Our study evaluates the effectiveness of discriminative and generative models in protecting privacy and standardizing medical data, focusing on Sensitive Health Information (SHI) extraction and Time Information Normalization (TIN) in medical records. Utilizing advanced pre-trained models, we introduce a dual-model fusion with a voting mechanism, diverse data forms, and data augmentation techniques. Our experiments demonstrate the superiority of generative models over traditional discriminative models in the SHI extraction task. This paper underscores the significance of strategic model selection, adept data processing, and the novel voting mechanism in enhancing the performance of SHI extraction and TIN tasks. Our methods achieve remarkable performance both on SHI extraction and TIN tasks. These results pave the way for advanced AI applications in healthcare, offering valuable insights for ongoing research in medical data management and privacy safeguarding.