Download presentation
Presentation is loading. Please wait.
Published byΦίλανδρος Σαβαώθ Μητσοτάκης Modified over 6 years ago
1
Meta-Analysis Glass (Education Researcher, 5:3-8; 1976) suggested that there were three types of analyses: Primary analysis tests the hypothesis for which the data were collected. Secondary analysis re-analyzes data already collected by someone else to answer questions raised in primary analysis or to address new questions. Meta-analysis combines evidence across studies.
2
Most often meta-analysis is based on summary measures derived from
the literature : difference in risk difference in means odds ratios … Meta-analyses appearing in the medical literature have most often dealt with summarizing treatment effects.
3
Some Formalization We have summary measures from k studies
and variances of these summary measures We assume the are known constants
4
Fixed Effects Model
5
A test statistic If the fixed effects model is true, then: and
6
libname meta "&path\metaiml";
/*the pre-eclampsia data Pre-eclampsia is high bp and urine protein after 20 weeks of pregnancy */ proc print data=meta.preeclampsia; run;
7
%let data=meta.preeclampsia;
%let effect=lnor; %let var=varlnor; %let outdata=tmp; proc sql noprint; select SUM(&effect*(1/&var))/SUM(1/&var) , 1/SUM(1/&var) into :fixedmn,:fixedvar from &data; select sum( 1/&var*(&effect-&fixedmn)**2), count(*)-1 into :qstat, :qstat_df quit; data &outdata; fixedeffectsmean=&fixedmn; fixedeffectsvar=&fixedvar; qstat=&qstat; q_df=&qstat_df; q_pval=1-cdf("chisq",qstat,q_df); lower95_ci=fixedeffectsmean-1.96*sqrt(fixedeffectsvar); upper95_ci=fixedeffectsmean+1.96*sqrt(fixedeffectsvar); run; proc print data=tmp;run;
8
The fixed effects macro
%macro fixedeff(effect=,var=,data=&syslast,outdata=fixedresults); /*====== Parameters: effect=variable with estimated effect size var=variable with variance of estimated effect size data=data set containing the effect and its variance outdata=name of dataset to contain results */ proc sql; select SUM(&effect*(1/&var))/SUM(1/&var) , 1/SUM(1/&var) into :fixedmn,:fixedvar from &data; select sum( 1/&var*(&effect-&fixedmn)**2), count(*)-1 into :qstat, :qstat_df quit; data &outdata; fixedeffectsmean=&fixedmn; fixedeffectsvar=&fixedvar; qstat=&qstat; q_df=&qstat_df; q_pval=1-cdf("chisq",qstat,q_df); lower95_ci=fixedeffectsmean-1.96*sqrt(fixedeffectsvar); upper95_ci=fixedeffectsmean+1.96*sqrt(fixedeffectsvar); run; %mend;
9
%fixedeff(effect=lnor,var=varlnor,data=meta.preeclampsia)
proc print data=fixedresults; run;
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.