Design considerations in a sib-pair study of linkage for susceptibility loci in cancer

BACKGROUND: Modern approaches to identifying new genes associated with disease allow very fine analysis of associaton and can be performed in population based case-control studies. However, the sibpair design is still valuable because it requires few assumptions other than acceptably high penetrance to identify genetic loci. METHODS: We conducted simulation studies to assess the impact of design factors on relative efficiency for a linkage study of colorectal cancer. We considered two test statistics, one comparing the mean IBD probability in affected pairs to its null value of 0.5, and one comparing the mean IBD probabilities between affected and discordant pairs. We varied numbers of parents available, numbers of affected and unaffected siblings, reconstructing the genotype of an unavailable affected sibling by a spouse and offspring, and elimination of sibships where the proband carries a mutation at another locus. RESULTS: Power and efficiency were most affected by the number of affected sibs, the number of sib pairs genotyped, and the risk attributable to linked and unlinked loci. Genotyping unaffected siblings added little power for low penetrance models, but improved validity of tests when there was genetic heterogeneity and for multipoint testing. The efficiency of the concordant-only test was nearly always better than the concordant-discordant test. Replacement of an unavailable affected sibling by a spouse and offspring recovered some linkage information, particularly if several offspring were available. In multipoint analysis, the concordant-only test was showed a small anticonservative bias at 5 cM, while the multipoint concordant-discordant test was generally the most powerful test, and was not biased away from the null at 5 cM. CONCLUSION: Genotyping parents and unaffected siblings is useful for detecting genotyping errors and if allele frequencies are uncertain. If adequate allele frequency data are available, we suggest a single-point affecteds-only analysis for an initial scan, followed by a multipoint analysis of affected and unaffected members of all available sibships with additional markers around initial hits.