Rejoinder on “ascertainment adjustment in complex diseases”

✍ Scribed by David V. Glidden

• Publisher: John Wiley and Sons
• Year: 2002
• Tongue: English
• Weight: 47 KB
• Volume: 23
• Category: Article
• ISSN: 0741-0395
• DOI: 10.1002/gepi.10198

✦ Synopsis

The discussions of and add considerable insight to the topic of ascertainment adjustment in genetic variance component models. The authors are in agreement on a broad range of issues, i.e., that the problem of ascertainment bias in the presence of unmeasured heterogeneity is compelling and more complex than in the classic paradigm; that correction in this context raises conceptual issues of a sample which differs systematically from the broad population; that ascertainment correction in this setting is theoretically straightforward but requires precise specification of 1) the ascertainment scheme and 2) the parametric form for the mixing distribution; and that slight misspecification of either will give markedly biased parameter estimates. However, the discussions of Burton [2002a] and point out intriguing lines for further development and discussion.

Further simulation studies should be undertaken to extend results on the effect of mixing distribution misspecification. Glidden and Liang [2002] performed simulations in GLMMs without individual covariate effects. These covariate effects may be more robust to misspecification of mixing distributions. Consider the model given in formula (1) of Glidden and Liang [2002]. It is possible to obtain a consistent estimate of b (but not a or s c ) from a so-called conditional likelihood. The conditional likelihood approach allows for latent heterogeneity without assuming a particular parametric form. The estimation issues are quite different for a and s c compared with b. It is possible that misspecification of the mixing distribution in this setting may not affect b as markedly. Simulations along these lines are required.

Epstein [2002] points out new problems that may occur in ascertainment correction, even when they are appropriately specified.