ADVERTISEMENT
Using Computational Models to Improve ADHD Diagnosis and Treatment
A recently published study suggests that using computational models to analyze neurocognitive testing data may help clinicians improve the diagnosis and treatment of attention-deficit/hyperactivity disorder (ADHD). Here, lead author Nadja Ging-Jehli, a doctoral candidate in computational psychiatry, explains the findings of the review, how clinicians can use it in practice, and her ongoing work in the field.
What led you and your colleagues to review the use of computational models to understand cognitive processes in ADHD?
My passion is helping people with ADHD (and other mental health conditions) by using my skills in mathematics and neuroscience and my experience in studying human behavior in laboratory experiments (eg, social and cognitive tasks). When I started reviewing the findings in the field of neurocognitive testing for ADHD, I noticed two aspects that motivated me to write this review article.
First, neurocognitive testing for ADHD seemed a confusing field because many different tests have been used and many different contradicting findings have been reported. At the same time, most studies concentrated on specific clinical samples (eg, boys with ADHD in a specific age range). I then realized that the findings became less contradictory the more I investigated the details of each study. For instance, specifics in the cognitive tasks conducted by clinicians can lead to important behavioral differences that can be understood if we consider the findings of research from cognitive psychology. Moreover, I noticed that certain tasks (eg, stop signal tasks) have been used to measure specific cognitive concepts (eg, inhibition failure) without compelling evidence that these tasks indeed measure those concepts.
Second, computational modeling has long been used in nonclinical research areas (eg, cognitive psychology, cognitive neuroscience, behavioral economics) to understand how people make decisions in cognitive and social tasks within a laboratory setting. Unfortunately, cognitive research after 1970 has had little influence on the neurocognitive tests used for clinical purposes.
Based on all of these observations, our review article focuses on integrating findings from cognitive psychology and computational modeling into clinical neurocognitive testing. We also provide a thorough overview of the current field of neurocognitive testing and we discuss barriers that need to be overcome to better utilize neurocognitive testing for clinical practices.
Please briefly describe the study method and your most significant finding(s).
We reviewed 50 clinical studies on a broad range of cognitive tasks. In so doing, we provide a synopsis of differences in sample characteristics, task procedures, and effect sizes across all studies (found in the appendix of our article).
The synthesis of our review suggests that there is more information available from neurocognitive testing, if its data are analyzed with computational models. This additional information may benefit clinicians not only in diagnosing ADHD, but also in selecting and assessing treatments. However, different research fields (eg, clinical science, cognitive psychology, computational modelers) need to work together to: 1. set new standards on how to report findings from those tests (eg, some report test scores and others report aggregated measures, which makes comparisons across studies impossible); 2. understand the difference between analyzing neurocognitive data with summary statistics (eg, mean response times) versus analyzing data with computational models; 3. adjust tasks to make them suitable for model applications. In our article, we provide an introduction to 3 commonly used computational models and we list the requirements that need to be met to apply those models.
Eventually, by integrating the findings of the reviewed studies, we also found evidence that computational models could be promising tools to characterize different ADHD endophenotypes.
How do you think computational psychiatry can change the process of diagnosing ADHD?
Only a few studies have used computational modeling to inform clinical practices such as diagnosis, treatment selection, and treatment assessment. Those studies which did use computational modeling with neurocognitive testing mostly focused on quantifying cognitive differences between individuals with and without ADHD. However, it is not enough for the field of computational psychiatry to simply apply models to clinical data from neurocognitive tests. It will be necessary for the field to show that the results of those models provide insights of clinical relevance; there is a difference between statistical significance and clinical significance. I hope that clinicians will find neurocognitive testing more useful in their daily practice if they start to combine it with computational model analyses.
To date, clinicians mostly rely on questionnaires and interviews to assess the clinical significance of symptoms such as inattention and hyperactivity-impulsivity. This means that clinicians need to be able to observe an individual’s behavior and/or individuals need to be able to articulate symptoms. Additionally, many children with ADHD can compensate for their deficits or find other mechanisms (eg, drugs, sports) that help them to cope with their symptoms. Sometimes, they are even able to “hide” symptoms for some time. Computational modeling is a tool that allows us to observe “how” those children solve school tasks or “how” those children interact with other children in computer games. Therefore, these tools help to quantify certain characteristics that may be dysfunctional but that do not become directly apparent in behavior.
There are multiple ways computational psychiatry may help clinicians—not only in the process of diagnosing ADHD, but also in the process of selecting treatments and assessing the efficacy of treatments. First, computational modeling provides a way to index and quantify biases in an individual’s decision-making process. These biases may be dysfunctional and a characteristic of the underlying disorder. For instance, different individuals with ADHD may vary in the underlying sources of inattention (eg, how much they have problems with getting started on a task relative to how much they have problems with staying focused on a task). Therefore, neurocognitive testing in conjunction with computational modeling may help to identify different endophenotypes of ADHD. The neurocognitive-based model parameters provide clinicians with additional information, asides from the questionnaires and interviews that are already used.
Second, understanding the underlying cognitive deficits of an individual may help clinicians to tailor cognitive behavioral therapies (CBTs) to the needs of that individual. This is because computational model parameters can be used to index and quantify the severity of a deficient cognitive component. These cognitive components can then be targeted in CBTs. Third, neurocognitive testing in conjunction with computational modeling may help clinicians to set eligibility criteria for an intervention. Specifically, computational modeling in conjunction with neurocognitive testing may represent factors that turn out to be important moderators of treatments (CBTs and pharmacological interventions). Fourth, pre- and post-treatment assessments with cognitive tasks and computational modeling may provide another way to assess the efficacy of treatments.
How can mental health clinicians use this review to improve their diagnosis and care of patients with ADHD?
One of our goals was to synthesize different research areas by creating summary tables that focus on the reviewed studies’ differences in: 1. sample characteristic; 2. test statistics; and 3. task procedures. We provide those summaries in the appendix of our review article. I hope that scientists will find it useful for gaining a better overview of this field and for relating the different findings into perspective with each other. I hope that this gained understanding then helps clinicians in their selection of neurocognitive tests. It also invites clinicians to critically rethink whether some of these tests do indeed measure what they were presumed to measure (eg, failure of inhibition control with a stop-signal task?).
Our article also provides clinicians with an overview and introduction into common computational models, explaining how they differ from the statistical measures used thus far. For those clinicians who want to create their own neurocognitive assessments or who want to apply computational modeling, we provide a list of recommendations how tasks and models need to be adjusted to be together applicable. Finally, we provide clinicians with ideas how they can integrate neuronal recordings such as those used with electroencephalography (EEG).
Are you conducting any more research in this area, and are there any related studies you feel are needed?
Computational modeling can be utilized with other measures (eg, neural data, eye tracking data) to create an integrated clinical assessment approach. To date, many studies used resting states of EEG to find biological markers (which has been found to be of limited help in the diagnosis of ADHD because they lack specificity and sensitivity). Other studies used EEG to understand the neural activity during task performance. What we need though is an integrated approach that combines behavioral and neural measures. Computational modeling is a tool that can be used to link these measures. My latest research focuses on creating such an integrative approach (together with machine learning techniques). In fact, we are already collecting data for it. In another study of mine, we currently investigate whether computational modeling can be used to predict who likely benefits from which treatment (eg, pharmacological and/or behavioral therapies).
Nadja Ging-Jehli is a doctoral candidate in computational psychiatry at Ohio State University (OSU), Columbus. She holds two master’s degrees – one from OSU in Psychology and Neuroscience, and one from University of Zurich, Switzerland, in Experimental and Behavioral Economics. She has expertise in constructing computer simulations and in studying human behavior in laboratory settings. The thrust of her research is to find better ways to diagnose and tailor treatments for psychiatric conditions such as attention-deficit/hyperactivity disorder (ADHD).
