Researchers Develop Global Approach for Using AI to Identify Kids with Rare Diseases

Rare Daily Staff

Diagnosing children with rare, genetic diseases often involves misdiagnoses and prolonged diagnostic odysseys but researchers at the University of Wisconsin have developed an algorithm that they say can identify kids likely to have a rare genetic disease.

In a study published in Orphanet Journal of Rare Diseases, the researchers and their collaborators at GeneDx and Ichan School of Medicine at Mount Sinai describe their digital phenotyping algorithm for identification of a pediatric population with a definitive or possible genetic disorder.

“Identifying pediatric patients across an entire population with or who possibly have a rare genetic disorder is critical for improving patient outcomes,” the authors wrote. “We and others have attempted to identify patients with specific genetic disorders using EMR data but have found that such a process is not straightforward, largely due to coding differences, unconfirmed diagnoses, variation in disease names and terminology, and inaccurate information represented in medical records.”

Rather than identify children with a specific disease, the PheIndex algorithm is intended to provide a global approach to address an unmet need to identify children with rare genetic disorders and potentially help overcome obstacles to getting an accurate diagnosis. To date, digital phenotyping studies using electronic medical record data have largely focused on identifying populations with specific individual diseases.

The PheIndex patients on 13 criteria to derive a score for children from birth to 3 years to determine if a child is presenting with an illness that may be a rare genetic disorder. These include such things as prolonged NICU stays, prolonged hospitalizations, multiple ER visits, feeding support, respiratory support, developmental delays, and more. The criteria rely on items that could be extracted from an electronic medical record.

They validated the algorithm through blinded chart review by a pediatrician and a clinical geneticist. The cohort for the study included 93,154 newborns linked to 68,893 mothers who delivered in the Mount Sinai Health System from 2007 to 2019. The researchers assessed the frequency of each of the 13 PheIndex digital phenotyping criteria in its cohort and summarized the number of children aged 0 to 3 years old that satisfied each of the 13 criteria.

The researchers said the PheIndex Score could be used as a clinical guide to shorten the diagnostic odyssey of hard-to-diagnose patients, timely administration of therapeutics by facilitating more rapid diagnosis, and assessing clinical benefit of genetic testing.