Background: Observational cohort studies have been little used in linkage analyses due to their general lack of large, disease-specific pedigrees. Nevertheless, the longitudinal nature of such studies makes them potentially valuable for assessing the linkage between genotypes and temporal trends in phenotypes. The repeated phenotype measures in cohort studies (i.e., across time), however, can have extensive missing information. Existing methods for handling missing data in observational studies may decrease efficiency, introduce biases, and give spurious results. The impact of such methods when undertaking linkage analysis of cohort studies is unclear. Therefore, we compare here six methods of imputing missing repeated phenotypes on results from genome-wide linkage analyses of four quantitative traits from the Framingham Heart Study cohort.
Results: We found that simply deleting observations with missing values gave many more nominally statistically significant linkages than the other five approaches. Among the latter, those with similar underlying methodology (i.e., imputation- versus model-based) gave the most consistent results, although some discrepancies remained.
Conclusion: Different methods for addressing missing values in linkage analyses of cohort studies can give substantially diverse results, and must be carefully considered to protect against biases and spurious findings.