The metafor Package

A Meta-Analysis Package for R

User Tools

Site Tools


tips:comp_mh_different_software

Differences

This shows you the differences between two versions of the page.

Link to this comparison view

tips:comp_mh_different_software [2019/05/05 16:21] – external edit 127.0.0.1tips:comp_mh_different_software [2021/11/08 15:48] (current) Wolfgang Viechtbauer
Line 1: Line 1:
 ===== Comparison of the Mantel-Haenszel Method in Different Software ===== ===== Comparison of the Mantel-Haenszel Method in Different Software =====
  
-The Mantel-Haenszel method is an approach for fitting meta-analytic fixed-effects models when dealing with studies providing data in the form of 2x2 tables or in the form of event counts (i.e., person-time data) for two groups (Mantel & Haenszel, 1959). The method is particularly advantageous when aggregating a large number of studies with small sample sizes (the so-called sparse data or increasing strata case).+The Mantel-Haenszel method is an approach for fitting meta-analytic equal-effects models when dealing with studies providing data in the form of 2x2 tables or in the form of event counts (i.e., person-time data) for two groups (Mantel & Haenszel, 1959). The method is particularly advantageous when aggregating a large number of studies with small sample sizes (the so-called sparse data or increasing strata case).
  
 The method is available in the metafor package via the ''rma.mh()'' function. By default, the results obtained may differ slightly from those obtained via the ''metan'' function in Stata (for more details, see Harris et al., 2008; Sterne, 2009), the [[http://tech.cochrane.org/Revman|Review Manager]] (RevMan) from the [[http://www.cochrane.org/|Cochrane Collaboration]], or [[http://www.meta-analysis.com/|Comprehensive Meta-Analysis]] (CMA). The reason for such discrepancies is explained further below using an illustrative dataset from a meta-analysis comparing the risk of catheter-related bloodstream infection (CRBSI) when using anti-infective-treated versus standard catheters in the acute care setting (Niel-Weise et al., 2007). The method is available in the metafor package via the ''rma.mh()'' function. By default, the results obtained may differ slightly from those obtained via the ''metan'' function in Stata (for more details, see Harris et al., 2008; Sterne, 2009), the [[http://tech.cochrane.org/Revman|Review Manager]] (RevMan) from the [[http://www.cochrane.org/|Cochrane Collaboration]], or [[http://www.meta-analysis.com/|Comprehensive Meta-Analysis]] (CMA). The reason for such discrepancies is explained further below using an illustrative dataset from a meta-analysis comparing the risk of catheter-related bloodstream infection (CRBSI) when using anti-infective-treated versus standard catheters in the acute care setting (Niel-Weise et al., 2007).
Line 44: Line 44:
 </code> </code>
 <code output> <code output>
-Fixed-Effects Model (k = 18)+Equal-Effects Model (k = 18) 
 + 
 +I^2 (total heterogeneity / total variability):  5.12% 
 +H^2 (total variability / sampling variability): 1.05
  
 Test for Heterogeneity: Test for Heterogeneity:
Line 51: Line 54:
 Model Results (log scale): Model Results (log scale):
  
-estimate       se     zval     pval    ci.lb    ci.ub +estimate     se    zval   pval   ci.lb   ci.ub 
-  -1.209    0.222   -5.434    <.001   -1.645   -0.773+  -1.209  0.222  -5.434  <.001  -1.645  -0.773
  
 Model Results (OR scale): Model Results (OR scale):
  
-estimate    ci.lb    ci.ub +estimate  ci.lb  ci.ub 
-   0.299    0.193    0.462+   0.299  0.193  0.462
  
-Cochran-Mantel-Haenszel Test:    CMH = 32.214, df = 1,  p-val < .001+Cochran-Mantel-Haenszel Test:    CMH = 32.214, df = 1,  p-val < 0.001
 Tarone's Test for Heterogeneity: X^2 = 25.679, df = 16, p-val = 0.059 Tarone's Test for Heterogeneity: X^2 = 25.679, df = 16, p-val = 0.059
 </code> </code>
Line 129: Line 132:
 </code> </code>
 <code output> <code output>
-Fixed-Effects Model (k = 17)+Equal-Effects Model (k = 17)
  
-Test for Heterogeneity: +I^2 (total heterogeneity / total variability):  2.48% 
 +H^2 (total variability / sampling variability): 1.03 
 + 
 +Test for Heterogeneity:
 Q(df = 16) = 16.406, p-val = 0.425 Q(df = 16) = 16.406, p-val = 0.425
  
 Model Results (log scale): Model Results (log scale):
  
-estimate       se     zval     pval    ci.lb    ci.ub  +estimate     se    zval   pval   ci.lb   ci.ub 
-  -1.149    0.214   -5.356    <.001   -1.569   -0.728 +  -1.149  0.214  -5.356  <.001  -1.569  -0.728
  
 Model Results (OR scale): Model Results (OR scale):
  
-estimate    ci.lb    ci.ub  +estimate  ci.lb  ci.ub 
-   0.317    0.208    0.483 +   0.317  0.208  0.483
  
-Cochran-Mantel-Haenszel Test:    CMH = 30.919, df = 1,  p-val < .001+Cochran-Mantel-Haenszel Test:    CMH = 30.919, df = 1,  p-val < 0.001
 Tarone's Test for Heterogeneity: X^2 = 22.033, df = 16, p-val = 0.142 Tarone's Test for Heterogeneity: X^2 = 22.033, df = 16, p-val = 0.142
 </code> </code>
  
-These are the exact same results as obtained with Stata, RevMan, and CMA. However, the results of Bradburn et al. (2007) suggest that the ''1/2'' adjustment should only be used with caution when applying the Mantel-Haenszel method. Also, alternative correction factors could be considered, which may actually lead to more accurate results (see Sweeting et al., 2004). Finally, the findings by Bradburn et al. (2007) suggest that Peto's method (as implemented in the ''rma.peto()'' function) can actually give the least biased results and may be preferrable when events are rare (as long as treatment and control groups are of approximately equal size within trials and the true odds ratio underlying the studies is not very large).+These are the exact same results as obtained with Stata, RevMan, and CMA. However, the results of Bradburn et al. (2007) suggest that the ''1/2'' adjustment should only be used with caution when applying the Mantel-Haenszel method. Also, alternative correction factors could be considered, which may actually lead to more accurate results (see Sweeting et al., 2004). Finally, the findings by Bradburn et al. (2007) suggest that Peto's method (as implemented in the ''rma.peto()'' function) can actually give the least biased results and may be preferable when events are rare (as long as treatment and control groups are of approximately equal size within trials and the true odds ratio underlying the studies is not very large).
  
 ==== References ==== ==== References ====
tips/comp_mh_different_software.txt · Last modified: 2021/11/08 15:48 by Wolfgang Viechtbauer