Topics Discussed during COVID Identified By Unsupervised Machine Learning Predict Distress Levels

We utilized Latent Dirichlet Allocation (LDA) to extract topics from open-ended responses to COVID-19 Exposure and Family Impact Scales (CEFIS). Seven emerged, and corresponding modal probabilities were used to predict stress at two time points. Time during pandemic and a topic associated with positive aspects had significant association with distress.