Design of Cross-Sectional Method Comparison Studies

• Activity Number: 66
• Type: Topic Contributed
• Date/Time: Sunday, July 31, 2011 : 4:00 PM to 5:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Farzad Noubary*+ and Michael Hughes
• Companies: Massachusetts General Hospital and Harvard School of Public Health
• Keywords: method comparison ; design ; sample size ; bias ; CD4 / HIV ; Bland-Altman

Abstract

Motivated by the development of point-of-care CD4 count enumeration technologies to inform clinical decision-making for HIV-infected patients, we discuss the design of cross-sectional method comparison studies to evaluate novel methods of measurement. We present formulae for the sample sizes required to estimate bias or variability in the novel method with desired precision, first by assuming that these statistics are constant across the range of measurement. We then relax this assumption using local regression methods that allow us to assess how bias changes with magnitude of measurement. The corresponding sample size formulae then require a minimum number of subjects in different strata of the outcome. To facilitate the accrual of patients with values in the tails of the distribution, we propose the use of variable probability sampling and discuss the analysis modifications required for such a design. We then show that the intuitive application of this sampling scheme based on patients' previous values from the gold standard can lead to misleading bias calculations because of regression to the mean.