Autism spectrum disorder (ASD) is a complex condition that is difficult to diagnose and treat — but new research in AI holds promise for alleviating the symptoms of ASD patients through precision medicine.
A new Jump ARCHES grant that will address that challenge has brought University of Illinois Electrical and Computer Engineering professor Ravishankar Iyer together with Dr. Adam Cross, who is a pediatric hospitalist and informaticist at the OSF HealthCare Saint Francis Medical Center in Peoria, and also directs the Children’s Innovation Lab at the Jump Trading Simulation & Education Center in Peoria.
ASD is being diagnosed with increasing frequency in children. It results in symptoms such as inattention, depression, repetitive behaviors (of movement or speech), short-temperedness, obsessive interests, a pronounced desire for routine, and unusual reactions to stimuli.
ASD can present a slippery problem for doctors. It is a disease of which other diseases, such as anemia of chronic disease (ACD) and depression, can be symptoms. Conversely, ASD can itself be a symptom of other disorders.
Diagnosis and treatment require experience and careful thought. No medications exist to target, treat or cure ASD; medications can only target signs or symptoms that are distressing to patients, their families or their physicians. Different patients respond differently to the same medications, so the identification of useful medications for a specific patient involves a process of trial and error that can be wasteful and disruptive for patients and their families.
Cross says, “Because many of these behaviors are so diverse, and because ASD is so diverse, it’s really hard to pin down a particular medication dosage strategy for the behavior of a patient with ASD.”
He has joined with Iyer in applying AI to the problem of individualized or precision medicine. Their project follows Iyer’s earlier research in AI-tailored precision medicine as applied to cancer pharmacogenomics, major depressive disorder, neurological and degenerative brain diseases including seizures and Alzheimer’s, and chronic diseases like diabetes. Iyer’s research group also has major collaborations with Mayo Clinic and Singapore’s National University Hospital and has won awards for its pioneering research in precision medicine.
“I can’t overemphasize how wonderful it is to work with such dedicated physicians,” comments Iyer.
There is no standard method of addressing ASD, explains Cross, so most physicians will ramp up a medication slowly, taper it back down slowly, and proceed with caution until an optimal prescription is achieved.
The AI solution will support physicians by suggesting proposed treatments based on datasets too complex for a physician to review manually, such as sets of genetic data.
Data the AI will assimilate include DNA, common laboratory values, diagnosis codes, procedure codes, billing records and past case histories.
Crucially, the AI will also rely on the physician’s written or typed notes to get a qualitative analysis of the effectiveness of medications, including evaluation of side effects and value judgments about the patient’s quality of life. Analysis of such longitudinal data in potentially abbreviated English requires processing of natural language that contains the physician’s observations, insights, and experience, and will be an important part of developing the AI model.
To add to the complexity, physician notes may include both notes and a separate problem list — a document that states the most important health problems facing a patient, such as illnesses or diseases, injuries suffered by the patient, and anything else that has affected or is currently affecting the patient.
“So, it’s very common that a physician will put in their note a diagnosis but won’t put it into the problem list for the patient. And they’ll bill for it. But the problem list and the billing diagnosis and the note are all in different places, and they don’t talk to each other,” explains Cross.
He adds, “So much of the way we document in health care revolves around either caring for the patient or billing for the patient, and there have been different structures in the patient’s medical record that have been created for those separate purposes and they don’t always align. Few of these pieces actually were created for the intended purpose of research.”
“You have to actually make very intelligent choices of how you manage this information, how you personalize it,” says Iyer. “In the end, we worry about the toxicity of medication; it might impact me more than it impacts you. I may have a better chance of absorbing the medication because of my genomic characteristics, but also due to metabolomics, how we metabolize the chemicals.”
When deployed, the proposed AI model will work with physicians with the goal of reducing the time, expense, and possible distress associated with finding an optimal prescription to target the signs and symptoms of ASD.
Iyer is also a researcher in the Coordinated Science Lab. The research was funded in part through the Jump ARCHES program, a collaboration of OSF HealthCare, the University of Illinois College of Medicine Peoria and UIUC’s Health Care Engineering Systems Center.