The occurrence and progression of diseases constitute a complex dynamic process, and cross-sectional data from a single time point are insufficient to fully characterize their evolutionary patterns. In contrast, longitudinal time-series data can capture the dynamic evolution of diseases over time, which is key to achieving precise disease prediction. In recent years, breakthroughs in deep learning have driven significant progress in intelligent predictive models based on longitudinal time-series data. However, challenges inherent to such data-including irregular sampling, high missing rates, and heterogeneity-also impose specific requirements on model design. This paper provides a systematic review of research advances in intelligent models for disease prediction based on time-series data. We conduct an in-depth analysis from three perspectives: data processing and representation, representative model architectures, and time-series modeling strategies, aiming to provide a reference for building more efficient and robust longitudinal disease prediction models.