Prediction of Autism Risk From Family Medical History Data Using Machine Learning: A National Cohort Study From...

Area of Science:

• Genetics and genomics
• Public health
• Neuroscience

Background:

• Family history of certain disorders is linked to autism spectrum disorder (ASD) risk.
• Previous studies suggest a familial aggregation of ASD.
• This research investigates the utility of family history data for ASD risk prediction.

Purpose of the Study:

• To assess the predictive power of comprehensive family history data for autism spectrum disorder (ASD) risk.
• To develop and evaluate machine learning models for ASD risk stratification based on three-generation family data.
• To compare the predictive accuracy of broad family history information versus immediate family history of ASD.

Main Methods:

• Utilized a Danish population-based cohort of 1,697,231 births (1980-2012) with prospective ASD diagnosis tracking.
• Linked birth records to three-generation family members, analyzing 438 morbidity indicators across 73 disorders.
• Employed machine learning to identify key indicators and develop a family history risk score.

Main Results:

• The best model included 41 indicators (8 mental, 9 nonmental conditions) across six family member types.
• The highest risk group (17.0% ASD prevalence) showed a 15.3-fold increased ASD risk compared to the lowest risk group (0.6% ASD prevalence).
• Individuals with an affected sibling had a 6.1-fold increased ASD risk, highlighting the value of broader family data.

Conclusions:

• A comprehensive family history encompassing multiple mental and nonmental conditions improves identification of individuals at highest risk for ASD.
• This approach surpasses the predictive capability of considering only immediate family history of ASD.
• A detailed family history framework is essential for future clinically relevant ASD risk prediction.