(416g) Biomarker Identification in Autism Spectrum Disorder: Common Pitfalls and Emerging Strategies

Biomarkers are measurements obtained from a patient that are predictive of some phenomenon, such as a disease [1] or environmental exposure [2]. Biomarkers link complex observations to a quantitative measurement (e.g., loss of consciousness in diabetes with hypoglycemia) and can be used for a variety of purposes, such as assigning diagnostic status, stratifying patients, or measuring clinical outcomes [3]. Consequently, biomarkers are integral to the understanding and treating disease. However, biomarkers are frequently assessed with inappropriate analytics methods [4] and restricted to single measurements [5], which contributes to the underwhelming numbers of proposed biomarkers that lead to clinical translation [5]. These issues with current biomarker development/reporting are further discussed below.

Distributions of biomarkers in the diseased and healthy cohorts are typically described by measures of central tendency and some measure of the dispersion or range. Furthermore, the most commonly reported difference in the biomarker distributions for diseased and healthy cohorts is a difference in the measure of central tendency [3]. However, this “statistical significance” is usually misinterpreted as being synonymous with “clinical significance”. Instead, diagnostic biomarkers should be assessed on the separability of the diseased and healthy distributions (e.g. sensitivity/specificity, receiver-operating-characteristic curve, etc.) in order to achieve clinically-relevant biomarkers.

Potential biomarkers usually arise from discussions of or mathematical modeling of a biological network and how this network connects to the phenomenon of interest. Traditionally, only single measurements or metrics are used to construct a potential biomarker and the remaining portions of the underlying network is ignored [3]. While this approach can work well for simpler diseases, more complex diseases lacking a completely understood disease mechanism (e.g., neurological diseases) can benefit from incorporating multiple measurements into the potential biomarker without specific relationships between multiple measurements imposed a priori [3], [6].

These issues with current biomarker reporting/development are highlighted through case studies of biomarker identification from Autism Spectrum Disorder (ASD). It is shown that effective biomarkers can be developed from multiple measurements and that p-Values from Student’s t-tests can lead to incorrect conclusions regarding biomarker quality. In comparison to this, multivariate statistical analysis and modeling techniques can result in the identification of biomarkers that can be used for disease classification and severity prediction even when univariate approaches fail [6].

[1] T. J. Lyons and A. Basu, “Biomarkers in Diabetes: Hemoglobin A1c, Vascular and Tissue Markers,” Transl. Res., vol. 159, no. 4, pp. 303–312, Apr. 2012.

[2] N. L. Benowitz, “Cotinine as a biomarker of environmental tobacco smoke exposure,” Epidemiol. Rev., vol. 18, no. 2, pp. 188–204, 1996.

[3] J. C. McPartland, “Considerations in biomarker development for neurodevelopmental disorders,” Curr. Opin. Neurol., vol. 29, no. 2, pp. 118–122, Apr. 2016.

[4] A. D. Barker, C. C. Compton, and G. Poste, “The National Biomarker Development Alliance accelerating the translation of biomarkers to the clinic,” Biomark. Med., vol. 8, no. 6, pp. 873–876, Jul. 2014.

[5] G. Poste, “Bring on the biomarkers,” Nature, vol. 469, no. 7329, pp. 156–157, Jan. 2011.

[6] D. P. Howsmon, U. Kruger, S. Melnyk, S. J. James, and J. Hahn, “Classification and adaptive behavior prediction of children with autism spectrum disorder based upon multivariate data analysis of markers of oxidative stress and DNA methylation,” PLOS Comput. Biol., vol. 13, no. 3, p. e1005385, Mar. 2017.