Predicting the trajectory of a disease at an early stage can aid physicians in offering effective treatment and prompt care to patients. Previous work has used Electronic Health Record (EHR) data and leveraged sequence models to forecast patients’ diagnosis, procedure, and medication codes. However, current deep learning models have difficulty learning from such EHR data—which comprises multivariate time series and multimodal data distribution. We propose a novel method called Clinical-GAN to tackle this learning issue, which combines Transformer and Generative Adversarial Networks (GAN) to forecast diagnosis, procedure, and prescription drugs while maintaining the interpretability of the model’s outcome. A Transformer mechanism is used as a Generator to learn from existing patients’ medical history and is trained adversarially against a Transformer-based Discriminator. In addition, we used multi-head attention of the Generator network to explain the model’s outcome. We evaluated our method using a publicly available dataset, Medical Information Mart for Intensive Care IV (MIMIC-IV) v1.0, with more than 500,000 visits completed by around 196,000 adult patients over an 11 year period from 2008-2019. Based on the patient’s medical codes of each visit, Clinical-GAN achieved 57.04% in Mean Average Recall (MAR)@250 and 76.57% in Mean Average Precision (MAP)@250, significantly outperforming baseline methods in forecasting the patient’s diagnosis codes for subsequent visits. In addition, our model also outperformed the existing work in sequential disease prediction by achieving a 65.27% in MAR@60. We used cosine similarity to calculate the top 20 associated codes of diagnosis based on the learned representation of embedded medical codes. Then we projected them onto a two-dimensional map using t-Distributed Stochastic Neighbour Embedding (t-SNE). We observed that our proposed method effectively learns the correlations between medical codes from the two-dimensional map. Overall, Clinical-GAN has achieved higher accuracy in forecasting the patient trajectory and sequential disease prediction than baseline and existing work, as demonstrated through various experiments.