New Developments in Screening Tools for PTSD

In this video, Robert Gibbons, PhD, Blum-Riese Professor of Biostatistics, Departments of Medicine, Public Health Sciences, and Psychiatry; Director, Center for Health Statistics, University of Chicago, Chicago, Illinois, discusses new developments in screening tools for post-traumatic stress disorder (PTSD) using computerized adaptive testing.

In part 2, Dr Gibbons will further discuss the machine-learning algorithm used to derive diagnosis of PTSD.

Read the Transcript:

Dr Gibbons:  How did we get into this? What led us to the development of these screening tools and measurement devices for post-traumatic stress disorder (PTSD)? To answer that question, let me step back for a moment and put all of this into the context of our research. First, you should know that I'm a statistician interested in the field of measurement.

I'm Robert Gibbons and I'm the Blum-Riese professor of statistics of the University of Chicago. In 2002, David Kupfer, who was then the chairman of the Department of Psychiatry at the University of Pittsburg and later became the head of DSM-5, he and I received the first five-year RO1 grant from the National Institute of Mental Health.

This was the first of many RO1 grants since that time to develop a completely new approach to psychiatric measurement. It was the beginning of 20 years of continuous funding of this program of research. We expect the funding to continue.

The idea was to elevate the quality of measurement in the social and behavioral sciences to that enjoyed in the physical sciences where I had done a considerable amount of work related to the measurement of trace-level elements in analytical chemistry.

It occurred to us that we could use many of the same ideas to improve the quality of mental health measurement dramatically increasing the precision of measurement while at the same time decreasing the burden of measurement. We began with the development of computerized adaptive tests based on multidimensional item response theory for the measurement of depression, anxiety, and mania.

We also developed the first computerized adaptive diagnostic screener for major depressive disorder. This idea of different approaches to diagnostic screening and dimensional severity measurement is fundamental to our work in general and PTSD in particular. We turned our attention to PTSD soon after because of its high prevalence in our nation's active military and veteran populations.

Meta-analyses have revealed that PTSD has a prevalence of 23% in Iraqi war veterans. Among adults who are in the US without a history of service in the military, the lifetime incidence of PTSD is 7%, making PTSD an important area of scientific attention both in the military and non-military populations and a fundamental concept in terms of our nation's public health.

For these reasons, we felt that it was of critical importance to use our new technology to further advance screening and measurement of PTSD. We began by developing and validating our diagnostic screener and dimensional severity measure in a veteran population working with Dr. Lisa Brenner and her colleagues at the Rocky Mountain VA.

Let me spend a moment explaining what the technology is in a non-technical way. What is item response theory and how does it differ from traditional mental health measurement that's based on classical test theory? I like sports. Let's use a sporting analogy.

Classical test theory is like the hurdles. If I change the number of hurdles, if I change the distance between them or their height, I can no longer compare the time it takes to run the race. However, item response theory is more like the high jump. Your ability is measured by the highest height of the bar that you can clear.

An experienced high jumper will place the bar around six feet. An inexperienced high jumper will place it at three or four feet. Nevertheless, the change in the task does not affect our ability to measure the ability of these two high jumpers in exactly the same metric. What can we do with this? We can do CAT or computerized adaptive testing.

What is CAT? Imagine you had a thousand-item mathematics test ranging in difficulty from simple arithmetic through advanced calculus and you had two examinees. You had a fourth-grader and one of my graduate students in statistics here at the University of Chicago.

I could give each of them this thousand-item test, but it would waste their time since I'd be giving arithmetic questions to a grad student and calculus questions to a fourth grader. I could, alternatively, construct a short-form test, maybe with just two questions, an arithmetic question, and a calculus question.

The child would get the arithmetic question correct and have a score of one. My graduate student would get both correct and have a score of two. I would come to the ridiculous conclusion that my graduate student had twice the mathematical ability of a fourth-grader. What I have done is I've sacrificed the precision of measurement for the speed of measurement.

Nevertheless, this is exactly how traditional mental health measurement works. A better approach would have been to give both examinees an algebra question. When the child got it wrong, move to easier items and when the graduate student got it right, move to harder items. That is what CAT is.

What we have developed are CATs for complex, multidimensional constructs like depression, anxiety, and PTSD. What did our study consist of? What were the methods? What were the most significant findings? We've developed two new tools for PTSD assessment. They are completely novel.

The first is a computerized adaptive diagnostic, or CAD, screener for PTSD. This is a binary indicator of whether or not you would meet criteria for the diagnosis of PTSD based on DSM-5 if you had spent an hour with a trained clinician doing a structured clinical interview. The second is a computerized adaptive test, or CAT, for the measurement of PTSD symptom severity.

The scientific methodologies that were used to develop the CAD and the CAT PTSD are quite unique but quite different statistical methodologies. To begin, we developed a bank of 211 items that were used in creating both the CAD and the CAT. For the CAT, we pre-calibrated the item bank using a multidimensional item response theory model called a bifactor model.

Don Hedeker, also a statistics professor here at the University of Chicago, he and I developed this in the 1980s. Unfortunately, the computations were so heavy that they could not be performed on even the fastest computers of the day. However, we can now screen millions of people each day on their smartphones for any of these mental health constructs.

Dr Robert Gibbons, PhD, is the Blum-Riese Professor of Biostatistics in addition to serving as a director in the departments of medicine, public health sciences, and psychiatry at the University of Chicago, Chicago, Illinois, as well as the director of the Center for Health Statistics. Dr Gibbons received his PhD in statistics and psychometrics from the University of Chicago. Dr Gibbons is a fellow of the American Statistical Association, where he received a lifetime achievement award and created a section for mental health statistics, and several other national and globally recognized statistical societies.