Outcome-dependent sampling provides a cost-effective yet powerful strategy to perform sequencing studies of quantitative traits. When the number of individuals that can be genotyped is limited, such sampling design can generate good power to detect associations between genetic variants and the trait. However, failure to account for the biased nature of the sampling can produce inflated type I error and loss of power for both the analysis of primary and secondary traits, especially when using meta-analysis to combine results from multiple studies with different selection criteria. Commonly used methods designed for random sampling design are not equipped to properly account for the non-random sampling scheme. Here we review the problems with naive approaches and propose an alternative likelihood-based approach that accounts for the biased sampling design. We illustrate our approach through simulation studies and demonstrate through simulation and real data analysis that our proposed approach maintains correct type I error and provide a powerful alternative to perform association analysis under outcome dependent sampling design.