Multilevel models are widely used in genome-wide association studies (GWAS) to account for population structure and relatedness. To study diseases that aggregate in the families of patients such as Alzheimer's disease, many genetic epidemiological studies recruit independent study participants, often referred to as probands, and collect information on their relatives through a family health history interview of the probands. However, dense genotypes are often collected only in probands but not their family members due to the high cost of in-person collection of blood samples or inaccessibility. To overcome this problem, we develop a regression model for genome-wide analysis of binary traits or age-at-onset traits that control confounding due to familial relatedness by using available family data. This approach adjust for the unobserved polygenic effects as well as the shared non-genetic familial effects estimated from the multilevel model by using the available pedigree information. We show in real data analysis that our method effectively controls for the confounding and achieves similar or better performance compared to using principal components analysis adjustment.