CHOP Researchers Develop Algorithm to Help Distinguish ADHD Symptoms from Related Conditions

Researchers from the Center for Applied Genomics (CAG) at Children’s Hospital of Philadelphia (CHOP) have developed an algorithm that uses existing electronic health records to distinguish patients with attention-deficit/hyperactivity disorder (ADHD) alone versus patients who present with ADHD and a variety of related comorbid conditions. The findings were recently published in the journal Neurodevelopmental Disorders.

ADHD is a complex condition affecting between about 5 and 8% of school-age children and 2 to 4% of adults, with many possible causes. Because ADHD comes in three different types and exists on a spectrum, it can be difficult to diagnose. Futher compounding diagnostic challenges are a series of other conditions, such as learning disorders, sleep disorders, or anxiety disorders, that affect about half of patients with ADHD, which can complicate whether a patient’s symptoms are being caused directly by the ADHD or one of these other comborbities.

To better distinguish these groups of patients, researchers developed an algorithm using existing electronic health records (EHR) to help distinguish ADHD from other related psychiatric disorders.

“Our goal with this algorithm was to establish a tool that could be used to automate future genetic analyses and improve diagnostic yield and precision in future studies,” said Hakon Hakonarson, MD, PhD, director of the Center for Applied Genomics at CHOP and senior author of the study.

The study team developed a multi-source algorithm, which allowed them to ascertain a more complete view of each patient’s EHR. Using CHOP data as well as data from the Bio-Bank at CAG from 2009 and 2016, the team performed a retrospective case-control study from a total of 51,293 patients. Of those patients, 5,840 were diagnosed with ADHD, and among those cases, 46.1% had ADHD alone while 53.9% had ADHD along with at least one other psychiatric comborbidity.

The researchers found that the algorithm had a positive predictive value of 95% for ADHD and 93% for controls, suggesting a very accurate test and one that could be used in prospective studies going forward. The algorithm also had a positive predictive value ranging from 60% to 100% for other psychiatric conditions, with the higher number of patients with those comborbities (i.e. anxiety and autism spectrum disorder) yielding more accurate results. While ADHD keywords did not appear to significantly help identify patients, ADHD-specific medications on EHRs increased the number of properly identified cases by 21%.

“With the high positive predictive values achieved by this algorithm, we believe we have developed a robust and useful tool for identifying appropriate datasets and successfully distinguishing between groups of patients,” Hakonarson said. “It’s possible that these groups with or without comoribities may respond differently to medication, which could help us design better and more effective methods for therapeutic intervention.”

This work was supported in part by CHOP’s Endowed Chair in Genomic Research, U01-HG006830 (NHGRI-sponsored eMERGE Network), a sponsored research agreement from Aevi Genomic Medicine Inc., and an Institutional Development Award from Children’s Hospital of Philadelphia. 

Slaby et al, “An electronic health record (EHR) phenotype algorithm to identify patients with attention defcit hyperactivity disorders (ADHD) and psychiatric comorbidities.” Neurodev Disord. 2022 Jun 11;14(1):37. DOI: 10.1186/s11689-022-09447-9.PMID: 35690720.