Online health forums provide a large repository for patients, caregivers, and researchers to seek valuable information. The extraction of patient-reported personal health experience from the forums has many important applications. For example, medical researchers can discover trustable knowledge from the extracted experience. Patients can search for peers with similar experience and connect with them. In this paper, we model the extraction of patient experience as a classification problem: classifying each sentence in a forum post as containing patient experience or not containing patient experience. We propose to exploit the sentence context information for such experience extraction task, and classify the context information into global context and local context. A unified Context-Aware Experience Extraction (CARE) framework is proposed to incorporate these two types of context information. Our experimental results show that the global context can significantly improve the experience extraction accuracy, while the local context can also improve the performance when less labeled data is available.