Autism is a pervasive developmental disorder characterized by impairments in communication, language, and reciprocal social interaction and by patterns of restricted and repetitive interests or behaviours, as de?ned by the Diagnostic and Statistical Manual of Mental Disorders DSM-IV-TR. Several growth-related hormonal biomarkers including insulin-like growth factors and growth hormone binding protein have been shown to significantly elevated among autistic children.
This paper addresses the issue of combing multiple biomarkers such as growth-related hormonal biomarkers to improve diagnostic accuracy of a disease, such as Autism. Among existing methods in the literature, linear combination is perhaps the most popular and practical method for combining biomarkers achieve better diagnostic accuracy. Unlike the likelihood ratio, however, the optimal linear combination is not transformation-invariant. We investigate the effect of transformations on the combination and accuracy and proposed corresponding combination methods. Growth-related hormone data from a case-control study on autistic children are used to exemplify the methods.