A Special Inference Problem in Repeated Measures Design-Test of Statistical Hypothesis on Accuracy Root Mean Square-Application to Pulse Oximetry Studies

• Published in: Statistics in biopharmaceutical research 2016-01, Vol.8 (1), p.60-76
• Main Authors: Ndikintum, Nfii K., Rao, Marepalli
• Format: Article
• Language: English
• Subjects: Devices; Hypothesis testing; Measurement error; Mixed effects models; Simulation
• ISSN: 1946-6315; 1946-6315
• DOI: 10.1080/19466315.2015.1101012

The objective of pulse oximetry studies is to demonstrate that paired repeated measurements from a pulse oximeter (SpO 2 ) and a co-oximeter (SaO 2 ) agree sufficiently with respect to a prespecified agreement score (A RMS ). The current methods for analyzing pulse oximetry method comparison studies are mostly descriptive. Sample size calculations in the context of a specific test of statistical hypothesis applied to the A RMS have been reported. The purpose of this article is to further study the distributional properties of the A RMS and recommend a test appropriate to this agreement score. Five tests are considered: Z-Wald test, Z-score test, chi-squared (score) test, chi-squared (Wald) test, and "Exact" test. We propose and evaluate via simulation studies a robust method for testing the hypothesis of the accuracy on a pulse oximeter. Overall, the Z-score test is superior to the other four tests, resulting in estimated α-levels that are closest to the nominal α-level. The chi-squared (Wald) test is liberal and the "Exact" test is conservative. The chi-squared (Wald) test provides better power compared to the score and Exact tests. The Z-score test is a viable method for assessing the accuracy of pulse oximeters.