Dr. Alex Cohen, clinical psychologist and professor at Louisiana State University (LSU), continues his efforts to adapt behavioral technologies for investigating a wide range of clinical issues, including suicidality, depression, psychosis, mania, and anxiety. His newest publication is “High Predictive Accuracy of Negative Schizotypy with Acoustic Measures,” published recently in the flagship clinical journal, Clinical Psychological Science. He is widely recognized for his work using automated computerized analysis of behavior and has been featured in top psychology and psychiatry journals.
In an interview with the Times, Dr. Cohen said, “I think we are pushing the boundaries of what clinical science can do in measuring symptoms of serious mental illness.
“Psychology has not sufficiently addressed many areas of human suffering, and innovation is needed. I believe that Psychology can’t fulfill its potential alone, and will require cooperation between other academic disciplines, and also community partners, big tech, advocacy and political groups, business, government regulators, law enforcement and above all, people from the communities we are serving,” he said.
“Finding ways to cooperate and overcome the inevitable ‘tower of babel’ problem between these groups, in my opinion, is essential to solving many of the big problems that we face right now. Who else is trained so effectively in bringing people together? I think psychology can occupy a central role in coordinating these efforts.”
What does he think are the major and most important findings of this new study? “We were trying to use objective vocal data to predict personality traits associated with psychosis risk,” Dr. Cohen said. “Given the nature of our data, we used supervised machine learning. Our models were highly accurate, generally 85% or so in classifying people with versus without the traits.
“More importantly, we were exploring how this kind of model building should be done, and this problem extends well beyond psychosis risk research. Our models didn’t actually predict personality traits or psychosis risk, but rather, people’s report on ‘gold-standard’ self-report scales. Predicting psychosis risk and predicting scores from a gold-standard measure are not the same, and our secondary analyses speak to this.
“Generally speaking, ‘gold standard’ measures are good enough for many purposes in psychology. If the goal is highly accurate prediction using objective data however, our measures are often inadequate. This is an unrecognized obstacle to implementing predictive analytics into psychology,” Dr. Cohen said.
This most recent work was a collaboration with the LSU Department of Psychology, the LSU Center for Computation and Technology, the Department of Psychiatry at University of Utah, and Department of Psychology at University of Central Florida.
Dr. Cohen worked with Dr. Christopher Cox on this project, an Assistant Professor of Psychology at LSU. Dr. Cox is involved in various research endeavors, including focusing on experimental machine learning tools, exploring the context sensitivity of semantic knowledge, building computational models of reading.
What was it like collaborating with Dr. Cox? “Dr. Cox is one of the most thoughtful people I have had the pleasure to work with,” said Dr. Cohen. “He is extremely bright and methodical, and cares deeply about students and learning. He seems to operate on a higher level of consciousness than most, and it wouldn’t surprise me if he sees the world in streams of binary data like Neo from the Matrix movies.”
Dr. Cohen is also an adjunct professor at Pennington Biomedical Research Center and LSU Health Sciences, where he manages a team of doctoral students and graduate assistants. His current research projects focus on understanding and helping those with severe mental illness, notably schizophrenia, and those at risk of developing various psychotic-spectrum disorders.
Dr. Cohen’s current research projects are multi-tiered. He is currently working on a project that involves adapting biobehavioral technologies for use in assessing mental well-being. This project involves a highly constructed collaboration between industry and academia and uses “Big Data” methods to measure and predict cognitive, affective, and behavioral states in those with serious mental illness.
A few years ago, LSU helped Dr. Cohen and some of his colleagues to commercialize his technologies for “digital phenotyping.”
“Digital phenotyping involves quantifying aspects of mental health using complex, objective data streams,” he said. “In our case, these data are from automated language, facial, vocal, location and movement analysis from a smart phone. Since then, we have created an app using these technologies to support clinical trials. We are starting to explore digital phenotyping to support clinical management of patients with serious mental illness, and I am proud to have community partners in Baton Rouge for this. The methods used in our clinical psychological science paper were central in advancing these technologies.”
Dr. Cohen is in collaboration with an international consortium involved in researching the links between disturbances in natural speech and symptoms of mental illness and genomics. Pattern recognition and advanced machine learning are being utilized in this research. In addition to these projects, he facilitates research investigating how emotion, cognition, motivation, and social functions in those predisposed for developing serious mental illness and those already combating serious mental illness. This project uses “small data” and basic psychological science methods are used, including self-report, behavioral and electrophysiological measures, and performance measures.
Dr. Cohen has been working for nearly 20 years on these innovations, and explains that with the help of many colleagues, “… we are getting closer – though this process has been anything but time efficient.”
“What I have found is that digital data and symptoms ratings rarely agree,” he said. “Using machine learning, one can engineer solutions that show impressive agreement in one setting, but they don’t generalize. What is considered flat and unresponsive speech in one setting by one group of people is considered unremarkable in another setting for other people. That is one major thing we found in the CPS paper, and have replicated in a number of other studies.”
“Why don’t they agree? Are clinicians wrong? Are digital technologies missing a critical human element? The answer is, of course, both. So we are trying to develop methods for optimizing and evaluating these digital technologies. This field is huge right now, but I am afraid many of the solutions being proposed are superficial and will fade quickly. I think my colleagues and I are in a unique position to advance this field.”
Some of Dr. Cohen’s recent work helps to explain these complexities. • Cohen, A. S., Rodriguez, Zachary Warren, K. K., Cowan, T. M., Masucci, M. M., Granrud, Ole Edvard Holmlund, Terje B Chandler, C., Foltz, P. W., & Strauss, Gregory, P. (2022). Natural Language Processing and Psychosis: On The Need for Comprehensive Psychometric Evaluation. Schizophrenia Bulletin, In Press.
“Evaluation of digital measures falls far short of what is expected of most psychological tests,” Dr. Cohen said. “This is part of a themed issue Brita Elvevåg and I are finalizing for the journal Schizophrenia Bulletin.”
• Cohen, A. S., Cox, C. R., Tucker, R. P., Mitchell, K. R., Schwartz, E. K., Le, T. P., Foltz, P. W., Holmlund, T. B., & Elvevåg, B. (2021). Validating Biobehavioral Technologies for Use in Clinical Psychiatry. Frontiers in Psychiatry, 12. https://doi.org/10.3389/fpsyt.2021.503323
“In this paper,” he said, “we compare evaluation of objective measures in other areas of science (e.g., physics, computer sciences, engineering) to that of psychology. There are some critical differences, particularly surrounding how ‘resolution’ is handled. The upshot is that psychology should do a better job of defining exactly when, where and how a phenomenon is occurring. . . at least with respect to validating objective measures.”
• Cohen, A. S., Schwartz, E., Le, T. P., Cowan, T., Kirkpatrick, B., Raugh, I. M., & Strauss, G. P. (2021). Digital phenotyping of negative symptoms: the relationship to clinician ratings. Schizophrenia Bulletin, 47(1), 44-53. https://doi.org/10.1093/schbul/sbaa065
“In this paper, we demonstrate how objective technologies often disagree with what a clinician says. We attempt to unpack why that is – with the idea that neither is inherently wrong. Rather, they are looking a different phenomenon,” he said.
• Cohen, A. S., Cowan, T., Le, T. P., Schwartz, E. K., Kirkpatrick, B., Raugh, I. M., Chapman, H. C., & Strauss, G. P. (2020). Ambulatory digital phenotyping of blunted affect and alogia using objective facial and vocal analysis: Proof of concept. Schizophrenia Research, 220, 141–146. https://doi.org/10.1016/j.schres.2020.03.043
“In this paper, we evaluate a method of evaluating aspects of psychosis using smart phone technologies. We are currently trying to implement these technologies with Capitol Area Human Services District ––though, in early stages.”
Besides pushing the boundaries of what clinical science can do in measuring symptoms of serious mental illness, what was the most enjoyable thing for about the work for Dr. Cohen?
“This is a necessarily multidisciplinary endeavor, and I really enjoy being challenged by my students and colleagues. When trying to objectify aspects of mental illness, we need to be very mindful of the role that demographics, culture and other individual differences play. I am blessed to have a network of colleagues from a variety of walks of life that can help challenge us to create culturally appropriate, and ultimately better, measures.”