In clinical practice, the administration of drugs and hormones can result in specific sequelae, often stemming from intricate and multifaceted mechanisms. While Artificial Intelligence (AI) techniques hold great potential in medicine, their direct application in predicting clinical sequelae faces obstacles such as limited cross-sectional data, poorly defined medical data features, and unclear patient cohorts. To bridge the knowledge gap between physicians and Al models, we've forged a lasting partnership with a prominent local hospital. This collaboration centers on a co-design process, refining input features and defining patient samples. This process comprises two phases: an initial retrospective analysis followed by continuous Human-Al collaboration. Through real-world case studies, quantitative comparison experiments, and a comprehensive user study, we gather valuable insights for future human-AI collaborative systems in specialized medical data analysis. The results affirm the effectiveness of our co-design framework in expanding analysis capabilities, streamlining workflows, and enhancing communication between humans and Al.
