使用for循环存储荟萃分析结果

Question

I am trying to conduct a meta analysis of a group of studies. I am trying to figure out the code to properly output what I would like. Essentially I am trying to run a fixed effects meta-analysis test using the metafor package and collect the coefficient estimates to store them inside a matrix.

I have several problems. For example, I want to run these tests only for research that have multiple results. So studies whose study numbers turn up more than once (see code for example). When I try this using my existing code, it does not work, it spits out a number, but it isn't the right one. Also, some study-numbers are larger than the actual amount of studies I have. In my personal dataset there is a study numbered 3500 for example. When I run my loop, R spits out the results for that particular fixed effects model on the 3500th row, instead of just placing it in the next empty row.

I have a basic example below that anyone can run in R.

library(metafor)
origdata <- data.frame(matrix(data=NA, nrow=15, ncol=3))
colnames(origdata) <- c("studynum", "Mail_b", "Mail_SE")
origdata$studynum <- c(1, 1, 1, 1, 1, 2, 3, 3, 3, 4, 7, 7, 7, 7, 7)
origdata$Mail_b <- c(1.8, 0.8, 1.2, 1, 1, 5, 3, 3, 6, 4, 5, 8, 5, 9, 2)
origdata$Mail_SE <- c(1.6, 0.8, 1.3, 1, 1, 1, 3, 2.9, 6, 4, 5, 8, 5, 8, 1)  

collapsedtest <- data.frame(matrix(data=NA, nrow=5, ncol=3))
colnames(collapsedtest) <- c("studynum", "Meta_b", "Meta_SE")
collapsedtest$studynum <- unique(origdata$studynum)

for(i in unique(origdata$studynum)) {
  if((table(origdata$studynum) == 1) == FALSE){
    collapsedtest[i, 2] <- (coef(summary(rma(yi=Mail_b[origdata$studynum == i], 
                                             sei=Mail_SE[origdata$studynum == i], 
                                             method="FE", 
                                             data=origdata)))$estimate)
    collapsedtest[i, 3] <- (coef(summary(rma(yi=Mail_b[origdata$studynum == i ], 
                                             sei=Mail_SE[origdata$studynum == i], 
                                             method="FE", data=origdata)))$ci.ub 
                            -
                              (coef(summary(rma(yi=Mail_b[origdata$studynum == i], 
                                                sei=Mail_SE[origdata$studynum == i], 
                                                method="FE", 
                                                data=origdata)))$estimate)) / 1.96
    } else {
      collapsed[i, 2] <- origdata$Mail_b[origdata$studynum == i]
      collapsed[i, 3] <- origdata$Mail_SE[origdata$studynum == i]
    }
}

Answer 1

Consider using nrow and adjust your row indexer with a conditional on studynum . Also, be sure to run your rma method only once and then use resulting estimates.

for(i in unique(origdata$studynum)) {
  coef_data <- coef(summary(rma(yi=Mail_b[origdata$studynum == i], 
                                sei=Mail_SE[origdata$studynum == i], 
                                method="FE", 
                                data=origdata)))

  if(nrow(origdata[origdata$studynum==i,]) > 1) {
    collapsedtest[collapsedtest$studynum==i, 2] <- coef_data$estimate
    collapsedtest[collapsedtest$studynum==i, 3] <- (coef_data$$ci.ub - coef_data$estimate) / 1.96

    collapsedtest[collapsedtest$studynum==i, 2] <- origdata$Mail_b[origdata$studynum == i]
    collapsedtest[collapsedtest$studynum==i, 3] <- origdata$Mail_SE[origdata$studynum == i]
  }
}

However, you can avoid the studynum filters by using by , the method that subsets a dataframe by factor(s) and runs subsets into an operation. Then after processing, row bind all subsetted dataframes with do.call :

df_list <- by(origdata, origdata$studynum, function(sub){
  coef_data <- coef(summary(rma(yi=Mail_b, sei=Mail_SE, method="FE", data=origdata)))

  if(nrow(sub) > 1) {
     df <- data.frame(studynum = sub$studynum[[1]], 
                      Meta_b = coef_data$estimate,
                      Meta_SE = (coef_data$$ci.ub - coef_data$estimate) / 1.96
  } else {
     df <- data.frame(studynum = sub$studynum[[1]], 
                      Meta_b = sub$Mail_b,
                      Meta_SE = sub$Mail_SE)
  }

  return(df)
})

collapsedtest <- do.call(rbind, df_list)