Crunching the autism equation in the DSM-5 era

solving the autism equation
An 'Information Commons' could better delineate the different faces of ASD by combining objective molecular, biochemical and neurological measures.
Alal Eran, PhD, studies the molecular basis of autism at Boston Children’s Hospital and Harvard Medical School.

Yet another redefinition of autism spectrum disorder (ASD) has stirred up debate. The new Diagnostic and Statistical Manual of Mental Disorders (DSM-5) now collapses four previously distinct conditions—autistic disorder, Asperger syndrome, childhood disintegrative disorder and pervasive developmental disorder not otherwise specified—under one umbrella label of ASD. It also collapses the traditional autistic triad (social deficits, communication impairments and restricted interests/behaviors) into two domains: social/communication deficits and restricted interests/behaviors.

While intended to increase accuracy and utility, the new diagnostic criteria for autism—the fifth revision since 1980—have attracted an unprecedented level of criticism by clinicians, researchers and families. The criteria for membership in DSM categories are much less robust than those for other clinical classification schemes—as evidenced by the rapid change in the DSM over the last 50 years. But more importantly, they are based only on behavioral symptoms.

Although multiple standardized instruments have been developed for the behavioral diagnosis of ASD, these measures still don’t impart sufficient diagnostic or prognostic accuracy to select an optimal course of treatment. Children with ASD show remarkable individual differences, including differences in treatment response, and many have increasingly recognized ASD comorbidities. These include seizure, sleep, bowel and autoimmune disorders, whose diagnoses could help guide the treatment pathway.

An autism Information Commons

A wide range of investigations have sought objective molecular, biochemical and neurological measures of ASD that could provide an earlier, more precise diagnosis to better guide clinical decisions and targeted therapies. These measures—which arguably, more closely reflect the disorder’s genetic and environmental causes—include genomic variation, gene expression, immunological function, metabolic profiles, eye fixation, electroencephalography (EEG) and functional MRI. Recently, aggregation of electronic health record data has identified four distinct ASD clusters that may help point to different etiologies or underlying vulnerabilities.

While many of these measurements can identify a group of individuals with ASD who fall outside the normal range, each also can identify many individuals without ASD. Therefore, it has been broadly hypothesized that the only way to accurately characterize the different faces of ASD—and accurately assign each patient to the group of patients who most resemble them—is to combine these measures.

This multimodal approach has been successful in other domains of medicine. The evaluation of heart disease, for example, combines echocardiography, troponin levels and stress tests. Yet it remains untested in ASD: The newer measurement modalities have rarely been used together on the same individuals or applied to a sufficiently large, well-characterized population.

We are now trying to change that. A group of investigators from the Boston Children’s Hospital Informatics Program and Research Connection, myself included, recently embarked on a mission to collect, compile and integrate research and clinical findings from children with ASD seen in a variety of labs and clinics across the hospital.

For such an autism Information Commons to succeed, data must be shared with other institutes and organizations, caregivers and parents. Institutional Review Board-approved regulatory frameworks must be in place to enable the acquisition and synthesis of diverse data types collected in multiple contexts, such as research, clinical care, school and home.

Systems biology and artificial intelligence techniques could then be applied to these data to create integrated autism maps. These maps would point to potential etiologies and patients’ expected clinical trajectories, providing better-defined populations for research—and ultimately enabling better, personalized care for people with autism.