OBJECTIVES: Every day, people share health-related messages and opinions online. Understanding and analyzing these insights could be useful for medical applications. Quality of life is medically defined as individuals’ perception of their place in life, in the cultural context and the value system in which they live, according to their objectives, standards and concerns. Mining social medias to understand the burden of pathologies and treatments is a new approach to better understand patients’ concerns and unmet needs. This study proposes a NLP algorithm able to detect health-related quality of life (HRQoL) impacts on social media posts.

METHODS: Messages relating a patient experience with a drug or disease were collected on French social medias, using a Web crawler. Two reviewers manually annotated messages following guidelines based on EQ-5D and SF-36 QoL measurements. Focus was put on physical, psychic, activity, relational and financial dimension of HRQoL. A first model detected impact presence. Then each impact was identified by a specific model. Impact’s expressions were used to generate features. Other features were based on the message content, such as expressed sentiment, grammar and conjugation.

RESULTS: Training set represented 1000 posts related to pathologies and 400 to treatments. Extreme gradient boosting was the chosen model for both impact detection and specific dimension identification. Models were trained using cross-validation and hyperparameter optimization. Over-sampling was used to augment infrequent dimensions. This allowed us to detect a general impact with a sensibility of 0.8 and specificity of 0.7, and physical (0.56 ; 0.857), psychic (0.58 ; 0.828), activity (0.71 ; 0.79), relational (0.675 ; 0.73),and financial (0.77 ; 0.814) dimensions.

CONCLUSIONS: We developed an algorithm, based on medically validated questionnaires, able to identify impacts of HRQoL in online patient’s messages. Social media studies could be a new source to understand how diseases and therapies represent a burden to patients.