Rule-out test for autism using machine-learning analysis of molecular temporal dynamics in hair - a multicenter study

Area of Science:

• Biomarker discovery
• Autism Spectrum Disorder diagnostics
• Pediatric health

Background:

• Early intervention is crucial for improving outcomes in children with autism spectrum disorder (ASD).
• Current diagnostic methods for ASD are often behavioral and typically suitable for children 18 months and older.
• A validated biosignature test for early autism detection or exclusion is currently unavailable.

Purpose of the Study:

• To develop and validate a hair-strand-based biomarker test as a diagnostic aid for autism spectrum disorder.
• To assist clinicians in ruling out autism in children from 1 month of age.
• To enable earlier and more precise delivery of early intervention services.

Main Methods:

• Analysis of elemental intensities in single hair strands from 1697 children across multiple countries using laser ablation-inductively coupled plasma-mass spectrometry.
• Generation of time-series data from hair shafts to capture elemental fluctuations over time.
• Application of a multi-layered machine learning architecture to analyze temporal data and optimize diagnostic performance (NPV and sensitivity).

Main Results:

• The diagnostic aid demonstrated an Area Under the Curve (AUC) of 0.75, with 97% Negative Predictive Value (NPV) and 96% sensitivity.
• In children 36 months or younger, the test showed improved performance with an AUC of 0.79, 97% NPV, and 96% sensitivity.
• Negative test results significantly reduce the odds of an autism diagnosis by approximately 82%, while positive results increase it by 25%.

Conclusions:

• This hair-strand biomarker test can estimate autism likelihood as early as one month after birth.
• The diagnostic aid can support clinicians in early autism diagnosis, enhancing clinical workflows.
• Early identification facilitated by this test allows for more precise and timely intervention for children with developmental support needs.