Comparing Monitoring Devices When a Gold Standard is Unavailable: Application to Pulse Oximeters

• Activity Number: 437
• Type: Topic Contributed
• Date/Time: Thursday, August 7, 2003 : 8:30 AM to 10:20 AM
• Sponsor: Biopharmaceutical Section
• Author(s): Gene A. Pennello*+
• Companies: Food and Drug Administration
• Keywords: latent class ; Bayes ; method comparison ; agreement

Abstract

Method comparison studies compare two or more methods that measure the same quantity of interest. For example at FDA, comparisons of pulse oximeters have been made to determine the safety and effectiveness of such devices relative to control devices. Often a gold or truth standard is not available, and the usual analysis is one of agreement, e.g., a Kappa analysis. However, these analyses are unsatisfactory because agreement studies can only establish equivalence between two methods, not superiority of one over the other. Recently, latent-variable statistical methods have been published that enable estimation of the true value from repeated measurements of that value by two or more methods (e.g., Kupinski et al. 2002). Thus the measurements by the methods can be directly compared with the true values to determine which method, if any, is superior. We use latent-variable methods to compare pulse oximeter device measurements of arterial oxygen saturation of hemoglobin with the unknown true values. In contrast with published methods, we consider Bayesian methods to obtain small-sample inferences under a flexible distribution assumed for the true values.