Clinical Scorecard: Collaborative Federated Learning for Equitable Screening of Autism Spectrum Disorder in Diverse Age Groups
At a Glance
Category
Detail
Condition
Autism Spectrum Disorder (ASD)
Key Mechanisms
Federated learning enables collaborative model training across multiple datasets without sharing raw patient data, preserving privacy and improving generalizability across age groups.
Target Population
Children, adolescents, and adults at risk for ASD
Care Setting
Healthcare and clinical screening environments utilizing behavioral and demographic data
Key Highlights
Federated learning models achieved high accuracy: 97.2% in children, 89.5% in adolescents, and 86.8% in adults for ASD screening.
The framework improves fairness and robustness under heterogeneous, non-IID data conditions compared to centralized machine learning models.
Privacy-preserving collaborative training enables scalable and ethical ASD screening without sharing sensitive patient data.
Guideline-Based Recommendations
Diagnosis
Use machine learning-based screening tools as adjuncts to identify individuals at risk for ASD across diverse age groups.
Recognize that ML screening models do not replace comprehensive clinical assessment by qualified professionals.
Management
Implement federated learning frameworks to facilitate large-scale, privacy-aware ASD screening.
Incorporate behavioral and demographic data from multiple sources to enhance screening accuracy and generalizability.
Monitoring & Follow-up
Evaluate model performance continuously across age cohorts to maintain accuracy and fairness.
Monitor computational efficiency and communication costs to optimize federated learning deployment.
Risks
Avoid overreliance on ML screening outputs without confirmatory clinical evaluation.
Be aware of potential biases due to data heterogeneity and ensure fairness in model predictions.
Patient & Prescribing Data
Individuals across children, adolescent, and adult age groups undergoing ASD screening.
Federated learning models provide accurate risk screening while preserving patient privacy, supporting early identification and referral for clinical assessment.
Clinical Best Practices
Use federated learning to collaboratively train ASD screening models without sharing raw patient data to maintain privacy.
Combine feature selection, data preprocessing, and oversampling techniques to address missing data and class imbalance.
Interpret ML screening results as preliminary risk indicators requiring further clinical evaluation.
Ensure equitable access to ASD screening tools across diverse populations and regions.
Continuously benchmark federated learning models against centralized approaches to validate performance and fairness.
