Gender-related stereotypes and biases can have severe consequences in the medical domain, especially in mental health therapy. In this study, we analyzed 91 psychotherapy transcripts from the Alexander Street database to investigate whether gender-related stereotypes differ in the treatment of patients by male versus female therapists using natural language processing and statistical analyses. We built a lexicon of ten high-level categories that capture sentence-level attributes and represent gender-related stereotypes. Our results suggest significant statistical differences in categories such as active, negatives, positives, etc., during the treatment of female patients by male therapists as compared to female therapists. We built logistic regression models using the ten high-level lexical categories to predict the gender of the therapist. We also provide recommendations on how our analytical methods can be used, along with other advanced deep-learning methods, to detect and reduce gender-related stereotypes in psychotherapy sessions.

Abstract Only