New Diagnostic Screen for ADHD Developed Using Machine Learning
FEBRUARY 22, 2018
Lenard Adler, MDResearchers applied a machine-learning algorithm to a newly-updated diagnostic tool, the Adult ADHD Self-Report Scale (ASRS), to help doctors quickly and effectively screen and diagnose patients with adult attention deficit hyperactivity disorder (ADHD), which is commonly overlooked.
The 6-question screen was first published in JAMA Psychiatry in April 2017, and an extended version was validated in a February study in The Journal of Attention Disorders. The authors plan to make it, its scoring algorithm, and educational materials available online to doctors within the next 6 months.
While researchers estimate that ADHD impacts 8–9 million adults, only about 11% are diagnosed and actually treated. Despite effective treatments, many doctors don’t screen for the disorder, or may miss it when it presents as a comorbidity to anxiety or depression.
“Our hope is that eventually the ADHD screener will be used as commonly as the PHQ-9 is. Everybody who comes in once a year for their physical gets a PHQ-9 for depression. We hope that people will be screened for ADHD because it is a highly common and impairing disorder,”
Lenard Adler, MD, professor of psychiatry, New York University School of Medicine, and an author on both studies, said, “[and] we do have good treatments available."
Currently, doctors use an 18-question checklist where each question corresponds to a symptom that impacts a patient “sometimes,” “often,” or “very often,” and each is weighted evenly. The questions are based off the DSM diagnostic criteria.
In 2013, the new Diagnostic and Statistical Manual, DSM V, was published and includes subtle, but important updates to the diagnosis of ADHD. The DSM V broadened the definition of the disorder, increasing the age of onset and decreasing the number of symptoms necessary for a diagnosis from 6 to 5. The manual emphasizes that the symptoms should be present in more than 1 setting such as at work and at home.
Adler and his team updated the ASRS checklist to reflect the new DSM V guidelines, then used a machine-learning algorithm to identify and weigh the most telling questions to develop a short and reliable 6-question screen. The researchers gave the checklist to 637 patients from 3 different populations including a household sample of 119 individuals, a managed care sample of 218, and a clinical sample of 300 individuals.
The machine-learning algorithm, RiskSLIM (Risk-Calibrated Super-sparse Linear Integer Model), compared the patients’ answers to their clinical diagnoses and developed the screen of 6 questions, some weighted more than others. The questions included 1 related to inattention, 3 related to hyperactivity-impulsivity, and 2 related to executive function, which is not a DSM core symptom of ADHD but commonly co-travels with the disorder.
They found that the updated 6-question scale was able to reliably identify patients who should undergo further evaluation for ADHD risk, with few false positives.
“We used artificial intelligence to establish the best loading up the symptoms," said Adler. “It's a sign of the times using artificial intelligence that now we're scoring it electronically. It's a change.”
The second study expanded the ASRS and the Adult ADHD Investigator Symptom Rating Scale AISRS to reflect the new guidelines and to specifically assess executive function deficits (EFDs) and emotional dyscontrol (EC), 2 items that are not symptoms of ADHD but often co-travel with the disorder. The pilot study included 297 respondents, 171 with adult ADHD, and demonstrated high validity of the expanded assessment.
“It's important to ask about [EC and EFDs.] They were not included in a diagnostic revision, even though many individuals feel, and much science supports, that when present they are highly impairing,” explained Adler, “[and], in many ways, drive the symptom presentation.”
Related Coverage >>>
White Matter Patterns Associated with Mental Illness, Cognitive Ability Identified
Working Memory, Vigilance Linked to ADHD
ADHD Shares Genetic Basis With Educational Under-attainment