<p>Investigating the causal relationships between characteristics (e.g., lifestyles or intervention strategies of patients) and the resulting expressions (e.g., physiological symptoms) play a critical role in healthcare analytics. Effective synthesis for the resulting expressions using given characteristics can make great contributions to health risk management and medical decision-making. Therefore, the objective of this study is to effectively synthesize the expressions based on given characteristics for healthcare analytics. However, there are two major challenges: (1) compared with expressions, the characteristics are usually relatively low dimensional, but most of the existing methods such as regression models are not able to effectively handle such mappings; and (2) the expressions may vary even the characteristics remain the same, as the relationship between characteristics and expressions may contain both deterministic and stochastic patterns. To address these challenges, this study proposed a generative adversarial network (GAN)-based approach, termed selective ensemble characteristic-to-expression synthesis (SE-CTES). The novelty of the proposed method can be summarized into three aspects: (1) a GAN-based architecture is leveraged to learn the mapping from relatively low dimensional space to relatively high dimensional space containing both deterministic and stochastic patterns; (2) the involved incorrect mapping information in the GAN-based architecture is better utilized by weight calibration, to reduce the estimation bias to the data distribution; and (3) a selective ensemble learning framework is proposed to further improve the synthesis stability and accuracy. To validate the effectiveness of the proposed approach, extensive numerical simulation studies and a real-world healthcare case study for cardiovascular disease (CVD) research were conducted, and the results demonstrated the great potential of the proposed SE-CTES method.</p>