Using Total Sample Size Weights in Meta-Analysis of Log-Odds Ratios |
| |
Authors: | Sunyoung Park S Natasha Beretvas |
| |
Institution: | University of Texas at Austin, Austin, Texas |
| |
Abstract: | The log-odds ratio (lnOR]) is commonly used to quantify treatments' effects on dichotomous outcomes and then pooled across studies using inverse-variance (1/v) weights. Calculation of the lnOR]'s variance requires four cell frequencies for two groups crossed with values for dichotomous outcomes. While primary studies report the total sample size (n..), many do not report all four frequencies. Using real data, we demonstrated pooling of lnOR]s using n.. versus 1/v weights. In a simulation study we compared two weighting approaches under several conditions. Efficiency and Type I error rates for 1/v versus n.. weights used to pool lnOR] estimates depended on sample size and the percent of studies missing cell frequencies. Results are discussed and guidelines for applied meta-analysts are provided. |
| |
Keywords: | Meta-analysis odds ratio robustness variance estimation weights |
|