Small sample size: statisticians tackle the problem

Vaccine and Infectious Disease Division

Small sample size: statisticians tackle the problem

A common vaccine clinical trial structure includes enrollment of the cohort, administration of the vaccine and follow-up of cohort status. This process yields tens of thousands of samples to analyze for efficacy, immune response and/or adverse effects, for example. Often, blood or plasma samples from patients are very small, which makes it difficult to perform multiple assays. As such, methods must be developed to produce statistically significant data from these minute and precious samples. A well-known such assay is an Enzyme-Linked Immunosorbent Assay (ELISA), which uses an antibody-binding, color change readout for protein content. Researchers at the HVTN have developed a more sensitive, high-throughput assay that requires an even smaller sample size than an ELISA and can measure the concentrations of up to 100 molecules (versus 1 for ELISA) at a time. This assay is called a Multiplex Bead Array (MBA).

With these types of assays, the need for statistical models becomes more essential; results from assaying small volumes of sample can be drastically affected by slight technical errors. To address these issues, statisticians use samples with a known amount of a compound or molecule at different concentrations and then can produce a fitted curve. From these data, they can make predictions as to molecule concentration and use calibration methods to deal with “experimental noise.” VIDD assistant member Youyi Fong, VIDD associate member Stephen DeRosa and VIDD assistant member Nicole Frahm developed a novel statistical model to address these issues and used data from the Rv144 trial to confirm their model.

MBA was used to study immune correlates from the Rv144 clinical trial. Specifically, HVTN researchers measured the amount of several dozen cytokines/chemokines, which are important immune response proteins. The results from this study showed their statistical model led to a substantial increase in prediction ability. This statistical model will be useful not only in clinical trial analysis, but in other high-throughput experimental systems, such as pharmakodynamic modeling.

Fong Y, Wakefield J, De Rosa S, Frahm N. A Robust Bayesian Random Effects Model for Nonlinear Calibration Problems. Biometrics. 2012 May 2.