This research delves into the intersection of big data and healthcare, particularly focusing on the critical task of predicting in-hospital mortality events. Leveraging Apache Spark’s capabilities alongside electronic medical records and advanced modeling techniques like Latent Dirichlet Allocation (LDA), the study explores a time-varying approach to improve prediction accuracy. The paper showcases a transition from the MIMIC II to MIMIC III dataset, emphasizing inclusivity in patient selection and utilizing logistic regression for model interpretability. Through rigorous experimentation and evaluation, the results demonstrate promising advancements in real-time medical predictions, highlighting the potential of clinical notes as invaluable resources for augmenting healthcare outcomes.