To evaluate the Pathogenetic Triad (PT) framework as a multilevel architecture for autism liability using out-of-sample prediction, highlighting its potential significance in understanding autism.
Approach:
Model Comparison: Compared theory-constrained PT models with strength-matched, domain-restricted atheoretical combinations using leakage-free nested cross-validation, focusing on predictive accuracy and model interpretability.
Key Findings:
Low-dimensional PT models ranked among the strongest models of comparable size.
Including all three PT domains (AP, CC, NB) provided systematic advantages over alternatives with similar univariate input strength.
Interpretation:
The findings suggest that the Pathogenetic Triad may serve as a multilevel framework for understanding autism liability, emphasizing the role of prespecified models in predictive evaluations.
Limitations:
Based on a small and demographically restricted cohort, which may introduce biases.
Results may not generalize beyond the studied population.
Conclusion:
The study illustrates how prespecified multilevel frameworks can be evaluated in neuropsychiatric samples, providing a structured approach to understanding autism liability.
