The invention provides a medical text-oriented patient privacy information desensitization method, which comprises the following steps of: extracting a patient medical text, and dividing patient privacy into three types according to all appearing patient privacy data, namely a digital type, a date type and a named entity type; designing a set of regular expressions for the digital privacy data, and removing digital information including patient ages, identity card numbers, contact information, postal codes, patient hospitalization numbers and the like; designing a random migration algorithm for the date type privacy data, identifying different types of date data, and migrating the date to a certain time of several future century; adopting a hidden Markov model for the named entity type privacy data, removing privacy information such as the patient name, the doctor name and the hospital name. According to the method, the personal sensitive information of the patient contained in the medical text is effectively removed, the privacy security of the patient in the medical data publishing process is guaranteed, and the method has important significance in medical data sharing and regional medical and health information big data platform construction.