Quantifying the Processes and Events of Psychotherapy at Scale

Abstract

Abstract Background In the wake of the COVID-19 pandemic telemedicine usage increased in the United States, especially in the field of mental health. The study aims to demonstrate the feasibility of collecting recordings of telemedicine psychotherapy, relevant electronic health records (EHR), and matched real-world data to create an aligned, multimodal dataset. We examine possible ways to use this dataset to train machine learning models, intending to explore the creation of tools that could assist psychotherapists. Methods This study was conducted through an outpatient, telemedicine-enabled, clinic in New York City. Participants were recruited from the existing treatment population and were already undergoing psychotherapy. After participants provided informed consent, each subsequent psychotherapy session was recorded, however, a participant could request that any individual session not be recorded without impact on study participation. Only sessions that occurred via telehealth were eligible for recording. This study also collected participants’ electronic health record (EHR) data from the study clinic as well as participants’ de-identified real-world data from aggregated records providers using a tokenized de-identification process provided by a third-party organization. Results We successfully collected 34 psychotherapy session recordings from 19 participants across seven different providers as well as EHR and other real-world health data from all participants. Preliminary machine learning analyses were applied to the data, and a further plan for data analysis is discussed. Conclusion Establishing this unique dataset is the first step to developing machine learning tools that can assist psychotherapists in their practice. This study demonstrates the feasibility of collecting more data of this nature, illustrates potential analyses that can be applied to the data, and how they may be used to help improve psychotherapy.