A microarray-based predictor is typically developed as a numerical index that combines the expression value of several signature genes. An established predictor in conjunction with one or more pre-selected cutoff points is used to define a categorical response of clinical interest (e.g. good/poor prognosis). Ideally, assessment of clinical outcome predictor reliability would be based the consistency of its predictive performance when the index is generated by two or more laboratories. We propose that the consistency of microarray-based predictors can be evaluated by assessing the agreement between the continuous indices generated by two or more different laboratories based on the overall concordance correlation coefficient for continuous measures. We present a methodology for power calculations of such studies and present results from simulated scenarios.