A study of personal health information posted online: Using machine learning to validate the importance of the terms detected by MedDRA and SNOMED in revealing health information in social media

With the increasing amount of personal information that is shared on social networks, it is possible that the users might inadvertently reveal some personal health information. In this work, we show that personal health information can be detected and, if necessary, protected. We present empirical support for this hypothesis, and furthermore we show how two existing well-known electronic medical resources MedDRA and SNOMED help to detect personal health information (PHI) in messages retrieved from a social network site, MySpace. We introduce a new measure - risk factor of personal information - that assesses the likelihood that a term would reveal personal health information. We synthesize a profile of a potential PHI leak in a social network, and we demonstrate that this task benefits from the emphasis on the MedDRA and SNOMED terms. Our study findings are robust in detecting sentences and phrases that contain users’ personal health information.