SynapDx Autism Gene Expression Analysis Study (STORY)

The primary aim of this study is to define a gene expression signature indicative of ASD and to establish its clinical sensitivity and specificity. Clinician diagnosis of ASD will be made using DSM-5 as the reference standard instrument. The widely used diagnostic instrument Autism Diagnostic Observation Schedule (ADOS-2) is a typical component of the clinical assessment for a child diagnosed with ASD and this evaluation will be performed on all study participants. Secondary aims of this study are (1) to define metabolite and lipid signatures indicative of ASD and establish their clinical sensitivity and specificity and (2) to determine clinical sensitivity and specificity of various combinations of gene expression signature, metabolite and/or lipid signatures, and presence of ASD-associated genetic variation detected by CMA and sequencing protein-coding regions of the genome

Analyses: Details of the analysis will be specified in a Statistical Analysis Plan (SAP), which will include procedures for handling outliers, missing data, and differences across sites. The SAP will be reviewed and approved by a committee of Principal Investigators (PIs) before unblinding of the validation set.

Primary analyses: The primary outcomes of the study will be the estimates of the clinical sensitivity and specificity of the SDX-002 test to classify subjects according to DSM-5 ASD diagnosis, with associated 95% confidence intervals. Sensitivity and specificity will be assessed on the Validation Phase population based on agreement with the clinical diagnosis of presence or absence of Autism Spectrum Disorder by DSM-5 (published May 2013).

The gene expression signature will be trained on the 500 subject Development Phase set, using 5-fold cross validation over the results of several machine learning algorithms, including partial least squares, support vector machines, and boosted decision trees. The training procedure will generate estimated ROC curves for each method, as well as confidence intervals for the area under the curve (AUC). The final choice of a machine-learning algorithm will be based on AUC, as well as on the estimated performance at the chosen operating point on the ROC curve. The operating point will be chosen to provide high sensitivity at an acceptable specificity.

Secondary analyses: In the majority of patients enrolled to date, consent was obtained for collection of an optional bio-repository sample. The intended analysis of the bio-repository samples has now been established (see also, secondary objectives listed above). The metabolomic and lipomic signatures will be similarly assessed on the subset of the 500 subjects in the Development Phase set who consent to a bio-repository sample. In addition to the gene expression signature, metabolite and/or lipid signatures and DNA analysis will be combined with the gene expression signature in various configurations and the impact of these additional measures on clinical sensitivity and specificity will be evaluated. If these additional metabolomic/lipomic signatures and/or DNA analyses improve test performance, these elements may be included in the SDX-002 assay.

Sensitivity and specificity the final SDX-002 assay will also be assessed among subpopulations using demographic information and the results of developmental testing. There are likely to be few subjects in many of these subpopulations and caution will be used in interpreting the results. Planned subpopulation analyses include gender, age, ethnicity, ASD DSM-IV-TR diagnostic subcategory, DSM-5 ASD severity level (social communication and restricted interests/repetitive behaviors) and ADOS-2 scores.