Issues in developing multivariable molecular signatures for guiding clinical care decisions

Omics technologies that generate a large amount of molecular data characterizing biospecimens have the potential to provide information about patients' disease characteristics above and beyond standard clinical features. By combining information from a large number of features into a multivariable model, called a biomarker signature, there is the opportunity to identify distinct subgroups of patients for whom treatment decisions can be personalized. The key challenge is to derive a signature with good performance and appropriately characterize its performance. We summarize the key statistical issues and methods for developing and validating biomarker signatures, using examples from the literature for illustration.