Presentation is loading. Please wait.

Presentation is loading. Please wait.

542-11-#1 Statistics 542 Introduction to Clinical Trials Meta Analysis.

Similar presentations

Presentation on theme: "542-11-#1 Statistics 542 Introduction to Clinical Trials Meta Analysis."— Presentation transcript:

1 #1 Statistics 542 Introduction to Clinical Trials Meta Analysis

2 #2 Meta-Analysis Alternatives?Occasionally Complementary?Yes Meta-Analysis Combination of similar studies using similar subjects and similar treatments and similar outcomes

3 #3 Figure 2 Figure 2 Odds Ratios and 95% Confidence Limits for Various Studies and a Pooled Estimate

4 #4 New Method of Analyzing Health Data Stirs Debate by Lawrence K. Altman Increasing use of a controversial statistical method to evaluate medical therapies and surgical procedures is beginning to affect profoundly the care of pregnant women and patients with cancer, heart disease and many other common conditions. The method, known as meta-analysis promises to plan an increasingly important role in determining health risks, environmental hazards and national policy on payment for medical care. Backers say technique can draw big, reliable conclusions from small, inconsistent findings. Meta-analysis is a term derived from the Greek meaning an analysis that is more comprehensive. The larger numbers obtained by combining studies provide a greater statistical power than any of the individual studies. Researchers are often able to draw more reliable inferences or new conclusions from the combined results than from the smaller studies that may be inconclusive individually. In earlier applications of meta-analysis, researchers evaluated intelligence quotients, government social welfare programs and many other topics. Meta-analysis has come to medicine late, but “it is now undergoing a boom in popularity,” said Dr. Thomas C. Chalmers, a distinguished physician of the Department of Veterans Affairs in Boston and a pioneer in methodology. The method involves an analysis of previous analyses. It combines the results of a wide range of existing smaller studies and then applies one of several statistical techniques to discover more precisely what is known from previous research. It may also produce a unified result from diverse, apparently contradictory studies. The technique has already shed new light on the effectiveness of medical therapies. Although it has not, in itself, revolutionized any medical treatment it has helped clear away the confusion caused by studies with scattered and apparently conflicting findings and has strengthen and confirmed findings from traditional clinical trials. NY Times 8/21/90

5 #5 Reference: NIH Proceedings Methodologic Issues in Overviews of Randomized Clinical Trials NIH Conference May 1986 Statistics in Medicine Vol 6, No. 3, 1987

6 #6 What is the Purpose? a.Testing for a treatment effect (rejecting the null hypothesis) b.Evaluating a safety issue (rare events) c.Estimating size of treatment effect in subgroups d.Design of new studies e.Develop practice guidelines

7 #7 Ideal Meta Analysis is Randomized Multi-center Control Trial Same protocol Same treatment Same type of subjects Same outcome measure

8 #8 Issues in Meta Analysis Differences Across Studies in: a.Treatment b.Control Group/Population c.Time Span (Disease, Background Therapy) d.Outcome Measures Publication Bias Completeness/Quality of Data Access to Data

9 #9 What Studies Should Be Included? All existing studies All published studies "Non-flawed" trials Other selection criteria

10 #10 Meta-Analysis: When? (1) Retrospective Analyses Test Treatment Effect When: –Definitive answer not yet available –No more studies likely –Need to salvage available results Develop Practice Guidelines Design New Studies

11 #11 Meta-Analysis: When? (2) Prospective Analyses Not recommended Better to design in advance proper multi-center trial(s)

12 #12 Meta-Analysis Methodology Not New Combining p-values, Fisher (1948) Analysis of Variance, Fisher (1938) Combining 2x2 Tables –Mantel-Haenszel (1959) –Cochran (1954)

13 #13 Odds Ratio more explicitly OR = ad/bc TC Saba + b Fcdc + d a + cb + d

14 #14 Methods of Meta-Analysis Collapsing can be misleading if there is qualitative interaction. 1.0 Collapse Data RCT-1 TC S155 F8595 OR = 3.35 RCT-2 TC S515 F9585 OR = 0.30 Collapsed TC OR = 1.0

15 #15 2.Graphical See Figure 95% CI for each study (ad / bc) exp { ± 1.96 (1/a + 1/b + 1/c + 1/d) } Methods of Meta Analysis

16 #16 Apparent effects of fibrinolytic treatment on morality in the randomised trials of IV treatment of acute myocardial infarction. Stat in Med 7:890: 1988.

17 #17 Comparison of meta-analysis of 12 RCTs of i.v.mixed drugs (double-blind) with i.v. metoprolol (double-blind) and i.v. atenlol (open study). Stat in Med 6(3): 320, 1987.

18 #18 Comparison of meta-analysis of mortality in 11 RCTs and reinfarction rates in 10 RCTs of i.v. streptokinase with large co-operative study (GISSI). Stat in Med 6(3): 320, 1987.

19 #19 Comparison of meta-analysis of 7 small RCTs of phenobarbital in the treatment of neonatal intra-cranial haemmorrhage with one large co-operative study (3 institutions). Endpoints are total infants with haemmorrhage and totals with severe haemorrhage (Grades III-IV) only. Stat in Med 6(3): 321, 1987.

20 #20 Odds Ratios and 95% Confidence Limits for Various Studies and a Pooled Estimate

21 #21 3.Blocking (Peto-MH) Overall Estimate Let O =  a i E =  E i E i = (a i + c i )(a i + b i ) n i V =  V i V i = (a i + c i ) )(b i + d i )(c i + d i )(a i + b i n i 2 (n i - 1) Z = O - E CPooled OR OR = exp { (O - E) / V } 95% CI = exp { (O - E) / V ± 1.96 / } Methods of Meta Analysis

22 #22 4.Averaging P-valuesFisher (1948) P i = P-value for i th trial Z = -2  log (P i ) ~  2 with 2N df 5.Averaging Test Statistics e.g. w i = n i Methods of Meta Analysis

23 #23 Meta-Analysis Examples Cardiology Post MI Treatments (e.g., beta-blockers, aspirin) Thrombolytic Therapy (e.g., streptokinase) Anticoagulants

24 #24 Registries/Databases Byar (1980) Biometrics D'Ambrosia, Ellenberg (1980) Biometrics Starmer et al. (1980) Biometrics Mantel (1983) Statistics in Medicine

25 #25 Registries/Databases Use Clinical Observational Series to: Describe Clinical Practice Identify Risk Factors "Evaluate" Treatment –Historical –Concurrent

26 #26 Databases Treatment Evaluation Comparison Requires Risk Factor Comparability –Measured –Not Measured or Unknown Statistical Models Usually Not Adequate –Association vs. Estimation –Model Only an Approximation –Small Portion of Outcome Explained

27 #27 Potential Biases Time Trends (Decline in CHD Death) Ascertainment –Changes in Diagnostic Criteria –Availability of Technology Selection Bias

28 #28 Compliance “Adjustment” ComplianceClofibratePlacebo < 80%24.6%28.2% > 80%15.0%15.1% All18.2%19.4% Coronary Drug Project (NEJM, 1980) 5 Year Mortality

29 #29 Registries Bias in Treatment Effect (Peto, Biomedicine, 1978) Trials of Anticoagulant Therapy DesignStudiesPatients Effect Historical % Reduction Concurrent % Reduction RCT % Reduction

30 #30 PTCA PTCA Registry –Tracked and compared usage –Lead to further trials –No PTCA vs. placebo TIMI-II –Compared immediate vs. delayed PTCA BARI –Compares PTCA vs. CABG

31 #31 CABG CASS RCT (Circulation, 1983) –Comparison of immediate vs. delayed CABG CASS Registry ( J Clin Inv, 1983) –Prognostic value of Angiography

32 #32Arboretum

Download ppt "542-11-#1 Statistics 542 Introduction to Clinical Trials Meta Analysis."

Similar presentations

Ads by Google