Using a latent class model to refine risk stratification in multiple myeloma

In multiple myeloma research, the GEP70 model is known to be capable of predicting a high risk patient group for disease progression based on the expression levels of 70 selected genes measured at baseline. The model consists of a continuous gene score that is a linear combination of the 70 genes along with a cutoff, such that patients with a score greater than the cutoff are categorized as high risk and otherwise low risk for disease progression. However, the continuous gene score may be confusing at times because of its open range nature. In addition, the present two-group model is sensitive to scores falling close to its cutoff. To facilitate patients' understanding of their prognosis, it is desirable to convert the continuous score into a probability that has an easier interpretation. In this article, we employ a latent class model to address this issue, and we also propose a superior grey zone model to refine the current risk stratification associated with the GEP70 model. Lastly, we demonstrate the robustness of the grey zone model with results from a simulation study.