Autism can be diagnosed from routine medical information collected within a month of birth, research in tens of thousands of children suggests.
Nearly half the children subsequently diagnosed as autistic could be identified from information recorded in electronic health records (EHRs) within 30 days of birth, according to results published in JAMA Network Open.
EHRs can include data on low birthweight, prematurity, low Apgar scores and other perinatal complications, as well as autism-related conditions such as postnatal hyperbilirubinemia, respiratory infections and issues with sleep, crying or feeding.
The accuracy of EHR diagnosis was competitive with caregiver surveys, which were collected much later at 18 to 24 months of age, and it improved further when data for 360 days were included.
“The results of this diagnostic study conducted in a large academic medical center suggest that EHR-based autism prediction reaches a clinically meaningful level of accuracy as early as 30 days after birth,” report Matthew Engelhard, PhD, from Duke University School of Medicine, and co-workers.
“We observed that almost half (45.5%) of autistic children can be identified at 30 days while maintaining high specificity (90.0%).”
The findings suggest that combining EHR information with caregiver surveys could improve the accuracy of early autism screening, enabling families to gain more timely access to behavioral support.
Noting that early detection of autism is associated with improved outcomes, the investigators conducted their retrospective diagnostic study using EHR data from 45,080 children seen before the age of 30 days within the Duke University Health System between 2006 and 2020.
Subsequent autism spectrum disorder and other neurodevelopmental conditions were identified using billing codes.
The team used 60 percent of the data to train and fine-tune EHR-based autism detection models, with the remaining 40 percent deployed as a test set to evaluate performance.
Overall, 18,032 children were randomly assigned to the test set, including 363 autism patients and 3721 control participants meeting the criteria for other neurodevelopmental conditions. A total of 3615 control participants were followed up through age eight and were included when calculating primary performance measures.
The EHR-based autism detection model achieved 45.5% sensitivity and 23.0% positive predictive value at 90.0% specificity at age 30 days.
This improved at 360 days to 59.8% sensitivity and 17.6% positive predictive value at 81.5% specificity, and 38.8% sensitivity and 31.0% PPV at 94.3% specificity.
The researchers note that the EHR-based tool provided information on the likelihood of autism that was complementary to the Modified Checklist for Autism in Toddlers (M-CHAT) screening tools, which are commonly used.
But they add: “In contrast with existing screening tools, such as the M-CHAT, EHR-based autism detection takes place at a much earlier age (≥30 days) and is entirely passive, meaning that it does not require any data collection other than that which already takes place during routine care.”
The team concludes: “The results suggest that EHR-based monitoring should be integrated with the M-CHAT, other caregiver surveys, and other screening tools to improve the accuracy of early autism screening.”
