The study of gene-environment interactions is an increasingly important aspect of genetic epidemiological investigation. Family-based study designs hold many advantages over samples of unrelated individuals, including quality control checks, imputation of ungenotyped samples and within-family contrasts to minimize confounding due to population stratification without additional genotyping. The QBATI [1] estimates and tests for a gene-environment interaction in families of arbitrary structure that are sampled independent of the phenotype of interest. However, it is vulnerable to inflated type I error if ascertainment is informative of the phenotypes. In this study, we investigated the type I error rate of the QBATI in non-randomly ascertained samples. We observed an elevated type I error rate in samples ascertained on the trait of interest when there is a main effect of the genetic factor of interest. We propose three corrective methods that allow use of the QBATI to test for gene-environment interactions in non-randomly ascertained samples. Our results indicate that the methods we propose can often restore the nominal type I error rate. In addition, our corrective methods allow the QBATI to retain good power to detect gene-environment interactions in non-randomly ascertained pedigrees. We also investigated the potential for confounding by ancestry when using the QBATI, and propose a correction method.