31097
The Clinical Feasibility of Deconstructing Autism into a Pathogenetic Triad: Classification Results from a Case-Control Cohort
Several mechanisms have been shown to influence the risk and severity of autism. Autism can be understood as resulting from a triad of contributing factors: genetic vulnerability from common susceptibility variants influences the pervasive and normally distributed broader autism phenotype (BAP; operationalized as the Autism-Spectrum Quotient (AQ)); cognitive compensatory ability (operationalized as the working memory subscale (WMIQ) of the Wechsler Adult Intelligence Scale (WAIS)) alleviates social difficulties and allows for adaptive behaviors and learning; environmental risk factors (operationalized as Heart Rate Variability (HRV)) negatively influence brain development and cognitive compensatory ability. While the sheer number, heterogeneity, and variety of environmental factors make quantification difficult, we propose HRV as a candidate surrogate variable that may capture enough of an individual's history to be useful for diagnostic purposes.
Objectives:
Risk factors for autism have previously been theoretically discussed. However, a concrete and clinically relevant framework that can guide clinical reasoning has not been outlined. The objective of this study was to empirically validate the clinical feasibility of a pathogenetic triad by investigating its ability to classify a case-control cohort.
Methods:
20 neurotypical controls and 21 high-functioning individuals with autism (all adult males, age- and total IQ-matched) underwent neuropsychological testing using the WAIS and AQ, as well as ECG recording during a face-processing experimental paradigm of a magnetoencephalographic session. The inter-beat intervals from the ECG recordings were extracted and used to calculate the logarithm of the Cardiac Vagal Index (CVI), a measure of HRV. The AQ, WMIQ and CVI were plotted in 3D using Python Matplotlib to identify the plane of maximal separation between the groups. The principal component of the WMIQ and CVI for that orientation was found, and plotted against the AQ. A second-degree polynomial regressor was applied and the residual for each individual was calculated. Sensitivity, specificity, diagnostic odds ratio (DOR), and prevalence-corrected (at 1%) positive and negative predictive values were calculated. The residuals for each individual were used in a Receiver Operator Characteristic, and the Area Under the Curve (AUC) was calculated.
Results:
It was found that classification of autism following deconstruction into a purported pathogenetic triad yielded high accuracy with a sensitivity of 90.5% and specificity of 90.0%. The DOR was 85.5 (lnDOR = 4.45, 95% CI [2.39-6.51]). The positive and negative predictive values were 0.084 and 0.999 respectively. The AUC was statistically significant at 0.963 (p < 0.0005, 95% CI [0.913-1.000], SE = 0.026).
Conclusions:
These results show the promise of deconstructing and operationalizing autism as a pathogenetic triad and its clinical feasibility in discriminating individuals with autism from neurotypical controls. A population-based study, with larger sample size, will be instrumental in proving the validity and clinical utility of the framework, as well as its specificity, as autism was the only patient group included. Strengths of the method include the simplicity, ubiquity and cheapness of the tests, as well as the short time needed for administration. If the high diagnostic accuracy were to be replicated, the method could easily be introduced into clinical work.