<p>Data heterogeneity and the multi-source nature of modern datasets present significant challenges for statistical methodologies. Motivated by the Stimulant Reduction Intervention using Dosed Exercise (STRIDE) study, we analyze primary outcomes from two complementary data sources to assess heterogeneity in treatment effects between experimental and control groups at medium and high quantiles. To address these challenges, we propose a novel approach that integrates quantile regression with weighted quantile loss and joint pairwise fusion penalties, enabling joint subgroup identification across data sources. Our method distinguishes between homogeneous and heterogeneous effects using a center regularization term and can detect sources lacking group structures. Theoretically, we establish weak Oracle properties, ensuring consistent estimation of group structures. Computationally, we employ the alternating direction method of multipliers (ADMM) and mitigate the burden of pairwise fusion through a k-nearest neighbors trimming method. The effectiveness of our approach is demonstrated through numerical simulations and an application to the STRIDE study.</p>