Meta-analysis of behavioral genetic studies would provide (i) tighter confidence intervals around parameter estimates, (ii) clarification of apparently discrepant study findings, and (iii) a mechanism for analyzing systematic causes of between-study differences. We examined some key issues that arise in the meta-analysis of categorical phenotypes. Data were simulated under a multifactorial threshold model that assumed an underlying normal liability distribution, and summary statistics (probandwise concordance rate, recurrence risk ratio, odds ratio, kappa) compared for given values of the liability correlation between relatives and given population prevalence. Although the odds ratio and kappa statistic performed well at moderate to high values of the population prevalence (15-50%), at low values all of these statistics were sensitive to overall prevalence. In cases where the assumption of a multifactorial threshold model is reasonable, direct estimation of genetic and environmental parameters from the summary statistics from all studies appears to be a preferable strategy. For cases where data from non-randomly ascertained samples are used, the impact of misspecification of the model for ascertainment was examined. For some parameter values, such misspecifications led to quite serious biases to estimates of genetic and environmental parameters. These biases varied in complex ways as a function of research design and of the true causes of variation in the population, so that the same misspecification could lead to an overestimate of the importance of genetic influences in twin data but an underestimation in adoption data or to an overestimate of the importance of genetic effects from twin data if shared environmental as well as genetic influences were simulated but an underestimate of genetic effects if shared environmental effects were assumed unimportant. These complexities emphasize the importance of being sensitive to the effects of misspecifying ascertainment corrections in any meta-analysis of behavioral genetic data.
- Categorical phenotypes
- Multifactorial threshold model