens is warranted. Am. J.Hematol. 86:18–24, 2011. VVC 2010 Wiley-Liss, Inc.

BACKGROUNDTherapy with melphalan plus prednisone (MP) has been

the mainstay of treatment for patients with newly diagnosedmyeloma who are not eligible for transplant for the last 4decades [1]. Recently, thalidomide (T) and bortezomib (B)has been added to the combination of MP to further improveoutcomes for myeloma patients ineligible for transplant [2–4]. The efficacy of adding thalidomide to MP (MPT) versusMP alone has been tested in several randomized controlledtrials (RCT) [2,4–7]. The results are encouraging but conflict-ing at the same time. Although some RCTs showed a sur-vival advantage with MPT versus MP [5,6], others did not[2,4,7]. Addition of bortezomib to the combination of MP(MPB) has resulted in improved outcomes of survival andevent-free survival (EFS) [3]. Introduction of MPT and MPBfor the treatment of multiple myeloma has also created a di-lemma for the treating physician and patients alike on theoptimal choice of treatment who would like to know if MPT isbetter than MPB or vice versa. However, a RCT comparingthese MPT versus MPB has not been performed. When adirect comparison between treatments is not available, oneway to assess the effects of competing treatments is to per-form an indirect comparison. To provide an unbiased esti-mates of treatment effects, such an indirect comparisonshould preserve the original randomization [8–11], so thatcomparison remain within each RCT and do not involve anyanalyses in which patients in one trial are compared withpatients in another trial. Therefore, to address the questionrelated to the treatment effects of MPT and MPB versus MPfor the treatment of multiple myeloma, we performed a sys-tematic review and meta-analysis of all RCTs. To assess thetreatment effects of MPT versus MPB we also performed anindirect meta-analysis via common comparator MP.

METHODSTrials and patients. Only phase III RCTs testing MPT or MPB ver-

sus MP were considered for the analysis. We used the comprehensivesearch strategies described by Haynes et al. [12] to identify all relevantRCTs published through December 2009 in the Medline (PubMed). We

also performed manual searches of abstracts from the annual meetingsof the American Society of Hematology (2007–2009), and American So-ciety for Clinical Oncology (2007–2009) to identify potential RCTs. Inaddition, experts in the field were contacted to identify any unpublisheddata in this subject area. No search limits were applied on the basis oflanguage.

Selection of trials. Studies were included if they were prospectivephase III RCTs involving assessment of MPT or MPB versus MPfor the treatment of patients with multiple myeloma and reported ex-tractable data on benefits and risks according to the intention-to-treatprinciple.

Data extraction. Two reviewers (A.K. and B.D.) independentlyscreened the titles and abstracts of all identified studies to assess theireligibility for inclusion. We extracted data on benefits in terms of overallsurvival (OS), event-free survival (EFS), complete response (CR), verygood partial response (VGPR), and partial response (PR). We alsoabstracted data on harms associated with treatments on following out-comes: treatment related mortality (TRM), any grade III/IV adverseevent, and deep vein thrombosis (DVT). We also extracted data on themethodological domains relevant to minimizing bias and random error(generation of allocation sequence, description of drop-outs, and analy-sis on an intention-to-treat basis) in the conduct and analysis of the tri-als [13,14]. We also extracted data on most important features of trialto judge if the trials were similar with respect to patients, intervention,control, and outcome to enable us to conduct an unbiased direct andindirect comparison. There were no discrepancies in data extractionbetween the reviewers.

1Center for Evidence-based Medicine & Health Outcome Research, Univer-sity of South Florida; 2Department of Health Outcomes and Behavior, Mof-fitt Cancer Center & Research Institute, Tampa, Florida; 3Department ofMathematics, Indiana University, Gary, Indiana; 4University of Birmingham,Birmingham, United Kingdom; 5Department of Hematology, Moffitt CancerCenter & Research Institute, Tampa, Florida

*Correspondence to: Ambuj Kumar, MD, MPH, 12901 Bruce B. Downs Bou-levard, MDC 27, Tampa, FL 33612. E-mail: akumar1@health.usf.edu

Accepted 5 October 2010

Am. J. Hematol. 86:18–24, 2011.

Published online 19 October 2010 in Wiley Online Library (wileyonlinelibrary.com).DOI: 10.1002/ajh.21904

Research Article

VVC 2010 Wiley-Liss, Inc.

American Journal of Hematology 18 http://wileyonlinelibrary.com/cgi-bin/jhome/35105

Statistical methods. For each included RCT, for the purpose ofanalysis, we calculated the logarithm of the hazard ratio (HR) or riskratio (RR), as applicable, and its standard error to perform meta-analysis. It is important to note, that indirect meta-analysis preservesthe randomization within a RCT while at the same time, combines allavailable comparisons between treatments [8,9]. These comparisonsincluded both the direct within trial comparisons between two treat-ment strategies and the indirect comparisons constructed from trialsthat have 1 treatment in common.

When more than one RCT was available for comparison (e.g.,MPT versus MP), we first calculated the pooled estimates usingstandard meta-analytic techniques for that comparison [15]. Usingsimilar meta-analytic techniques, we obtained a pooled estimate fromRCTs that compared other interventions (e.g., MPB versus MP).Because both comparisons used MP as control, the summary esti-mates obtained from the respective meta-analysis (MPT versus MPand MPB versus MP) can be used to provide estimates of the HR orRR for the indirect comparison of MPT versus MPB. The adjustedindirect comparisons were performed using the method described byBucher et al. [8] and Glenny et al. [9] that were extended to calcu-late HR.

According to this method, a less biased indirect comparison of inter-ventions MPT versus MPB can be obtained by adjusting the results oftheir direct comparisons with a common intervention of MP. If weassume that MPTMA is the estimate of direct comparison betweenintervention MPT versus MP and MPBMA is the direct comparison ofintervention MPB versus MP, then the estimate of the adjusted indirectcomparison of intervention MPT versus MPB (MPBTad_indirect) (such aslog HR or log RR etc.) is estimated by MPBTad_indirect 5 MPTMA 2MPBMA. Because the estimates are obtained from different studies,the results are statistically independent and its variance can beobtained by Var(log(MPTBad_indirect) 5 Var(log(MPTMA) 1 Var(log(MPBMA) [8,9,16].

Two hypotheses tested were (a) is MPT or MPB superior to MPfor the treatment of multiple myeloma, and (b) do effects of treat-ments obtained in MPT trials differ from MPB trials? An adjustedindirect comparison at conventional difference of P < 0.05 levelsusing the random effects model was calculated [8,9,16]. Time-to-event data (e.g., OS, EFS), were pooled and reported as HR whiledichotomous data (CR, DVT, TRM, etc.) were expressed as RR,respectively, using a 95% confidence interval (CI), under a randomeffects model [17]. All data of interest were obtained either directlyfrom the published articles or supplemented by the information fromthe investigators.

A formal statistical test for heterogeneity using the chi square test[17] and I2 [18] was performed. The possibility of publication bias wasalso assessed using the Begg and Egger funnel plot method [19,20].The meta-analysis was performed using STATA software (STATA Corp,College Station, TX) and Revman [21], which was used to generatecorresponding forest plots. The work was performed and reportedaccording to the PRISMA guidelines for reporting of systematicreviews [22].

RESULTS

TrialsAs displayed in Figure 1, a comprehensive search of

Pubmed, meeting abstracts of American Society for ClinicalOncology, and American Society for Hematology yielded1,013 articles, of which six studies met the predeterminedinclusion criteria. Five RCTs tested MPT against MP(2–5,7) and one MPB versus MP [6]. These six trials enrolled atotal of 2,798 patients. All RCTs were reported as full publi-cations [2,3,5,6]. [4,7]. Study characteristics and the meth-odological quality of included RCTs have been summarizedin Table I. Details on regimen doses and cycles have beendepicted in Table II. The Begg and Egger funnel plot(not shown) for the outcomes of OS (P 5 0.447) and EFS(P 5 0.990) showed a symmetric distribution indicating nopublication bias.Data on OS, EFS, CR, VGPR, PR, and DVT were ex-

tractable from all RCTs. Data on TRM and grade III/IVadverse events were extractable from five [2,3,5–7] out ofsix RCTs. Figure 2 shows the network of comparisonsemployed in our analysis. The summary results for

all comparisons are shown in Table III and depicted inFigures 3–6.

MPT versus MPBenefits Five RCTs enrolling a total of 1,571 patients

compared the treatment effects of MPT versus MP [2,4–7].Pooled results showed a statistically non-significant differ-ence with MPT use versus MP for OS. The combined HRfor OS was 0.82 (95% CI 0.64–1.05; P 5 0.12). There wasa statistically significant difference for the outcome of EFS(HR 0.66; 95% CI 0.56–0.77; P < 0.00001), CR (RR 3.58,95% CI 2.20–5.81; P < 0.00001), VGPR (RR 3.61, 95% CI2.52–5.18; P < 0.00001), and PR (RR 1.47, 95% CI 1.09–1.98; P 5 0.01) favoring MPT versus MP.There was a statistically significant heterogeneity among

included trials for the outcomes of OS (I2 72%, P 5 0.006),and PR (I2 85%, P < 0.0001).Harms There was a statistically nonsignificant difference

in TRM (RR 1.11, 95% CI 0.64–1.92; P 5 0.71) betweenMPT and MP. Use of MPT was associated with a statisti-cally significant risk for DVT (RR 2.41, 95% CI 1.05–5.53;P 5 0.04), and any grade III/IV adverse events (RR 2.44,95% CI 1.84–3.22; P < 0.00001) compared with MP.There was a statistically significant heterogeneity among

included trials for the outcomes of DVT (I2 61%, P 5 0.04).

Sensitivity analysesTo assess the robustness of the effect of thalidomide

addition to the MP regimen and to explain the heterogene-ity observed for the outcome of OS additional sensitivityanalyses were performed. As shown in Table III overalltrials were similar with respect to distribution of patientcharacteristics according to age, disease stage, and meth-odological quality of RCTs. However, three trials providedthalidomide as maintenance treatment [2,4,23] and two didnot, [5,6] and accordingly, we performed a subgroup analy-ses according to provision of maintenance thalidomide (yesor no). The pooled HR for OS in trials using maintenancethalidomide was 0.96 (95% CI 0.75 to 1.24; P 5 0.76) ver-sus 0.63 in trials without maintenance thalidomide (95% CI0.51–0.78; P < 0.0001) indicating no effect of maintenancethalidomide on OS. Additionally, there was a statisticallynonsignificant heterogeneity among pooled studies for theoutcome of OS within subgroups (for thalidomide mainte-nance group I2 was 30.6%, and P 5 0.23; for the no main-tenance thalidomide group I2 was 0% and P 5 0.5). Therewas a statistically significant heterogeneity between sub-groups indicating that trials with maintenance thalidomidewere distinct from trials without maintenance thalidomide.

Figure 1. Flow diagram depicting the selection process.

American Journal of Hematology 19

research article

TABLEI.

StudyCharacteristicsandtheMethodologicalQuality

ofIncludedRandomizedControlledTrials

Medianage(range)

Distributionofpatients

accordingto

ISS

Methodologicalquality

Author(Yearof

publication)

Numberof

patients

ExperimentalRx

Standard

Rx

ExperimentalRx

N(%

)Standard

Rx

N(%

)Primary

end-point

Allocation

concealm

ent

Blinding

Apriorisample

size

calculationsperform

ed

Descriptionof

withdrawls/drop-outs

Intention-to-treat

analysis

MRTvs.

MP

Hulin

etal.(2009)

229

78.5

(75–89)

125(25%)

26(25%)

Ove

rallsurvival

Adequate

No

Yes

Yes

Yes

239(40%)

47(45%)

334(35%)

31(30%)

Palumboetal.(2008)

331

72(N

R)

72(N

R)

132(23%)

32(25%)

Responserates

andProgression-

freesurvival

Adequate

No

Yes

Yes

Yes

266(48%)

59(46%)

338(29%)

37(29%)

Wijerm

ansetal.(2010)

333

72(65–87)

73(65–84)

144(27%)

39(23%)

Eve

ntfree

survival

Adequate

No

Yes

Yes

Yes

242(25%)

39(23%)

332(19%)

29(17%)

Unknown

47(28%)

61(36%)

Waageetal.(2010)*

357

74.6

(65–79)

74.1

(65–79)

123(13%)

31(18%)

Ove

rall

survival

Adequate

Patient,

inve

stigator

andmonitors

Yes

Yes

Yes

269(37%)

75(43%)

366(36%)

52(30%)

Unknown

24(14%)

16(9%)

Faconetal.(2007)

447

NR

NR

138(34%)

61(34%)

Ove

rall

survival

Adequate

No

Yes

Yes

Yes

242(38%)

67(37%)

332(29%)

54(30%)

MPBvs.

MP

Mateosetal.(2010)

682

71(57–90)

71(48–91)

165(19%)

64(19%)

Tim

eto

disease

progression

Adequate

No

Yes

Yes

Yes

2162(47%)

159(47%)

3117(34%)

115(34%)

*Meanagesare

reportedinsteadofmedianforthis

trial.

TABLEII.CharacteristicsofDosageandRegim

ensAcrossIncludedTrials

Author(Year

ofpublication)

MP

Regim

en

MP-T

Cycle

length

inweeks

Doseof

maintenance

thalidomide

Melphalan

Prednisone

Melphalan

Prednisone

Thalidomide

Hulin

etal.(2009)

0.2

mg/kg

2mg/kg

Days1–4

0.2

mg/kg

2mg/kg

100mg/daycontinuously

for72weeks

60

Palumboetal.(2008)

4mg/m

240mg/m

2Days1–7

4mg/m

240mg/m

2100mg/daycontinuously

during6cycles

4100mg/day

Wijerm

ansetal.(2010)

0.25mg/kg

1mg/kg

Daily

for5dayseve

ry4weeks

0.25mg/kg

1mg/kg

200mg/daycontinuously

until4weeksafterlastMPcycle

450mg/day

Waageetal.(2010)

0.25mg/kg

100mg/day

Daily

for4dayseve

ry6weeks

0.25mg/kg

100mg/day

200mg/dayfor1week,then400mgdaily

6200mg/day

Faconetal.(2007)

0.25mg/kg

2mg/kg

4Dayspercycle

0.25mg/kg

2mg/kg

�400mg/daycontinuously

duringthe12MPcycles

60

MP

MP-B

Melphanlan

Prednisone

Melphalan

Prednisone

Bortezo

mib

Doseofmaintenancebortezo

mib

Mateosetal.(2010)

9mg/m

260mg/m

2Days1–4

9mg/m

260mg/m

21.3

mg/m

2(days1,4,8,11,22,25,29,32,cycles1–4)

61.3

mg/m

2(days1,8,22,29,cycles5-9)

20 American Journal of Hematology

research article

MPB versus MPBenefits OneRCTenrollinga totalof682patientscompared

the treatmenteffectsofMPBversusMP[3].Theresultsshowedastatistically significantbenefitwithMPBuseversusMPforOS(HR0.65, 95%CI 0.51–0.84;P5 0.0008), EFS (HR0.48, 95%CI 0.37–0.63;P< 0.00001), CR (RR8.39, 95%CI 4.82–14.60;P < 0.00001), and VGPR (RR 2.12, 95% CI 1.12–4.01; P 50.02). However, there was a statistically nonsignificant differ-ence between MPB and MP for PR. The RR for PR was 1.08,95%CI0.86–1.35;P50.50).Harms There was a statistically non-significant difference

in TRM (RR 0.42, 95% CI 0.11–1.63; P 5 0.21), and DVT(RR 0.66, 95% CI 0.19–2.32; P 5 0.52) associated withMPB use compared with MP. However, MPB use was asso-ciated with a significantly worse any grade III/IV adverse

events (RR 1.28, 95% CI 1.06–1.54; P 5 0.009) comparedwith MP.

MPB versus MPT (Indirect comparison)Benefits An indirect comparison showed a statistically

non-significant difference between MPB versus MPT for theoutcome of OS (HR 0.80, 95% CI 0.56–1.14; P 5 0.21),EFS (HR 0.73, 95% CI 0.54–1.00, P 5 0.053), VGPR (RR0.59, 95% CI 0.28–1.23; P 5 0.167), and PR (RR 0.852,95% CI 0.598–1.215; P 5 0.377). There was a statisticallysignificant benefit with MPB for the outcome of CR (RR2.34, 95% CI 1.12–4.90; P 5 0.02).Harms There was a statistically nonsignificant difference

between MPB and MPT for the outcomes of TRM (RR0.38, 95% CI 0.09–1.631; P 5 0.194) and DVT (RR 0.27,95% CI 0.06–1.23; P 5 0.092). Compared with MPB, MPTwas associated with a significantly higher risk for anygrade III/IV adverse events (RR 0.53, 95% CI 0.38–0.73;P < 0.0001).

DISCUSSIONThe results of systematic review and meta-analysis

show that MPT is superior to MP in improving EFS,CR,VGPR, and PR, and use of MPT was not associatedwith increased TRM compared with MP. However, use ofMPT was not associated with a significantly superior sur-vival and was associated with significantly higher rates ofDVT and grade III/IV adverse events. Treatment with MPBwas associated with a significant improvement in OS,EFS, CR, and VGPR without a significant increase inTRM or DVT compared with MP. However, use of MPBwas associated with increased risk for other grade III/IVadverse events. The indirect comparison between MPBversus MPT showed that except for grade III/IV adverseevents and CR (both favored MPB versus MPT), there isno difference between two treatments. However, theincreased toxicity with MPT compared with MPB shouldbe interpreted with caution. Because, the higher risk forany grade III/IV toxicity with MPT may be driven by a sig-nificantly higher risk DVT. Additionally, MPB and MPT regi-mens are evolving into less toxic treatment schedules dueto DVT prophylaxis.Both, MPT and MPB regimens are considered important

breakthroughs in the management of myeloma and arecredited for changing natural history of this, once dismal,disease [24]. However, because of lack of direct compari-sons between these regimens, physicians are currently notsure what treatment they should choose when they seepatients with newly diagnosed multiple myeloma patientswho are ineligible for transplant. This has also beenreflected in the guidelines recommendations. For example,influential National Comprehensive Cancer Network guide-lines [25] recommend each of these treatment options asequally acceptable alternatives. As a result, it is not surpris-ing to witness uncertainty in the contemporary practiceamong patients and physicians alike with the consequenceof huge variation seen in practice. But, treatment choiceshave to be made. What should physicians (and guidelinesdevelopers) do in the situations like this? Common practiceis to perform direct comparisons of treatments based onnonrandomized trials. However, such a practice is discour-aged because it is invariably biased [9]. When nonrandom-ized comparisons are made, it is simply impossible toreduce selection biases and disentangle the effect of treat-ment from other prognostic variables on outcomes [9]. Theonly way to avoid these biases is to preserve the originalrandomization [8,9]. This is exactly what adjusted indirectcomparison does: it preserves the original randomization,so patients in one trial are not directly compared with thosein another trial [8,9]. All trials are analyzed separately with

Figure 2. Network of comparisons. [Color figure can be viewed in the onlineissue, which is available at wileyonlinelibrary.com.]

Figure 3. Meta-analysis of randomized controlled trials on the treatment effectsof MPT and MPB versus MP on OS (A) and event free survival (B). The pooledsummary effect estimate (HR/RR) for each outcome is indicated by black rectan-gles, with the lines representing 95% confidence intervals (CIs). The vertical lineindicates no difference between two treatments. [Color figure can be viewed in theonline issue, which is available at wileyonlinelibrary.com.]

American Journal of Hematology 21

research article

TABLE III. Summary Results for All Comparisons and Outcomes

Outcome

MPT versusMP (5 RCTs;N 5 1571)

Presence ofheterogeneityamong trials Conclusion

MPB versusMP (1 RCT;N 5 687) Conclusion

MPB versusMPT (IndirectComparison) Conclusion

Overall SurvivalHazard ratio (HR) HR 5 0.828 Yes No difference HR 5 0.65 MPB better HR 5 0.80 No difference(95% Confidence Intervals) (0.64, 1.05) I2 5 72% (0.51, 0.84) (0.56, 1.14)P-value P 5 0.12 P 5 0.006 P 5 0.0008 P 5 0.162

Event-free survivalHazard ratio (HR) HR 5 0.66 No MPT better HR 5 0.48 MPB better HR 5 0.73 No difference(95% Confidence Intervals) (0.56, 0.77) I2 5 51% (0.37, 0.63) (0.54, 1.00)P-value P < 0.00001 P 5 0.09 P < 0.00001 P 5 0.055

Treatment-related mortalityRisk ratio (RR) RR 5 1.11 No No difference RR 5 0.42 No difference RR 5 0.38 No difference(95% Confidence Intervals) (0.64, 1.92) I2 5 32% (0.11, 1.63) (0.09, 1.61)P-value P 5 0.71 P 5 0.22 P 5 0.21 P 5 0.194

Deep vein thrombosisRisk ratio (RR) RR 5 2.41 Yes MP better RR 5 0.66 No difference RR 5 0.27 No difference(95% Confidence Intervals) (1.05, 5.53) I2 5 61% (0.19, 2.32) (0.06, 1.23)P-value P 5 0.04 P 5 0.04 P 5 0.52 P 5 0.092

Any grade III-IV adverse eventsRisk ratio (RR) RR 5 2.44 No MP better RR 5 1.28 MP better RR 5 0.53 MPB better(95% Confidence Intervals) (1.84, 3.22) I2 5 56% (1.06, 1.54) (0.38, 0.73)P-value P < 0.00001 P 5 0.08 P 5 0.009 P < 0.0001

Complete responseRisk ratio (RR) RR 5 3.58 No MPT better RR 5 8.39 MPB better RR 5 2.34 MPB better(95% Confidence Intervals) (2.20, 5.81) I2 5 0% (4.82, 14.60) (1.12, 4.90)P-value P < 0.00001 P 5 0.46 P < 0.00001 P 5 0.019

Very good partial responseRisk ratio (RR) RR 5 3.61 No MPT better RR 5 2.12 MPB better RR 5 0.59 No difference(95% Confidence Intervals) (2.52, 5.18) I2 5 36% (1.12, 4.01) (0.28, 1.23)P-value P < 0.00001 P 5 0.18 P 5 0.02 P 5 0.167

Partial responseRisk ratio (RR) RR 5 1.47 Yes MPT better RR 5 1.08 No difference RR 5 0.38 No difference(95% Confidence Intervals) (1.09, 1.98) I2 5 85% (0.86, 1.35) (0.09, 1.07)P-value P 5 0.01 P < 0.0001 P 5 0.05 P 5 0.377

MPT, melphalan, prednisone, thalidomide; MPB, melphalan, prednisone, bortezomib; MP, melphalan and prednisone.

Figure 4. Meta-analysis of randomized controlled trials on the treatment effects of MPT and MPB versus MP on CR (A), PR (B), and VGPR (C). The pooled summaryeffect estimate (HR/RR) for each outcome is indicated by black rectangles, with the lines representing 95% confidence intervals (CIs). The vertical line indicates no differ-ence between two treatments. [Color figure can be viewed in the online issue, which is available at wileyonlinelibrary.com.]

22 American Journal of Hematology

research article

summary statistics calculated for each trial; these summarystatistics of the same interventions are added together inthe standard meta-analysis and then the effects betweendifferent interventions are estimated by adjusting the resultsof their direct comparisons with a common intervention asdescribed earlier (see Statistical Section). Thereby, influen-ces of potential prognostic imbalances on any particulartrial are avoided and the results can be consideredunbiased. It is for this reason that an indirect comparison isconsidered superior to direct comparison from non-random-ized studies [9].In light of this discussion, does this mean that we should

recommend MPB over MPT since MPB lead to superiorCR over MPT and was also associated with fewer adverseevents than MPT? We believe that such a conclusion wouldbe premature. The main reason for this is that the existingbody of evidence is relatively insufficient due to limitation ofthe power of our analysis. This is the main limitation of indi-rect meta-analysis. It has been estimated that four times as

many similarly sized trials are needed for the indirectapproach to have the same power as directly randomizedcomparisons [9]. In addition, it is always controversial tomake comparisons where supporting evidence is derivedfrom one trial only, as in case of MBT comparator. There-fore, it is not surprising that our results were accompaniedwith the wide confidence intervals. For example, confidenceintervals for survival HR for comparison between MPB andMPT ranged from 0.56 to 1.14 (Fig. 6). This means that theresults are consistent with MPB improving mortality by 44%as well as MPT improving survival by 14%. The accompa-nied ‘‘non-significant’’ P 5 0.162 should not be interpretedas the effects of these two treatment are no different.Rather this is a classic example illustrating that ‘‘absenceof evidence is not the same as evidence of absence’’ i.e.,evidence of no difference between these two regimensshould not be construed as if their effects is identical [26].In summary, the main value of this analysis is to identify

the current gaps in knowledge, articulate the remaining

Figure 5. Meta-analysis of randomized controlled trials on the treatment effects of MPT and MPB versus MP on treatment-related mortality (A), any grade III/IVadverse events (B), and DVT (C). The pooled summary effect estimate (HR/RR) for each outcome is indicated by black rectangles, with the lines representing 95%confidence intervals (CIs). The vertical line indicates no difference between two treatments. [Color figure can be viewed in the online issue, which is available atwileyonlinelibrary.com.]

Figure 6. Summary results (individual studies not shown) from indirect comparison of MPB versus MPT using the results of meta-analysis. The pooled summary effectestimate (HR/RR) for each outcome is indicated by black rectangles, with the lines representing 95% confidence intervals (CIs). The vertical line indicates no differencebetween two treatments. [Color figure can be viewed in the online issue, which is available at wileyonlinelibrary.com.]

American Journal of Hematology 23

research article

uncertainties, and help prioritize future research for this dis-ease. We believe that our analysis clearly identified the eq-uipoise [27], and the direct head-to-head trial comparingMPB versus MPT regimens is clearly warranted. To theextent that one believes that all derivatives of thalidomidesuch as lenolidomide (L) have the same effects with thali-domide, the same recommendation applies to comparisonbetween MBP and MPL.

Author ContributionsConception and design: Benjamin Djulbegovic and Ambuj

Kumar; Collection and assembly of data: Ambuj Kumar andBenjamin Djulbegovic; Data analysis and interpretation:Ambuj Kumar, Benjamin Djulbegovic, Keath Wheatley andIztok Hozo; Manuscript writing: Ambuj Kumar, BenjaminDjulbegovic, Keath Wheatley and Iztok Hozo; Final approvalof manuscript: Ambuj Kumar.

REFERENCES1. Alexanian R, Haut A, Khan AU. Treatment for multiple myeloma combination

chemotherapy with different melphalan dose regimens. JAMA 1969;208:1680–1685.

2. Palumbo A, Bringhen S, Liberati AM, et al. Oral melphalan, prednisone,and thalidomide in elderly patients with multiple myeloma: Updated resultsof a randomized controlled trial. Blood 2008;112:3107–3114.

3. Mateos M-V, Richardson PG, Schlag R, et al. Bortezomib plus melphalanand prednisone compared with melphalan and prednisone in previouslyuntreated multiple myeloma: updated follow-up and impact of subsequenttherapy in the phase III VISTA trial. J Clin Oncol 2010;28:2259–2266.

4. Waage A, Gimsing P, Fayers P, et al. Melphalan and prednisone plus thalido-mide or placebo in elderly patients with multiple myeloma. Blood 2010;116:1405–1412.

5. Facon T, Mary JY, Hulin C, et al. Melphalan and prednisone plus thalidomideversus melphalan and prednisone alone or reduced-intensity autologous stemcell transplantation in elderly patients with multiple myeloma (IFM 99–06): arandomised trial. Lancet 2007;370:1209–1218.

6. Hulin C, Facon T, Rodon P, et al. Efficacy of melphalan and prednisone plusthalidomide in patients older than 75 years with newly diagnosed multiple my-eloma: IFM 01/01 trial. J Clin Oncol 2009;27:3664–3670.

7. Wijermans P, Schaafsma M, van Norden Y, et al. Melphalan 1prednisone versus melphalan 1 prednisone 1 thalidomide in induction ther-apy for multiple myeloma in elderly patients: final analysis of the dutch coop-erative group HOVON 49 study. Blood (ASH Annual Meeting Abstracts)2008;112:649.

8. Bucher HC, Guyatt GH, Griffith LE, Walter SD. The results of direct and indi-rect treatment comparisons in meta-analysis of randomized controlled trials.J Clin Epidemiol. 1997;50:683–691.

9. Glenny AM, Altman DG, Song F, et al. Indirect comparisons of competinginterventions. Health Technol Assess 2005;9:1–134, iii–iv.

10. Song F, Glenny AM, Altman DG. Indirect comparison in evaluating relative ef-ficacy illustrated by antimicrobial prophylaxis in colorectal surgery. ControlClin Trials 2000;21:488–497.

11. Song F, Altman DG, Glenny AM, Deeks JJ. Validity of indirect comparison forestimating efficacy of competing interventions: Empirical evidence from pub-lished meta-analyses. Bmj. 2003;326:472.

12. Haynes RB, McKibbon KA, Wilczynski NL, et al; for the Hedges T. Optimalsearch strategies for retrieving scientifically strong studies of treatment fromMedline: analytical survey. BMJ 2005;330(7501):1179.

13. Altman DG, Schulz KF, Moher D, et al. The revised CONSORT statement forreporting randomized trials: Explanation and elaboration. Ann Intern Med2001;134:663–694.

14. Juni P, Altman DG, Egger M. Systematic reviews in health care: Assessingthe quality of controlled clinical trials. BMJ 2001;323:42–46.

15. Egger M, Smith GD, Altman D. Systematic Reviews in Health Care. Meta-Analysis in Context.2nd ed. London: BMJ; 2001.

16. Lumley T. Network meta-analysis for indirect treatment comparisons. StatMed 2002;21:2313–2324.

17. DerSimonian R, Laird N. Meta-analysis in clinical trials. Control Clin Trial1986;7:177–188.

18. Higgins JPT, Thompson SG, Deeks JJ, Altman DG. Measuring inconsistencyin meta-analyses. BMJ 2003;327:557–560.

19. Begg CB, Mazumdar M. Operating characteristics of a rank correlation testfor publication bias. Biometrics 1994;50:1088–1101.

20. Egger M, Smith DG, Schneider M, Minder C. Bias in meta-analysis detectedby a simple, graphical test. BMJ 1997;315:629–634.

21. RevMan. Review Manager. Copenhagen: The Nordic Cochrane Center, TheCochrane Collaboration; 2008.

22. Moher D, Liberati A, Tetzlaff J, Altman DG. Preferred reporting items for sys-tematic reviews and meta-analyses: the PRISMA statement. Ann Intern Med2009;151:264–269,W264.

23. Wijermans P, Schaafsma M, Termorshuizen F, et al. Phase III study of thevalue of thalidomide added to melphalan plus prednisone in elderly patientswith newly diagnosed multiple myeloma: The HOVON 49 study. J Clin Oncol2010;28:3160–3166.

24. Djulbegovic B, Kumar A. Multiple myeloma: Detecting the effects of new treat-ments. Lancet 2008;371:1642–1644.

25. National Comprehensive Cancer Network (NCCN) Clinical Practice Guidelinesin Oncology. Multiple Myeloma V. 2.2009.

26. Alderson P. Absence of evidence is not evidence of absence. BMJ 2004;328:476–477.

27. Djulbegovic B. Articulating and responding to uncertainties in clinicalresearch. J Med Philos 2007;32:79–98.

24 American Journal of Hematology

research article