Evaluation of a bayesian model integration-based method for censored data.

OBJECTIVE Non-random missing data can adversely affect family-based linkage detection through loss of power and possible introduction of bias depending on how censoring is modeled. We examined the statistical properties of a previously proposed quantitative trait threshold (QTT) model developed for when censored data can be reasonably inferred to be beyond an unknown threshold. METHODS The QTT model is a Bayesian model integration approach implemented in the PPL framework that requires neither specification of the threshold nor imputation of the missing data. This model was evaluated under a range of simulated data sets and compared to other methods with missing data imputed. RESULTS Across the simulated conditions, the addition of a threshold parameter did not change the PPL's properties relative to quantitative trait analysis on non-censored data except for a slight reduction in the average PPL as a reflection of the lowered information content due to censoring. This remained the case for non-normally distributed data and extreme sampling of pedigrees. CONCLUSIONS Overall, the QTT model showed the smallest loss of linkage information relative to alternative approaches and therefore provides a unique analysis tool that obviates the need for ad hoc imputation of censored data in gene mapping studies.