How to Conduct a Meta-Analysis

Document Sample
How to Conduct a Meta-Analysis Powered By Docstoc
					How to Conduct a Meta-Analysis

     Arindam Basu MD MPH
        About the Author
       Required Browsing

Following the course, you will be able to:
•   Define Meta-analysis
•   Select Studies for a Meta-analysis
•   Identify different types of Models
•   Calculate Summary Effects
•   Interpret Results of a Meta-analysis
    What is Meta-analysis?
• Synthesis of previous studies
• Providing a Summary estimate
• Steps
  –Identify studies
  –Define Eligibility Criteria
  –Abstract Data
  –Do Statistical Analysis
  Identify Studies: Overview
• Steps:
 –Search Your Personal Files First
 –Search Electronic Databases
 –Review Reference Lists of Articles
 –Contact Experts and Researchers
 –Retrieve New Articles
• Evaluate Quality of the Studies
• Set Up Eligibility Criteria
Searching Electronic Databases
• First, Define a Search Strategy
• Limitations of Databases
 –incomplete and imperfect queries
 –language problems
 –problems with fugitive literature
• Publication Bias is important
 –What is publication bias
 –How to deal with publication bias
       Evaluating Study Quality

•   Define Study Quality Criteria Early
•   Set Up A Good Scoring System
•   Develop A Form for Assessment
•   Calculate Quality for each Study
•   Use this for Sensitivity Analysis
    –stratify studies according to quality
Defining Eligibility of Studies
Select Eligible Studies Based On:
• Study Designs
• Years of Publication
• Language
• Choice among multiple articles
• Sample-size or follow-up issues
• Similarity of Exposure and/or Rx
• Completeness of information
   Abstract Data - Review!
• Identify Relevant Articles
• Sort out Eligible Articles
• Set up a Form for Abstraction
• Enter the Eligible Studies
• Use this as your database
Statistical Analysis is next...
    Statistical Analysis - Overview
•   Select An Estimate of Effect
•   Choose An Effects Measure
•   Select An Effects Model
•   For Each Model:
    –Calculate Summary Effect Size
    –Calculate Confidence Intervals
    –Calculate Q-statistic for Homogeneity
• Perform Sensitivity Analysis
    Selecting Estimate of Effect
• Choose Only One Estimate
• For RCTS, choose the one with
 –Once randomized always randomized
• For nonrandomized trials, choose:
 –estimate adjusted only for age
 –that, and for a known confounder
 –the “most adjusted” estimate
 –estimate presented in the abstract
 Choosing An Effect Measure
• RCTs or Cohort Studies
 –Rate Difference between Treatment
  and Control Groups
 –Ratio of Disease Rates
• Case Control Studies
 –Odds’ Ratio
 –Rate Ratio
 Selecting An Effects Model

• Available Types:
 –Fixed Effects Model
 –Random Effects Model
• Difference Between the Two
• Special Cases:
 –When Outcomes are not binary
• Methods to be Used for them
     Fixed Effects Model
• Methods:
 –Mantel Haenszel Method
 –Peto’s Method
 –General Variance Based Methods
   • For Rate Difference
   • For Rate Ratios
   • When only RR and 95 CI given
• Tests of Homogeneity
 –Calculation of Q Statistic
     Mantel Haenszel Method

• Download Spreasheet Calculator
• Strength of Mantel Haenszel
 –Very powerful
 –Widely Used
• Limitation
 –Cannot Control For Confounding!
           Peto’s Method
•   Similar to Mantel Haenszel
•   Download Calculator
•   Simpler Computation
•   No Control for Confounding
•   Good for RCTs
•   Requires 2 X 2 Table
    Variance-Based Methods
• Download Calculators For:
  –Rate Difference
  –For only Relative Risk and 95% CI
• Strengths and Limitations
  –Good For Rate Differences
  –Computationally Intensive
      Tests of Homogeneity
• Establish Null Hypothesis that
  Effect Sizes Are Equal in All of the
• Tested By Using Q-statistic
• Q-statistic is distributed as chi-
  square distribution with degree of
  freedom = n-1 where n = number of
      Random Effects Model

• Download The Calculator!
• Strengths and Limitations:
 –Can Generalize the Conclusions
 –Computationally Intensive
     Continuous Outcomes
• Measurement Scale: Continuous
• Outcome Measured in Same Scale
 –Download Spreadsheet Calculator!
• Essentially Extension of ANOVA
• Useful For Integrating Social
  Science Research Data
 Sensitivity and Publication Bias

• Conduct Sensitivity Analysis
 –Stratify studies by their quality rating
 –Compare Fixed and Random Effects
• Deal With Publication Bias
 –Construct A Funnel Plot

• Presentation of Meta-analysis or
  Literature Synthesis
• Download The Graph Maker
• More Resources...

Shared By: