New parent-reported questionnaire may accelerate autism detection at 18 months

Researchers develop a streamlined, machine-learning-based parent questionnaire that improves early autism detection at 18 months, promising more timely interventions and better childhood outcomes.

At the 18-month check-up, paediatricians are being asked to make a rapid judgment that can shape a child’s future: whether early social and communication development is on track or needs closer attention. The American Academy of Pediatrics recommends universal autism screening at 18 and 24 months, alongside broader developmental screening at routine visits, because early support works best when it begins early. Yet autism is still often diagnosed well after a child’s fourth birthday in the United States, leaving a long gap between the first warning signs and intervention. Research published in the Journal of Autism and Developmental Disorders suggests that one of the best tools for narrowing that gap may be a better questionnaire completed by parents.

The study, led by Raymond Sturner at Johns Hopkins University School of Medicine and the Center for Promotion of Child Development through Primary Care, focused on what works best at the age when many screening tools struggle most. Existing autism screens, especially the Modified Checklist for Autism in Toddlers, Revised with Follow-Up, have known weaknesses in younger toddlers. At around 18 months, its predictive value falls sharply, and the follow-up interview designed to improve accuracy is often difficult to complete in routine practice. A systematic review has also found that the tool’s performance varies widely from one group of children to another, reinforcing concerns that a single screening approach is not enough.

To build a stronger alternative, the researchers drew on items from several parent-report instruments already used in autism and developmental screening, including the M-CHAT-R/F, the Parent’s Observation of Social Interaction, the Q-CHAT-10-O, the First Year Inventory and the Parent Observation of Early Markers Scales. They also added expressive vocabulary questions from the MacArthur-Bates Communicative Development Inventory, reflecting evidence that early word production can be an important clue at 18 months even though language delay alone is not part of autism’s formal diagnostic definition. In all, parents of 11,878 children aged 16 to 20 months completed online screening before their 18-month visits, and a final diagnostic sample of 408 children underwent follow-up assessment with examiners who did not know the screening results. The larger clinical context matters: the CDC says children should be screened for developmental delays at 9, 18 and 30 months, and for autism at 18 and 24 months, precisely because repeated surveillance can catch problems that emerge gradually.

The team then used machine-learning methods to shape a shorter, more precise tool. The resulting screen, called the Toddler Autism and Developmental Adaptive Screen, reduces the item set to 19 features and places particular weight on expressive vocabulary and joint attention, the early behaviours that include pointing to share interest, following another person’s point, showing objects and checking a caregiver’s reaction. Those social behaviours have long been seen as among the earliest and most consistent markers of autism. The model also retains questions about pretend play, imitation, unusual finger movements and sound sensitivity, and it uses adaptive vocabulary testing rather than a fixed checklist, which lowers the burden on families. The AAP has long stressed that repeated screening at 18 and 24 months is important because some children later diagnosed with autism first show language plateau or regression during that window, making sensitive parent-report tools especially valuable.

On the study’s holdback sample, the new model identified autism with higher sensitivity than the older tools it was compared with and also performed better for developmental delay than the standard Ages and Stages Questionnaire. The authors say it was the only screen in this age range to reach generally recommended performance for both autism and developmental delay at the same time, with scores above .70 for sensitivity and specificity on both outcomes. Its positive predictive value for autism remained limited, however, which means false positives are still part of the picture and no screening test can fully escape the problem of low prevalence. That is a reminder that screening is not diagnosis: its purpose is to flag children who need further evaluation, not to label them.

The practical appeal may be as important as the statistics. The new approach can be completed with fewer parent items than the usual combination of autism and developmental screens, requires no follow-up interview and could be integrated into electronic health records or administered on a smartphone. The study’s authors are candid about its limits: the sample leaned towards more highly educated parents, some families dropped out, and children exposed to English less than half the time were excluded. They also note that not every child with autism can be identified at 18 months, so continued surveillance remains necessary. Even so, the work points towards a more efficient way of using information parents already have about their child, and a larger, more diverse trial is now under way. If those findings hold, a simple questionnaire could become a more powerful early warning system for both autism and developmental delay.