The medical information extraction task plays a crucial role in assisting doctors by extracting patient-related information from clinical cases. Detailed processing of the extracted data enables the identification of key information within medical records. Through comprehensive analysis, the model can determine the specific disease affecting the patient, thereby providing valuable reference information to clinicians and alleviating their workload. In this paper, we present a medical information extraction and classification framework developed for an open competition task at the 10th Conference on Health Information Processing (CHIP 2024). This task involves processing complex and information-dense patient case texts to extract the specific lymphoma diagnosis and subsequently generate tumor codes based on the extracted lymphoma information and other case details. We propose a large model-based framework for information extraction and classification. To tailor the model to this medical task, we fine-tune it using an augmented dataset. To facilitate accurate extraction of information relevant to code generation, we employ a chain-of-thought (CoT) approach, guiding the model to incrementally extract key information.  Given the inherent challenges in generative tasks, such as the risk of producing similar but incorrect codes, we transform the generative task into a classification task. Specifically, the model is tasked with selecting the correct code from a set of possible options, thus improving accuracy. Furthermore, we carefully design and optimize prompt strategies to guide the model in producing reliable outputs. Our experimental results demonstrate that these methodologies enable the model to achieve strong performance in this task.