Hot flashes linked to linguistic markers of cognitive impairment: observational study of social media posts.

Résumé

OBJECTIVES: This study analyzes social media data from Reddit, using artificial intelligence and natural language processing, to explore cognitive changes in the menopause transition and their associations with hot flashes and hormone therapy (HT).METHODS: Posts from 2,387 women in the menopause subreddit and 3,710 matched controls (~340,000 posts) on Reddit were analyzed. A subset of 338 posts was manually labeled for age, hot flashes, cognitive complaints, mood disturbances, and sleep issues. A machine learning model (area under the receiver operating curve [AUC]: 0.67-0.92) applied these labels to the remaining posts. A cognitive scale, developed by correlating cognitive complaints with known linguistic markers of cognitive function, such as text length and lexical diversity, was employed.RESULTS: Hot flashes were reported by 41% of participants and were strongly associated with increased mood disturbances (6-fold), cognitive complaints (10-fold), and sleep difficulties (20-fold). Cognitive complaints correlated positively with length of text (AUC: 0.755) and negatively with unique word usage (AUC: 0.754). Women with hot flashes using HT initially exhibited lower cognitive scores, which improved with treatment over time but remained below those of asymptomatic untreated women.CONCLUSIONS: Hot flashes are associated with subjective cognitive impairment and with text markers of cognitive performance during the menopause transition. This study highlights social media's potential as a valuable research tool for assessing cognitive changes in menopause.