Computerized Adaptive Tests as Measure of Depression for Low-Income Black and Latina Women

In this video, Robert Gibbons, PhD, Blum-Riese Professor of Biostatistics; Professor, Public Health Sciences; Director, Center for Health Statistics, University of Chicago, Chicago, Illinois, discusses his recent research, which examined the concordance between computerized adaptive tests (CAT) and traditional measures of depression among low-income Black and Latino women.  

According to Professor Gibbons, “The most significant and intriguing findings of our study is the difference in the rates of depression and anxiety across the trimesters and postpartum between Black and Latina low-income women.”

In the upcoming part 2, Professor Gibbons will discuss best practices for clinicians to use CAT to address health disparities, and future research in this field.

Read the Transcript:

Professor Gibbons:  Hi, I'm Robert Gibbons. I'm the Blum-Riese Professor of Biostatistics at the University of Chicago. I develop the statistical machinery under the hood, if you will, of the CAT-MH. I'll be telling you about that today.

A note about the Blum-Riese professorship, the previous recipient of the Blum-Riese professorship at the University of Chicago was Dr. Janet Rowley, who discovered the molecular basis for leukemia and cancer in general.

Every day when I come into the Biological Sciences Division at the University of Chicago, there are pictures at every entrance of Dr. Janet Rowley receiving the Presidential Medal of Freedom from then-President Barack Obama.

She looks at me and says, "Regression towards the mean is not acceptable.” You cannot imagine how insecure a person I am in the huge, huge legacy and huge footprint that I have to fill. That's a prelude to what you'll hear about my work.

What led you to examine the concordance between computerized adaptive tests and traditional measures of depression among low-income Black and Latina women?

Professor Gibbons:  The question is what led me to examine the concordance between computerized adaptive testing and traditional measures of depression in low-income Black and Latina women. Of course, this is collaborative work with my colleagues at the University of Illinois, at Chicago, and the University of Iowa.

To begin, I've had a lifelong interest in the development of statistical methods for the advancement of mental health research. I think it's a wonderful area to work in. From my early days as a graduate student in statistics at the University of Chicago, I sought out applications of advanced statistical methods for applications in psychology and psychiatry.

Since that time, I've made many statistical contributions to mental health research, including methods for the analysis of longitudinal data, the safety of pharmaceuticals, suicide prevention, and most relevant to this, mental health measurement.

In 2011, my colleague and friend from Biostatistics at Columbia University, he and I founded the Mental Health Statistics Section of the American Statistical Association. That section, Mental Health Statistics, has now over 740 members.

That's more members than the Sports Statistics section, which we think is amazing and very happy to see now the plethora of well-trained statisticians working in this incredibly important field.

Now, traditional measurement in the area of mental health ignores important sources of variation in measured constructs such as depression or anxiety or substance use disorder or suicidality, particularly across racial and ethnic subpopulations.

Life events such as pregnancy and postpartum are also unique situations, sexual identity, and settings such as primary care or emergency departments versus measurement in universities of college students and criminal justice settings. All have important contributions to measurement of mental health constructs. Those experiences, those settings can bias mental health measurements.

These are very important factors that not only differentially affect the incidence of mental health disorders but also affect the way in which we experience these different kinds of clinical symptomatology and the way that we report them.

We've previously studied sources of bias and mental health measurement as a function of race and ethnicity, as well as differences in the expression of depression and anxiety in perinatal women versus the general population.

For example, it should not be surprising to anyone that while fatigue is an important somatic symptom of depression in a general population, it is all but meaningless in a perinatal population, because all pregnant and postpartum women experience fatigue whether or not they have clinical depression.

What we've done is statistically to adjust our tools for these sources of bias. That's why the use of the CAT that makes it so important in this particular application.

As a consequence, when my colleagues at the University of Illinois at Chicago, Colleen Mackie, and her group approached me about the proposed study of depression and anxiety using CAT-MH in low-income Black and Latina women, I was extremely enthusiastic to participate.

Please briefly describe the study method and your most significant findings.

Professor Gibbons:  In terms of a description of the study methods and our most significant findings, I need to start with a brief overview of what it is that we actually have created.

In order to understand the study methods, it's first important to understand how we've significantly advanced the field of mental health measurement through the development of computerized adaptive tests based on multi-dimensional item response theory.

That's quite a mouthful. Let me explain. Traditional mental health measurement uses small, fixed sets of symptoms that are identical for all people, regardless of the severity of their underlying mental health disorder. For example, depression, anxiety, suicidality, or substance misuse.

For a given person there's actually very little information in those items that are really targeted to that person-specific severity level at that moment in time. As such, traditional tests sacrifice the precision of measurement for the speed of the measurement. Adaptive tests, by contrast, quickly learn the severity level of the person based on their responses to questions during the assessment.

Then target subsequent items from a much, much larger bank of items. Sometimes containing hundreds and even thousands of items to the specific severity level of the person. As such, they maximize the precision of measurement while completely eliminating clinician burden and minimizing patient burden.

In this way, we can actually study huge populations, entire states, or even the entire United States, as I'll describe in some of our work later because the same items are not repeatedly administered over and over again, there's no response bias produced by a person recalling previous answers.

These tests can be repeatedly administered at any interval in time. For example, working with our colleagues, John Mann and Mike Bruno Baumann at Columbia University, we are doing studies of ketamine infusion on our tests of depression and suicidality using assessments every 30 minutes.

We're also doing novel long-term treatment trials where we assess the symptomatology of depression every single day for a period of a year. These tests are based on self-reports and take advantage of cloud computing environments so they can be administered anywhere on the planet. We're studying people, schizophrenics, in India. These can be done in or out of the clinic on any Internet-capable device, such as your smartphone.

The most significant and intriguing findings of our study is the difference in the rates of depression and anxiety across the trimesters and postpartum between Black and Latina low-income women.

As an example, among women with at least one positive depression or anxiety screen at any time during their pregnancy or postpartum, 64% of Black women had positive screens for anxiety in the first trimester and 50% postpartum.

Those 2 frequencies, those 2 percentages are pretty similar. By contrast, for Latino women, the rate of anxiety was 77% in the first trimester, much higher, versus only 11% postpartum.

There is a huge difference in terms of the expression of anxiety during pregnancy between early pregnancy, the first trimester, and then postpartum in Latino women, which we do not see in Black women. In the total sample, the rate of anxiety in Black women goes from 2.6% in the first trimester to 5.8% postpartum.

In Latina women, it's the opposite. The rate goes from almost 14% in the first trimester to 0% postpartum. We found similar effects for depression. These are huge differences, and they have major implications on surveillance of low-income Latina women in particularly during the perinatal period.

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.