The study, published in npj Digital Medicine in May 2022, notes that ASD can be reliably diagnosed as early as 18 months, but diagnostic delays persist in the US. ASD is also one of the most common developmental disorders among children, and early intervention has been shown to improve long-term outcomes.
The researchers further noted that diagnostic delay in the US is due, in part, to the increase in demand for ASD evaluations, which has outpaced specialist capacity and led to prolonged wait times. ASD diagnosis in the US also relies on a limited number of pediatric subspecialists and time-intensive, team-based behavioral evaluations, which can take as long as 18 months from initial screening to final diagnosis.
The use of diagnostic aids in primary care settings to augment ASD care has shown significant potential in previous studies, and the researchers aimed to test the accuracy of one such tool. The tool, an AI-based software-as-a-medical-device (SaMD), offers recommendations for the primary care provider after analyzing behavioral features from a caregiver questionnaire, a healthcare provider questionnaire, and two short home videos.
The tool bases its recommendations on a machine-learning (ML) algorithm that selects behavioral features predictive of ASD. The algorithm was developed and validated using patient record data from children with diverse presentations, conditions, and comorbidities who were either diagnosed with ASD or confirmed not to have it using standard diagnostic tools.
To test the tool’s accuracy, the researchers conducted a double-blinded study in which 711 participants across six states were enrolled. These participants were patients aged 18 months to 6 years old. Though a caregiver or a healthcare provider had concerns that these patients had a developmental delay, none had yet been diagnosed. Following evaluation, the tool’s output was compared to a diagnosis made by a specialist and validated by one or more additional blinded specialists.
Because of COVID-19 restrictions, only 425 participants completed the study. Of these, specialists determined that 61.9 percent had one or more non-ASD developmental or behavioral conditions, 28.7 percent were ASD positive, and 9.4 percent were ASD negative and neurotypical.
The tool classified patients as ASD positive, ASD negative, or “indeterminate,” which indicates that the information input was insufficient for the algorithm to render a highly predictive output. The tool provided a classification of ASD positive or ASD negative for 31.8 percent of study participants.
Of the participants who received an ASD diagnosis by a specialist, 52.5 also received a determinate result from the tool. These were all correctly classified by the device except for one (1) false negative. Of those who received an ASD negative and neurotypical diagnosis by the specialist, 35 percent received an ASD negative result by the tool, and none were misclassified as ASD positive.
There were no detected tool performance differences across participants’ race/ethnicity, gender, education level, or income.
These findings indicate that AI-based diagnostic aids may have significant potential to assist clinicians in primary care settings with ASD diagnosis. However, the researchers state that the study will need to be replicated, and other studies evaluating other devices are required before these tools can be effectively implemented in clinical settings.
REFERENCE: Health IT Analytics (xtelligent HEALTHCARE media); 23 MAY 2022; Shania Kennedy