Often, in medicine, there is some disease defined as a function of a biomarker, eg For risk models, the disease is then modeled as a binary random variable. Why not model the biomarker directly (eg as a continuous rv, if the biomarker is continuous)? See more here.