Generative AI models have transformed real-world applications, particularly in Natural Language Processing (NLP), enabling zero-shot inference across various tasks. They have also advanced zero-shot visual understanding, Agentic AI, and information retrieval. Unlike traditional models requiring task-specific fine-tuning, GenAI models handle complex NLP tasks without prior adjustments. This study leverages these capabilities for next-disease prediction, focusing on Chronic Kidney Disease (CKD). Using clinical note events from MIMIC-IV, we frame the task as a sequence classification problem, applying positive and negative sampling for binary classification to predict CKD occurrence based on a patient’s medical history. Our experiments compare traditional models like Logistic Regression and BERT with open-source LLMs of varying sizes and architectures. We also assess different fine-tuning methods to evaluate model generalizability. Larger models, particularly Mistral-Small-24B with supervised fine-tuning (SFT), achieved the highest performance, significantly outperforming other models and traditional baselines.