Title: Automated Screening of Expressed Emotion
Legend: Receiver operating characteristic (ROC) curves show the performance of classifiers in predicting overall expressed emotion in A) mothers of children with schizophrenia (Wave 1 of data collection); B) mothers of children with schizophrenia (Wave 2 of data collection); C) mothers of children with autism spectrum disorder. The area under the ROC curve (AUROC) shows the level of success in classification, reflecting the false-positive rate versus the true-positive rate resulting from the classification task. If AUROC is equal to 1, it means the classifier was able to successfully assign all of the cases to the correct class. Our random forest classifier was able to successfully, identify individuals with high level of overall expressed emotion form those who were classified to have lower level of overall expressed emotion. Our fully automated machine learning approach is fast, cost effective and efficient.
Citation: Movaghar, A., Mailick, M., Page, D., Saha, K., Rynn, M., Greenberg, J. Automated screening of expressed emotion, Under review.
Abstract: Expressed emotion (EE) is a measure of the family’s emotional climate and has been shown to be associated with a range of symptoms and psychiatric outcomes in individuals with various disabilities. In addition, growing evidence suggests that high levels of family distress are associated with high EE. Rapid assessment of EE has great clinical relevance because evidenced-based family psychoeducation treatments have been shown to be effective in improving the quality of life of families and patients, in part by reducing levels of EE. Here we present a novel, fully automatic method to replace the time-consuming, cumbersome and costly process of evaluating expressed emotion manually from long interviews; the new method relies on natural language processing and machine learning approaches. By analyzing five-minute speech samples, we were able to successfully characterize the emotional climate in 631 families who had children with schizophrenia or autism spectrum disorder. Use of computational techniques will significantly reduce resources required for measuring EE and potentially could be used for “real time” screening in clinics.
About the Lab: The Lifespan Family Research Program is dedicated to the advancement of knowledge about families who have a member with a disability, with a special emphasis on how these families change over the lifespan. Our program of research encompasses longitudinal and population-based studies of autism, fragile X syndrome, schizophrenia, other developmental disabilities and severe mental illnesses. In addition we study parents as caregivers to children with disability and develop evidence-based interventions for affected families.