The Use of Meta-Analysis in Cost-Effectiveness Analysis

The Use of Meta-Analysis inCost-Effectiveness AnalysisIssues and Recommendations

Sanjay Saint,1* David L. Veenstra2 and Sean D. Sullivan3

1 Division of General Medicine, University of Michigan Medical Center, Ann Arbor, Michigan, USA2 Department of Pharmacy, University of Washington, Seattle, Washington, USA3 Department of Health Services, University of Washington, Seattle, Washington, USA* Work conducted while an employee at the Department of Medicine, University of Washington,

Seattle, Washington, USA

Abstract Meta-analysis is used to statistically pool the results from individual studies,usually randomised trials, to obtain an estimate of the summary effect size acrossstudies. The summary measure from a meta-analysis is often used to derive theprobability of treatment success in a cost-effectiveness analysis. Recently, LeLorierand colleagues questioned the ability of meta-analysis to accurately predict theresults of a subsequent large-scale trial, implying that the use of a summarymeasure from a meta-analysis may be inappropriate in an economic evaluation.We comment on this potential shortcoming by first providing an outline of theuse of meta-analysis results in a cost-effectiveness analysis. Then, using examplesof discrepancies between meta-analyses and subsequent large trials noted byLeLorier and colleagues, we examine the potential impact of using the resultsfrom a small trial versus a meta-analysis.

We found that the meta-analyses were comparable to or better than small trialsat predicting the results of subsequent large trials. We, therefore, argue that ameta-analysis of homogeneous studies can provide a reasonable estimate of thetreatment effect for use in a cost-effectiveness analysis when no large, definitiveclinical trial has been performed. However, care must be taken not to over-interpret the precision of the estimate, since both the homogeneity and quality ofthe primary studies need to be considered. We conclude by providing guidanceon the appropriate use of summary measures derived from meta-analyses forcost-effectiveness studies.

CURRENT OPINION Pharmacoeconomics 1999 Jan; 15 (1): 1-81170-7690/99/0001-0001/$04.00/0

© Adis International Limited. All rights reserved.

Cost-effectiveness analysis (CEA) of pharma-ceuticals has become increasingly important as amethod for informing resource allocation and for-mulary listing decisions. Indeed, some countries (e.g.Australia and Canada) require favourable cost-effectiveness evaluations before drugs are reimbursedunder government-sponsored insurance schemes.

These cost-effectiveness studies frequently take theform of economic or simulation models that com-bine data from a myriad of sources in order toproject costs and effectiveness of alternative treat-ments.

Cost-effectiveness evaluations of healthcare in-terventions depend on solid clinical evidence in

order to establish the benefits and risks. The valid-ity of the data is crucial to the overall validity ofthe predictions from the model. Estimates derivedfrom large-scale, multicentre randomised trials arewidely considered the ‘gold standard’ for assessingefficacy,[1,2] but these data often are not availableat the time the cost-effectiveness study is requiredfor decision-making. Thus, many cost-effectivenessevaluations rely on a single randomised trial orobservational study to estimate a treatment effect.If several small trials are available, however, theanalyst must decide whether to use data from just1 of these studies or statistically pool the resultsusing meta-analysis.

Recently, the ability of meta-analysis to predictthe results of subsequent large clinical trials hasbeen questioned. The purpose of this paper is tocomment on the use of meta-analysis results in cost-effectiveness studies in light of this issue. We arguethat using the summary estimate and confidencebounds derived from a meta-analysis in a cost-effectiveness study is appropriate, with certain qual-ifications, when data from a large randomised trialare unavailable. Finally, we provide guidance onhow best to use summary estimates from a meta-analysis in an economic evaluation.

1. Meta-Analysis

Meta-analysis is a ‘... quantitative approach forsystematically combining the results of previousresearch in order to arrive at conclusions about thebody of research.’[3] First proposed as a researchmethodology by Light and Smith, the term ‘meta-analysis’ was coined by Glass in 1976.[4] The pastdecade has seen a large growth in the interest inmeta-analysis. Using the keyword ‘meta-analysis’and searching the Medline database, we found thatthe number of retrieved reports has increased from42 in 1987 to 923 in 1997. Indeed, the CochraneCollaboration has undertaken the enormous taskof developing systematic meta-analyses of publishedand unpublished clinical trials of treatment efficacy.[5]

Like any research method, meta-analysis hasstrengths and drawbacks.[3,4,6-9] The well-knownstrengths of meta-analysis include: (i) a reduction

in type II error when small studies show no signif-icant treatment effect; (ii) a provision of a summaryand confidence bound estimate of effectiveness inthe absence of a definitive trial; and (iii) a probableexplanation of heterogeneity in disparate results ofexisting trials. However, meta-analysis is not with-out limitations. These include: (i) the possible in-troduction of bias during selection of studies (e.g.limiting to English language[10] or published stud-ies[11]); (ii) the heterogeneity of studies such thatstatistical pooling is not valid;[12] and (iii) empha-sis on the summary estimate of effect even thoughthe quality of pooled studies may be poor.[13]

Recently, the ability of meta-analysis to predictthe results of a large and presumably definitive ran-domised trial has been questioned in a provocative,yet controversial study by LeLorier et al.[14] Theinvestigators compared the results of 12 large, ran-domised trials published in 4 leading general med-ical journals with the results of 19 meta-analysespublished earlier and addressing similar study ques-tions. The investigators found that the agreementbetween the meta-analysis and a subsequent largetrial was only fair (� = 0.35) and that the positivepredictive value of the meta-analysis result was 68%,leading some to question the reliability and thus theclinical utility of meta-analysis.[15] This evaluationby LeLorier[14] is the latest in a series of studiesthat have compared meta-analysis with clinical tri-als.[16,17] Recent work by Ioannidis et al.[18] clarif-ies the discrepancies identified by LeLorier et al.[14]

suggesting that the conclusions drawn by LeLorieret al.[14] perhaps should be re-examined.

Importantly, LeLorier et al.[14] found that the oddsratio (OR) from the meta-analysis and from the largetrial were on the same side of unity 80% of the time.In fact, no case was found in which the estimateswere both statistically significant and on oppositesides of unity. According to the definitions used byLeLorier et al.[14], all the discrepancies between theresults of the meta-analysis and those from the largeclinical trial occurred because 1 showed a statisti-cally significant treatment effect whereas the otherdid not. The challenges for economic evaluationrelate to the inherent shortcomings of meta-analysis

2 Saint et al.

� Adis International Limited. All rights reserved. Pharmacoeconomics 1999 Jan; 15 (1)

and how the use of pooled estimates affects theresults and interpretation of cost-effectiveness stud-ies.

2. Use of Meta-Analysis inCost-Effectiveness Analysis

A meta-analysis provides a point estimate andconfidence interval to represent the summary treat-ment effect across studies. Both of these valuesshould then be incorporated into a CEA using ap-propriate techniques. The results of a meta-analysis,however, must often be adapted for use in a CEA.The summary outcome measure calculated in a meta-analysis is usually a measure of relative risk such asthe OR or risk ratio. However, economic evalua-tions typically use decision analytic or mathemat-ical models that require an estimate of the condi-tional probability or absolute (rather than relative)risk. Absolute risks can be determined directly fromthe primary studies using common statistical meth-ods to combine proportions for both interventionand control groups.[19,20]

Several recent papers have used pooled treat-ment effects from meta-analyses in an economicevaluation.[20-22] For example, O’Brien et al.[20] stud-ied the cost effectiveness of the treatment of recur-rent duodenal ulcer. Focusing on recurrence riskbetween 6 and 12 months, the individual trials usedin the analysis gave a summary OR of 0.34 [95%confidence interval (CI): 0.05 to 2.44] when com-paring patients treated with ranitidine with thosegiven placebo, indicating that ranitidine decreasedthe odds of ulcer recurrence by about 66%. As men-tioned above, an OR provides estimates of relativerather than absolute risk. Instead, by using methodsfor combining proportions,[19] the analyst can deriveestimates for treatment and placebo groups that canbe directly incorporated into an economic analysis.Pooling the proportion of patients experiencingrelapse produces an absolute risk of recurrence ofapproximately 6% (95% CI: 0 to 13%) in the rani-tidine group and 34% (95% CI: 20 to 47%) in theplacebo group.[20]

An examination of the uncertainty in a CEA causedby uncertainty in the treatment effect estimate (as

well as other estimates) is essential for a rigorousanalysis.[23,24] In deterministic models, this is doneby performing a sensitivity analysis using, com-monly, the upper and lower bounds of the 95% CIfor the effect estimate. In stochastic models, theuncertainty in the effect estimate is inherently in-corporated into the model by using a probabilitydistribution to define the effect estimate, and thestandard deviation of the distribution is typicallybased on the 95% CI of the effect estimate.[25] Thus,the uncertainty in the denominator of the cost-effectiveness ratio is usually based on the CI of theeffect estimate in both deterministic and stochasticmodels. As discussed in section 3, this has impor-tant implications for the evaluation of the validityof meta-analysis, particularly in economic studies.

3. Consequences of Using EstimatesDerived from Meta-Analyses

When presented with a series of smaller, non-definitive trials, the analyst conducting a cost-effectiveness study must decide whether to use meta-analytic techniques to derive a pooled estimate or tosimply use an estimate from the largest or highestquality trial available. Specifically, the investiga-tor must often decide to use information from ei-ther: (i) a small, nondefinitive trial; or (ii) a meta-analysis. We examine the potential consequencesof these 2 strategies by evaluating a subset of thestudies analysed by LeLorier et al.[14] as an exam-ple. We provide OR rather than absolute risks forconsistency with the reporting of most meta-analy-ses.

The most concerning discrepancies in the studyby LeLorier et al.[14] from the perspective of a CEAare those in which the CIs of the meta-analysis andthe subsequent large randomised trial do not over-lap. In such an instance, a sensitivity analysis con-ducted as part of an economic analysis would notinclude the 95% CI of the subsequent large trial.There are 5 instances in 4 meta-analyses of thistype of disagreement.[26-29]

In figure 1, we present the OR and 95% CI forthe largest trials included in 3 of these meta-analysesalong with the results from the meta-analysis itself

Meta-Analysis and CEA 3


and the subsequent large randomised trial to illus-trate the potential problems.

Leizorovicz et al.[26] reported a summary OR of1.02 (95% CI: 0.90 to 1.16) for major bleedingepisodes in general surgery with the use of low-mo-lecular-weight heparin versus unfractionated heparinin an analysis of 23 studies (fig. 1, panel A). Theauthors specifically note the heterogeneity in thedefinitions of major bleeding in the different tri-als.[26] If an estimate is derived from the largeststudy[30] used in the meta-analysis, the OR for ma-jor bleeding is 1.01 (95% CI: 0.79 to 1.30), whichis based on the number of patients requiring bloodtransfusions. Thus, the point estimates of the meta-analysis and trial are similar although the CI is nar-rower for the meta-analysis. LeLorier et al.[14] re-port an OR for major bleeding from a more recent,large trial of 0.50 (95% CI: 0.28 to 0.89).[31] How-ever, the definition of major bleeding used in thistrial was much broader. Using the requirement for

blood transfusions to define major bleeding in thesame trial[31] produces an OR of 1.01 (95% CI: 0.70to 1.46), in close agreement with the previous smallertrial[30] and with the meta-analysis.[26]

There are 2 lessons here. First, the outcome def-initions of seemingly similar trials can have a sig-nificant impact on the effect size. Second, a meta-analysis may not result in a point estimate far differentfrom the largest trial included in the analysis, but theCI will usually be smaller. This will impact the sen-sitivity analysis of the CEA.

An often-cited example of disagreement betweenmeta-analysis and subsequent trials is the use of ni-trates following myocardial infarction.[14,17,32] Yusufet al.[27] reported an OR for short term mortality of0.55 (95% CI: 0.39 to 0.76) in a meta-analysis thatincluded a total of 851 patients (fig. 1, panel B).However, the Gruppo Italiano per lo Studio dellaSopravvivenza nell’infarto (GISSI-3) trial, with

A

B

C

Major bleeding with LMWH vs unfractionatedheparin in general surgery

Mortality with nitrates vsplacebo post-MI

Intrauterine growth retardationwith aspirin vs placebo

Previous trial: EFS[30]

Previous trial: Beufils et al.[36]

Previous trial: MacParland et al.[37]

Meta-analysis: Imperiale & Petrulis[28](>18 wks)

Meta-analysis: Imperiale & Petrulis[28](all trials)

Subsequent trial: CLASP[35]

Previous trial: Jugdutt & Warnica[34]

Meta-analysis: Yusuf et al.[27]

Subsequent trial: GISSI-3[33]

Meta-analysis: Leizorovicz et al.[26]

Subsequent trial: Kakkar et al.[31] (bleeding)

Subsequent trial: Kakkar et al.[31] (transfusions)

0 0.5 1 1.5 2

OR (95% CI)

Fig. 1. Selected examples of studies used in the study by LeLorier et al.[14] A comparison of the results from meta-analysis and thelargest trial before and after the meta-analysis. CI = confidence interval; CLASP = Collaborative Low-Dose Aspirin Study in Pregnancy;EFS = European Fraxiparin Study; GISSI-3 = Gruppo Italiano per lo Studio della Sopravvivenza nell’infarto; LMWH = low-molecular-weight heparin; MI = myocardial infarction; OR = odds ratio.

4 Saint et al.


43 047 patients, found no benefit associated withthe use of nitrates (OR: 0.94; 95% CI: 0.84 to 1.05).[33]

What are the implications of this discrepancyfor CEA? Suppose the largest trial from the meta-analysis, the trial performed by Jugdutt and War-nica,[34] is used as the basis for an economic eval-uation. In this trial, the OR for early mortality withnitrates versus placebo is 0.21 (95% CI: 0.09 to0.49).[34] This result is not in agreement with thesubsequent GISSI-3 trial, yet it is consistent withthe meta-analysis. Panel B of figure 1 shows thatthe CIs of the earlier trial and the meta-analysisoverlap. Using the result from either the single largesttrial or the meta-analysis in this case might indicatethat use of nitrates after myocardial infarction wascost effective or even dominant.

Which study should be chosen for use in an eco-nomic evaluation? It is not clear based on theseresults alone. The trials used in a meta-analysis mustbe examined individually for homogeneity of studycharacteristics including the nature of the interven-tion and similarity of treatment patterns. Indeed,Borzak and Ridker[32] have outlined possible rea-sons for the differences between the above trialsevaluating nitrates after myocardial infarction, in-cluding changing treatment patterns and off-trialuse of nitrates. These results emphasise that a CEAshould evaluate a patient population and interven-tion similar to those in a trial or represented in ameta-analysis.

In another example, Imperiale and Petrulis[28]

analysed trials investigating the efficacy of low-dose aspirin in preventing pregnancy-induced hy-pertension. LeLorier et al.[14] indicated discrepantresults compared with the Collaborative Low-doseAspirin Study in Pregnancy (CLASP) trial[35] forthe adverse outcome of intrauterine growth retar-dation. Apparently, LeLorier et al.[14] reported thepooled results from just the 2 studies in the meta-analysis with more than 18 weeks of treatment, whichgive a summary OR of 0.22 (95% CI: 0.08 to 0.61),in comparison with the OR from the CLASP trialof 0.92 (95% CI: 0.80 to 1.07)[35] [fig. 1, panel C].However, using all 6 trials in the meta-analysis,the summary OR was 0.59 (95% CI: 0.35 to 0.99),

while the 2 largest trials included in the meta-analysishad ORs of 0.22 (95% CI: 0.07 to 0.75) and 1.10(95% CI: 0.36 to 3.40).[36,37] In this instance, themeta-analysis including all 6 trials provides a rea-sonable estimate incorporating the disparate find-ings of 2 similarly sized trials and avoids the po-tential bias introduced by selecting only 1 of thestudies.

A final example concerns the broad applicationof meta-analysis results in economic models wherethe intervention is either poorly or not representedin any of the individual trials. The meta-analysisby Ravnskov[29] reports discrepant results for 2 out-comes, coronary death and all cause mortality, com-paring hypercholesterolaemia treated with variouscholesterol-lowering agents to the Scandinavian Sim-vastatin Survival Study (4S).[38] Simvastatin was thecholesterol-reducing intervention in the 4S studybut was not among the interventions included inthe trials used for the meta-analysis. Using the re-sults of a meta-analysis in which the individual tri-als do not include the intervention of interest inorder to imply a class effect in an economic modelof a specific intervention should be undertaken withappropriate caution. In addition, the US Food andDrug Administration (FDA) has viewed this prac-tice with skepticism.[39]

These examples highlight several importantissues. First, the interventions and outcomes stud-ied in a meta-analysis should be homogeneous andsimilar to the intervention of interest for the eco-nomic evaluation. Some of the discrepancies foundby LeLorier et al.[14] are likely to be due to thesedifferences. Second, the point estimate from a meta-analysis is often similar to that of the largest trialincluded, but the CI will be smaller. Finally, meta-analysis may be better at predicting subsequent de-finitive trials than the largest trial available to date.Thus, although the strategy of using data from asingle nondefinitive trial is usually not inappropri-ate, we believe that using meta-analytic techniquesenables the incorporation of greater informationand a less biased estimate when the appropriateconditions of homogeneity are met.



4. Recommendations forUsing Meta-Analysis inCost-Effectiveness Studies

One of the most important steps in a cost-effec-tiveness study is a review and synthesis of theavailable evidence of intervention effectiveness. Out-come estimates from a single trial can be used if thetrial provides either the only established evidenceor the best evidence because of statistical power,higher quality, or is more reflective of the patientpopulation, intervention, time horizon and specificoutcome of interest. If a single data source is usedfor an estimate of effect, a brief review of the avail-able evidence and the reasons for choosing the trialshould be provided.

When it is not clear that a single study is mostappropriate for modelling outcomes, data from mul-tiple studies can either be selected based on a qual-itative review or statistically pooled using meta-analytic techniques. A qualitative review offers theadvantage of comprehensively describing the de-tails of study design and quality, 2 aspects that aredifficult to incorporate in a statistical synthesis. How-ever, important potential disadvantages of a quali-tative review include the introduction of bias in theselection of studies and the loss of information fromexcluded studies. Meta-analysis, on the other hand,can incorporate evidence from multiple studies intoa summary estimate of effect that can be used in aneconomic model and can minimise selection bias.Indeed, a properly conducted meta-analysis makesit more difficult to disregard information that maynot support the beneficial outcome of the interven-tion.

It is essential to note, however, that proper clin-ical and methodological judgement must be usedwhen conducting and interpreting a meta-analysis.In particular, the interventions, patient populationand outcomes measured should be similar amongthe individual trials. When these characteristics aredissimilar, misleading conclusions about both theresults and validity of meta-analysis can be made.[18]

Potential sources of heterogeneity need to be rigor-ously explored in order for the meta-analysis to beuseful.[12]

A meta-analysis incorporated into a cost-effec-tiveness study should be conducted according tostandard methods.[40] A systematic and reproduci-ble search should be conducted and preferably in-clude non-English language and unpublished stud-ies. Study exclusion and inclusion criteria shouldbe explicitly stated, the outcomes of interest clearlydefined and data abstraction conducted by at least2 investigators. Proper statistical techniques shouldbe used for deriving pooled estimates of probabil-ities, although pooled estimates of relative risk (e.g.risk ratio, OR) should also be calculated for compar-ison with trial data. The methods used to statisti-cally pool data should be clearly presented as partof the economic evaluation.

A cost-effectiveness study using outcome esti-mates from a meta-analysis should employ a sen-sitivity analysis including, at minimum, the lowerand upper bounds of the 95% CI for the summarytreatment effect. Sensitivity analyses using theresults from individual trials and subgroups of tri-als may be indicated when the trials are of varyingquality, use different interventions or outcome meas-ures, or report significantly different results. CEAsusing estimates from meta-analyses with statisticalheterogeneity should be interpreted with caution.In the setting of statistical heterogeneity, however,it may be appropriate for the economic analysis tofocus on just 1 of the populations or interventionsincluded in the meta-analysis if that subgroup isrelatively homogeneous.

5. Conclusions

Estimates of the probability of a treatment suc-cess in an economic model may be derived fromseveral sources, including clinical trials and meta-analysis. When data from a definitive clinical trialare not available, the analyst must decide whetherto use estimates derived from either a single trialor a meta-analysis if more than 1 trial is available.Recently, many of the pitfalls of meta-analysis havebeen highlighted by comparisons with subsequentlypublished large, ‘definitive’ clinical trials. It is alsoclear that when overemphasis is placed on summarypoint estimates without an exploration of differences

6 Saint et al.


among studies, misleading conclusions can be made.However, when a series of homogeneous trials areavailable, meta-analysis offers important advantagesover using either data from a single trial or esti-mates based on a qualitative review. As such, ameta-analysis conducted as part of an economicevaluation should be performed and reported in athorough fashion. This should include reporting ofa sensitivity analysis that addresses variation in in-dividual study results as well as variation in thepooled summary measure. Despite the recent con-troversy regarding the validity of meta-analysis,we believe that estimates derived from meta-analysisremain useful in certain circumstances.

Acknowledgements

Sanjay Saint was a Robert Wood Johnson Clinical Scholarwhen this work was completed. David L. Veenstra is supportedby a Roche postdoctoral fellowship.

References1. Horwitz RI. ‘Large-scale randomized evidence: large, simple

trials and overviews of trials’: discussion. A clinician’s per-spective on meta-analyses J Clin Epidemiol 1995; 48: 41-4

2. Peto R, Collins R, Gray R. Large-scale randomized evidence:large, simple trials and overviews of trials. J Clin Epidemiol1995; 48: 23-40

3. Petitti DB. Meta-analysis, decision analysis, and cost-effective-ness analysis. New York (NY): Oxford University Press, 1994

4. Sacks HS, Berrier J, Reitman D, et al. Meta-analyses of ran-domized controlled trials. N Engl J Med 1987; 316: 450-5

5. Bero L, Rennie D. The Cochrane Collaboration. Preparing,maintaining, and disseminating systematic reviews of the ef-fects of health care. JAMA 1995; 274: 1935-8

6. Lau J, Ioannidis JP, Schmid CH. Summing up evidence: oneanswer is not always enough. Lancet 1998; 351: 123-7

7. Liberati A. ‘Meta-analysis: statistical alchemy for the 21st cen-tury’: discussion. A plea for a more balanced view of meta-analysis and systematic overviews of the effect of health careinterventions [comment]. J Clin Epidemiol 1995; 48: 81-6

8. Spitzer WO. The challenge of meta-analysis [editorial; com-ment]. J Clin Epidemiol 1995; 48: 1-4

9. Feinstein AR. Meta-analysis: statistical alchemy for the 21stcentury J Clin Epidemiol 1995; 48: 71-9

10. Gr’egoire G, Derderian F, LeLorier J. Selecting the language ofthe publications included in a meta-analysis: is there a Towerof Babel bias? J Clin Epidemiol 1995; 48: 159-63

11. Lau J, Ioannidis JP, Schmid CH. Quantitative synthesis in sys-tematic reviews. Ann Intern Med 1997; 127: 820-6

12. Thompson SG. Why sources of heterogeneity in meta-analysisshould be investigated. BMJ 1994; 309: 1351-5

13. Schulz KF, Chalmers I, Hayes RJ, et al. Empirical evidence ofbias: dimensions of methodological quality associated withestimates of treatment effects in controlled trials. JAMA 1995;273: 408-12

14. LeLorier J, Gr’egoire G, Benhaddad A, et al. Discrepanciesbetween meta-analyses and subsequent large randomized,controlled trials. N Engl J Med 1997; 337: 536-42

15. Bailar JC, 3rd. The promise and problems of meta-analysis [ed-itorial; comment]. N Engl J Med 1997; 337: 559-61

16. Villar J, Carroli G, Beliz’an JM. Predictive ability of meta-anal-yses of randomised controlled trials. Lancet 1995; 345: 772-6

17. Cappelleri JC, Ioannidis JP, Schmid CH, et al. Large trials vsmeta-analysis of smaller trials: how do their results compare?.JAMA 1996; 276: 1332-8

18. Ioannidis JPA, Cappelleri JC, Lau J. Issues in comparisons be-tween meta-analyses and large trials. JAMA 1998; 279:1089-93

19. Laird NM, Mosteller F. Some statistical methods for combiningexperimental results. Int J Technol Assess Health Care 1990;6: 5-30

20. O’Brien B, Goeree R, Mohamed AH, et al. Cost-effectivenessof Helicobacter pylori eradication for the long-term manage-ment of duodenal ulcer in Canada. Arch Intern Med 1995;155: 1958-64

21. Taylor JL, Zagari M, Murphy K, et al. Pharmacoeconomic com-parison of treatments for the eradication of Helicobacter py-lori. Arch Intern Med 1997; 157: 87-97

22. Wong JB, Koff RS, Tine F, et al. Cost-effectiveness of inter-feron-alpha 2b treatment for hepatitis B e antigen-positivechronic hepatitis B. Ann Intern Med 1995; 122: 664-75

23. Drummond MF, O’Brien B, Stoddart GL, et al. Methods for theEconomic Evaluation of Health Care Programmes. 2nd ed.New York (NY): Oxford University Press, 1997

24. Gold MR, Siegel JE, Russell LB, et al. Cost-effectiveness inhealth and medicine. New York (NY): Oxford UniversityPress, 1996

25. Doubilet P, Begg CB, Weinstein MC, et al. Probabilistic sensi-tivity analysis using Monte Carlo simulation: a practical ap-proach. Med Decis Making 1985; 5: 157-77

26. Leizorovicz A, Haugh MC, Chapuis FR, et al. Low molecularweight heparin in prevention of perioperative thrombosis. BMJ1992; 305: 913-20

27. Yusuf S, Collins R, MacMahon S, et al. Effect of intravenousnitrates on mortality in acute myocardial infarction: an over-view of the randomised trials. Lancet 1988; 1: 1088-92

28. Imperiale TF, Petrulis AS. A meta-analysis of low-dose aspirinfor the prevention of pregnancy-induced hypertensive dis-ease. JAMA 1991; 266: 260-4

29. Ravnskov U. Cholesterol lowering trials in coronary heart dis-ease: frequency of citation and outcome. BMJ 1992; 305:15-9

30. The European Fraxiparin Study (EFS) Group. Comparison of alow molecular weight heparin and unfractionated heparin forthe prevention of deep vein thrombosis in patients undergoingabdominal surgery. Br J Surg 1988; 75: 1058-63

31. Kakkar VV, Cohen AT, Edmonson RA, et al. Low molecularweight versus standard heparin for prevention of venous throm-boembolism after major abdominal surgery: the Thrombo-prophylaxis Collaborative Group. Lancet 1993; 341: 259-65

32. Borzak S, Ridker PM. Discordance between meta-analyses andlarge-scale randomized, controlled trials: examples from themanagement of acute myocardial infarction. Ann Intern Med1995; 123: 873-7

33. Gruppo Italiano per lo Studio della Sopravvivenza nell’infarto(GISSI) Miocardico. GISSI-3: effects of lisinopril and trans-dermal glyceryl trinitrate singly and together on 6-weekmortality and ventricular function after acute myocardial in-farction. Lancet 1994; 343: 1115-22



34. Jugdutt BI, Warnica JW. Intravenous nitroglycerin therapy tolimit myocardial infarct size, expansion, and complications:effect of timing, dosage, and infarct location. Circulation 1988;78: 906-19

35. Collaborative Low-dose Aspirin Study in Pregnancy (CLASP)Collaborative Group. CLASP: a randomised trial of low-doseaspirin for the prevention and treatment of pre-eclampsia among9364 pregnant women. Lancet 1994; 343: 619-29

36. Beaufils M, Uzan S, Donsimoni R, et al. Prevention of pre-eclampsia by early antiplatelet therapy. Lancet 1985; I: 840-2

37. McParland P, Pearce JM, Chamberlain GV. Doppler ultrasoundand aspirin in recognition and prevention of pregnancy-in-duced hypertension. Lancet 1990; 335: 1552-5

38. Randomised trial of cholesterol lowering in 4444 patients withcoronary heart disease: the Scandinavian Simvastatin SurvivalStudy (4S). Lancet 1994; 344: 1383-9

39. Food and Drug Administration Modernization Act of 1997. PublicLaw No. 105-115, section 114

40. Cook DJ, Sackett DL, Spitzer WO. Methodologic guidelinesfor systematic reviews of randomized control trials in healthcare from the Potsdam Consultation on Meta-Analysis. J ClinEpidemiol 1995; 48: 167-71

Correspondence and reprints: Dr Sanjay Saint, Division ofGeneral Medicine, University of Michigan Medical Center,3116 Taubman Center, 1500 E. Medical Center Drive, AnnArbor, MI 48109-0376, USA.E-mail: [email protected]

8 Saint et al.


Documents

The Use of Meta-Analysis in Cost-Effectiveness Analysis