Adapting natural language processing and sentiment analysis methods for an intervention in older adults: Positive perceptions of health and technology.

Abstract

Older adults frequently participate in behavior change studies, yet it is not clear how to quantify a potential relationship between their perception of the intervention and its efficacy. We assessed the relationship between participant sentiment toward the intervention from follow-up interviews with physical activity and questionnaires for the perception of health. Sentiment was calculated using the transcripts of exit interviews through a bag of words approach defined as the sum of positive and negative words in 28 older adults with obesity (body mass index ≥30kg/m2). Mean age was 73 years (82% female), and 54% lost ≥5% weight loss. Through linear regression we describe a significant association between positive sentiment about the intervention and weight loss; positive sentiment on technology and change in PROMIS-10 physical health and reduced physical activity time, while controlling for sex and age. This analysis demonstrates that sentiment analysis and natural language processing in program review identified an association between perception and topics with clinical outcomes.