OBJECTIVES: Social media are increasingly used by patients and the content of these postings uniquely reflects the views and perspectives of patients. Historical discussions span several years, are updated in real time and are available for large numbers of patients. We examined the valence or sentiment of the language patients use over their treatment history utilising natural language processing (NLP) and sentiment analysis.

METHODS: We analysed a corpus of 8,433 postings from 483 Stage 4 RCC patients. Mean sentiment scores over time were computed for all patients for two distinct cases; Over follow-up from time since diagnosis and longitudinally over the 6-year period prior to 2016, regardless of time of diagnosis. The analysis of overall sentiment is presented with reference to a time-line representing regulatory approval of key targeted therapies.

RESULTS: Language used over time since diagnosis showed a positive sentiment with a mean value of +1.2 at time of diagnosis (range -5 for negative to +5 for postive), with a gradual decline toward neutral over time. Longitudinal mean sentiment change over time showed a distinct increase in positive sentiment (+0.8 to +1.4 over two years) occurring directly following the approval of 3 targeted therapies in 2010.

CONCLUSIONS: Sentiment is much used in other industries, although its use in medical fields is less well explored. The results showing an increase in mean sentiment following the approval of three targeted therapies in 2010 indicate that sentiment scoring can provide information regarding changes in the valence of language used over time, and potentially thereby quantifying a reaction to important events in a patient population. Patient mean sentiment remained positive over time since diagnosis, an overall positive sentiment may be a slightly counter intuitive finding (to non-patients) and is reflective of a predominance of positive wording being used by RCC patients.