Artificial Intelligence Tool May Help Speed Diagnosis
An artificial intelligence (AI) model based only on the occurrence of health conditions previously and newly associated with fragile X syndrome accurately identified fragile X patients five years earlier than diagnosis, according to a study based on nearly 5,600 people in Wisconsin.
Among early signs of fragile X were not only neurologic and cognitive conditions commonly linked to the disease, but also respiratory and ear infections, developmental impairment, and teeth and gum problems.
“A lot of people are still not getting the proper diagnosis or, they have to go through a really long process before being diagnosed,” Arezoo Movaghar, PhD, the study’s first author and a postdoctoral fellow at the Waisman Center at the University of Wisconsin-Madison (UW-Madison), said in a university press release.
These early findings suggest this type of AI model could be used to automatically alert physicians to the risk of fragile X in patients showing several conditions linked to the disease and thereby promote early genetic testing and diagnosis.
Still, the model has to be validated in other geographic regions.
“Now we have to see whether those new conditions also appear if we were to ask the same question in another data source,” said Marsha Mailick, PhD, the study’s senior author at Waisman and the emeritus vice chancellor of research and graduate education at UW-Madison.
The study, “Artificial intelligence–assisted phenotype discovery of fragile X syndrome in a population-based sample,” was published in the journal Genetics in Medicine.
Fragile X syndrome is the most common inherited cause of intellectual disability and autism, but it remains significantly underdiagnosed in the general population.
While there is not yet a cure, earlier fragile X diagnosis “has important implications for patients and families and would allow for timely intervention, appropriate genetic counseling, and family planning,” the researchers wrote.
Fragile X is associated with a variety of cognitive, developmental, and behavioral symptoms, as well as physical features, which differ in severity and occurrence among patients. Additional medical issues, such as ear infections, overly flexible joints, flat fleet, gastrointestinal disorders, and heart valve defects also have been reported in these patients.
“However, the prevalence of these medical issues in these individuals has not yet been estimated using population data, nor has the difference in prevalence between those with FXS [fragile X syndrome] and those in the general population been investigated,” the researchers wrote.
Understanding the full spectrum of lifetime medical conditions associated with fragile X would not only help accelerate diagnosis, but also “inform public health policies regarding services needed by families and patients,” the team added.
“There are patterns in the data in the electronic health records that can reveal important [clinical information],” on the full manifestation of the disease, Mailick said.
However, “there’s no way that we can look at 2 million records and just go through them one by one,” Movaghar said, adding that machine learning “help us to learn from what is in the data.”
Machine learning is a form of AI that uses algorithms to analyze large amounts of data, learn from its analyses, and then make a prediction.
In the current study, Mailick and his team used machine learning to identify patterns among the electronic health records of 55 fragile X patients (44 males, 11 females) and 5,500 unaffected individuals (used as controls) matched to patients in a 1–to–100 ratio for age and sex.
These individuals were selected from a pool of de-identified records of more than 1.7 million people collected over 40 years by the large, not-for-profit Marshfield Clinic Health System, which serves northern and central Wisconsin. The presence of fragile X and other co-occurring health conditions was identified through specific medical codes.
This discovery-oriented approach was able to identify fragile X-associated co-occurring conditions beyond the well-known mental and neurological disorders. These included circulatory, endocrine and metabolic, dental, gastrointestinal, and genitourinary conditions.
Particularly, heart valve disorders were five times more frequently recorded among fragile X cases than in the general population, confirming that regular screening for circulatory disease is critical in this patient population.
Stomach disorders also were five times more common among patients, and gastrointestinal problems were generally diagnosed at a younger age than in controls.
Diseases of hard tissues of teeth were the most frequently reported conditions, in 43.6% of patients, compared with 11.9% among the general population. Notably, the greatest difference was found for developmental problems, which were reported nine times more often among patients than in controls (27.3% vs. 2.6%).
“Although FXS is often thought of primarily as a neurological disorder, it is in fact a multisystem syndrome involving many co-occurring conditions … and they are associated with a considerable burden on patients and their families,” the researchers wrote.
Based on these associations and machine learning, the team “successfully created predictive models to identify cases at least five years earlier than the time of clinical diagnosis of FXS without using any genetic or family history information,” the researchers wrote.
The model showed that patients received diagnostic codes for intellectual disability, anxiety, attention-deficit/hyperactivity, depression, upper respiratory infections, sore throat, problems in teeth and supporting structures, developmental impairment, and ear infections, at least five years prior to being diagnosed with fragile X.
Particularly, while these patients were diagnosed with cognitive, developmental, and behavioral problems between a median age of 3 and 13.5, they were not diagnosed until a median age of 23.5 years.
“Incorporation of artificial intelligence approaches into the medical system could serve as a prescreening tool and create a structure to automatically alert physicians about the presence of multiple FXS-related phenotypes in the patient’s medical records,” the researchers wrote.
This approach “could accelerate the diagnostic process and be instrumental in identifying undiagnosed adults in the population and addressing their health conditions,” they added.
Early diagnosis also may help identify relatives with fragile X or fragile X-related conditions.
“Just knowing and receiving the proper diagnosis gives you the answers to this question that you always had of why you’re experiencing these health conditions or what’s happening to your child,” Movaghar said.