Consequences of multiple imputation of missing standard deviations and sample sizes in meta-analysis