Same-wave cross-sectional risk stratification for parent-reported current anxiety among US children with ADHD who received behavioral treatment: development in the 2023 NSCH and evaluation in the 2024 NSCH
This study developed and validated parsimonious regression models using 2023 National Survey of Children's Health data to classify parent-reported current anxiety in US children with ADHD receiving behavioral treatment, demonstrating moderate discrimination and superior calibration compared to machine learning approaches when evaluated in an independent 2024 sample.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
In the landscape of childhood health, attention-deficit/hyperactivity disorder, or ADHD, is a well-known condition that affects how a child focuses and controls their impulses. It is common for children with ADHD to also carry other emotional burdens, such as anxiety, which is a state of persistent worry or fear. While doctors often treat the ADHD itself, sometimes with behavioral strategies that teach coping skills, the accompanying anxiety can remain hidden or untreated. This is a significant concern because anxiety changes how a child feels and functions, and it often requires its own specific attention. The challenge for health systems is to figure out which children, even those already receiving help for their ADHD, are also struggling with current anxiety, so they can be directed toward the right support.
A team of researchers set out to build a tool to help identify these children using data from a massive, nationwide survey of American families. They focused specifically on children between the ages of six and seventeen who had been diagnosed with ADHD and had received behavioral treatment in the past year. The goal was not to predict if a child would develop anxiety in the future, but rather to create a snapshot that could classify whether a child was currently experiencing anxiety based on information parents had already provided. To do this, the researchers gathered information on twenty different factors, ranging from the child's age and family income to their sleep habits, whether they had been bullied, and the mental health of their parents. They used this information to build four different mathematical models, which are essentially sets of rules designed to weigh these factors and estimate the likelihood of anxiety.
The researchers first built these models using data collected in 2023. They tested the models against each other to see which ones were best at sorting children into "likely to have anxiety" and "unlikely to have anxiety" groups. Then, to ensure the models were not just lucky with the 2023 data, they locked the rules in place and applied them to a completely new group of families surveyed in 2024. This second group was different people, surveyed a year later, which served as a rigorous test of whether the models could work in the real world. The results showed that the simpler models, which relied on standard statistical methods, performed just as well as the more complex, computer-heavy machine learning models. In fact, the complex machine learning model actually performed worse, producing predictions that were too extreme and less reliable.
When the researchers looked at the numbers, they found that about half of the children in their study who had observed anxiety data were currently experiencing anxiety. The models developed from the 2023 data were able to distinguish between children with and without anxiety in the 2024 group with a moderate level of accuracy. The best-performing models correctly identified the presence or absence of anxiety in roughly 72 out of 100 cases. More importantly, these models were well-calibrated, meaning the probabilities they assigned to children matched the actual rates of anxiety seen in the group. The complex machine learning model, by contrast, tended to overestimate the risk for some children and underestimate it for others, making its predictions less trustworthy.
The study makes it clear that these models are not diagnostic tools that a doctor can use to tell a parent, "Your child definitely has anxiety." Instead, they are research instruments designed to sort children into risk groups. They work by looking at a single moment in time, using the same information a parent gave about their child's current state. Because the data comes from a single survey where the parent answers questions about both the child's behavior and their mental health at the same time, the models cannot prove that one thing causes the other. They simply show which combinations of factors are most common among children who currently have anxiety.
Ultimately, the researchers found that a straightforward approach using regression models was sufficient for this task. These models offered a balanced view, correctly identifying children who needed further attention without the instability of the more complex algorithms. The study suggests that for this specific population of children with ADHD who are already in treatment, simpler tools might be more effective at flagging those who also need help for anxiety. However, the authors emphasize that these findings are a starting point. Before such a tool could be used in a clinic to guide real-world decisions, it would need to be tested in different settings, checked for fairness across different groups of people, and evaluated to see if using it actually leads to better outcomes for children. For now, the work stands as a careful demonstration of how to build and test a risk-stratification tool using large-scale national data, highlighting the value of simplicity and the need for caution when applying these findings outside of research.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.