Purpose/Significance To mine the deep information of the diagnosis and treatment experience of patients with chronic diseases, and to provide references for optimizing the hierarchical diagnosis and treatment policy. Method/Process The interview texts of 539 patients with chronic diseases from 4 hospitals in Guangdong province and Chongqing are collected. Topic identification and sentiment analysis are conducted based on the BERT-base-Chinese model. Result/Conclusion A total of 4 234 question-answer pairs are analyzed, from which 11 topic categories are identified. Topics related to medical services and institutions account for the highest proportion (37.72%), followed by the appointment and treatment process (27.23%) and health education (10.75%). Sentiment analysis reveals that positive emotions toward topics such as “medical staff” and “appointment registration,” while “referral” and “financial burden” elicits negative sentiments. Machine learning methods can effectively identify the shortcomings in chronic disease services.