Class prediction models of thrombocytosis using genetic biomarkers

Criteria for distinguishing among etiologies of thrombocytosis are limited in their capacity to delineate clonal (essential thrombocythemia [ET]) from nonclonal (reactive thrombocytosis [RT]) etiologies. We studied platelet transcript profiles of 126 subjects (48 controls, 38 RT, 40 ET [24 contained the JAK2V(617)F mutation]) to identify transcript subsets that segregated phenotypes. Cross-platform consistency was validated using quantitative real-time polymerase chain reaction (RTPCR). Class prediction algorithms were developed to assign phenotypic class between the thrombocytosis cohorts, and by JAK2 genotype. Sex differences were rare in normal and ET cohorts (< 1% of genes) but were male-skewed for approximately 3% of RT genes. An 11-biomarker gene subset using the microarray data discriminated among the 3 cohorts with 86.3% accuracy, with 93.6% accuracy in 2-way class prediction (ET vs RT). Subsequent quantitative RT-PCR analysis established that these biomarkers were 87.1% accurate in prospective classification of a new cohort. A 4-biomarker gene subset predicted JAK2 wild-type ET in more than 85% patient samples using either microarray or RT-PCR profiling, with lower predictive capacity in JAK2V(617)F mutant ET patients. These results establish that distinct genetic biomarker subsets can predict thrombocytosis class using routine phlebotomy. (Blood. 2010;115:7-14)