[Early prognostic of ASD: A challenge]

Area of Science:

• Neuroscience
• Genetics
• Developmental Pediatrics

Background:

• Autism Spectrum Disorders (ASD) are complex neurodevelopmental conditions with diagnoses typically occurring between 3-5 years of age.
• Current diagnostic timelines delay early intervention, potentially impacting long-term outcomes.
• Intrauterine genetic or environmental factors are implicated in ASD development.

Purpose of the Study:

• To test the hypothesis of identifying infants at risk for ASD at birth.
• To facilitate early psychoeducative interventions to mitigate symptom severity.
• To explore novel predictive markers for ASD.

Main Methods:

• Utilized a machine learning analysis on comprehensive maternity data (biological and ultrasound) from French maternities.
• Data collected included in utero and postnatal information.
• Employed a 'without a priori' approach to identify predictive patterns.

Main Results:

• The model successfully identified 96% of infants who would later be diagnosed as neurotypical at birth.
• Approximately 50% of infants who would later be diagnosed with ASD were identified at birth.
• Identified several unexpected predictive parameters with no previously known association with ASD.

Conclusions:

• This machine learning approach enables early identification of infants at risk for ASD.
• The methodology holds potential for later ASD diagnosis and understanding ASD heterogeneity.
• Early detection facilitates timely intervention, potentially improving developmental trajectories.