<p>When longitudinal data with individual-specific endpoints are observed, we treat them as functional data while accounting for differences in endpoints. We call this type of data variable-domain functional data. We propose a method that simultaneously performs model estimation and variable selection in a logistic regression model where the predictors are multiple variable-domain functional data. Coefficient functions in the model depend on both time and the endpoints for each individual. They are estimated using a penalized likelihood method with sparsity-inducing penalties that enable variable selection. We conduct simulation studies to investigate the effectiveness of the proposed method. The proposed method is applied to the analysis of gene expression data from patients with autoimmune diseases to detect genes associated with the treatment response to the disease.</p>