Machine learning for longitudinal applications of neuropsychological testing

Abstract Psychiatric patients, such as those suffering from depression or schizophrenia, often need to be monitored with frequent clinical interviews by trained professionals to avoid costly emergency care and preventable events. However, there simply are not enough clinicians to monitor these patients on a regular basis. Furthermore, infrequent clinical evaluations may result in clinicians missing subtle changes in patient state that occur over time. These limitations can affect both the quality, timeliness, and monetary expense of treatment. Therefore, we leveraged smart devices to implement traditional neuropsychological assessments such that they could be collected frequently, remotely, and - when viable - self-administered by the participants themselves. This approach enables the generation of an enormous quantity of data across time and different assessments. Machine learning-based methods hold the potential to automatically analyze streams of behavioral and cognitive data, such as speech and movement, and convert them to actionable events. We examined the viability of the automation of a comprehensive assessment pipeline, from administration of neuropsychological tests, to transcription of spoken responses, to an analysis of data to predict clinical states. In the present research, we examined this pipeline in 353 participants (of whom 134 were patients with a range of diagnoses of psychosis spectrum disorders, substance abuse disorders, and affective disorders, and 219 were non-patient volunteers who were presumed to be healthy). We found that machine learning-based methods can be applied to this data in order to reliably and accurately assess the neuropsychological function of individuals. Among other applications, we were able to automatically score completion of a verbal recall task and predict emotional state via spoken language, thereby opening the potential for regular, frequent analyses of cognitive and mental states.