|
Abstract:
|
Binary regression models, such as logistic models, are widely used in biological research, in areas such as prognostic modeling in clinical genomics. However, when there is a significant rate of misclassification of the reported status of binary response variables, such models can fail miserably. We address this in utilization of a logistic model adjusting for misclassification of response variables. The adjusted model can be fitted either using MCMC methods for posterior evaluation, or using Newton-Raphson algorithm to directly compute MAP/ML estimates. As an example, we have applied and evaluated the adjusted model in predicting lymph node (LN) status of breast cancer patients. There is a 10-20% false-positive misclassification that this approach addresses and adjusts for. We discuss some computational issues in evaluating the adjusted models as well as this application.
|
