Statistical Considerations in the Development of Companion Diagnostic Device

Since the concept of companion diagnostics was introduced in 1990s, personalized medicine has become more and more prominent. A well-established example of personalized medicine is the approval of the drug Herceptin in patients with breast cancer who tested positive for human epidermal growth factor receptor 2 (HER2). This is also one of the earliest examples of a co-development drug and companion diagnostic (hereafter CDx) model before there was a formal regulatory process in place. The device (or test) technology, cutoff points, and performance of a CDx can all vary, and thus different tests are likely to identify different populations for a given drug. These aspects are critical components to clinical trial design for CDx corresponding therapeutic product. Drug companies like to have a predictable regulatory route of drug development when companion diagnostic is involved. It is also important for drug companies to understand specific sets of statistical and clinical trial design questions to be addressed. In this chapter, we will discuss statistical considerations pertaining to companion diagnostics. In the first section, we will provide the overview of personalized medicine from statistical perspective. The second section discusses statistical issues relative to CDx development such as primary endpoint and data analysis need to support CDx clinical validation. In the third section, we will discuss statistical considerations including cutoff/threshold determination for CDx, Phase II/III clinical trial study designs; CDx bridging studies, and pre-screening bias. In the last two sections, we will present conclusion and recommendations. We hope that the book chapter will provide a good reference for statisticians, clinicians, and researchers from industry for statistical issues pertaining to companion diagnostic device in personalized medicine.