How to Determine Sample Size for IVD Studies

What sample size do I need?

Sample size impacts budget, feasibility, and success of regulatory applications for in vitro diagnostic tests (IVDs). At least as important is having an unbiased study – at minimum, the study sample should represent the intended use population and the “ground truth” against which performance of the IVD will be measured must be trustworthy. Once those are set up, sample size calculations can begin; they depend on several things including a study’s endpoints, expectations for how well the test will perform, and minimum acceptable performance. For example, a new test, “AVTA,” claims to detect awful virus type A.

An initial study was performed with approximately 1000 subjects, around half of whom were symptomatic. Approximately 30% of the subjects were actually infected, including some subjects who were asymptomatic. Sensitivity, the proportion of infected people who test positive, was estimated as 93.4%, and specificity, the proportion of people who are not infected and have negative test results, was estimated as 99.3%.

Are these high enough? What else do we need to know before we can decide? Prevalence of awful virus type A in the population is about 7%. Should the pivotal study be enriched with > 7% of subjects being symptomatic? How can sample size be computed to protect power against dropout? This example is discussed in the invited book chapter “Statistical Considerations in Clinical Trials for In Vitro Diagnostic Medical Devices”, a resource for non-statisticians based on Dr. Alicia Y. Toledano's 25-year history of obtaining FDA approval for medical diagnostic devices and co-authored by Meridith Blevins Peratikos, M.S., her long-standing collaborator. The chapter appears in the Food and Drug Law Institute (FDLI) book Diagnostics at a Crossroads: Navigating IVD Regulation in a Changing Environment.