Power and type I errors of meta-analysis of single variant tests in the presence of missing data for continuous outcomes.

Datasets were simulated according to the genetic and phenotype model described in METHODS. Meta-analysis was performed to combine 20 cohorts with 1500 individuals each. For each replicate, summary association statistics were generated, and a certain fraction of the generated summary statistics were masked as missing. Scenarios with different combinations of known variant effects, candidate variant effects and fractions of missingness were considered. Six analysis strategies were considered: 1) PCBS; 2) SYN+; 3) ImpG+meta; 4) COJO; 5) DISCARD and 6) REPLACE0. Type I error and power were evaluated using 105 replicates under the significance threshold of α = 0.005.