Some T-Tests for N-Of-1 Trials with Serial Correlation

• Activity Number: 436 - SPEED: Tests, Trials, Biomarkers, and Other Topics in Biometrics
• Type: Contributed
• Date/Time: Tuesday, July 31, 2018 : 3:05 PM to 3:50 PM
• Sponsor: Biometrics Section
• Author(s): Ji-Ling Tang* and Reid D. Landes and Anne Holbrook and Mark S Mennemeier and J. Tyler Floyd
• Companies: University of Arkansas for Medical Sciences and University of Arkansas for Medical Sciences and McMaster University and University of Arkansas for Medical Sciences and University of Central Arkansas
• Keywords: single-case experimental design; cross-over studies; time-series; repeated measures analyses; autocorrelation

Abstract

N-of-1 randomized trials allow inference between treatments given to an individual. Because these trials repeatedly measure the individual, serial correlation exists within a treatment. When treatments are paired, most existing methods account for correlation between treatment pairs (e.g., paired t-tests), but ignore serial correlation. Those that do account for serial correlation are often computer intensive, using simulations or resampling. Here, we develop t-tests that adjust for serial correlation both when treatments are paired and when they are sufficiently independent. These t-tests detect differences in treatment means (level-change) and treatment slopes (rate-change). Monte Carlo simulation demonstrates these tests hold nominal significance and power levels more closely than usual t-tests in realistically sized N-of-1 trials. We present examples of these tests in a biopharmaceutical and a medical device setting. These t-tests provide an easily implemented, formula-based method to account for serial correlation, which we hope will increase the application of appropriate statistical methods to N-of-1 trials in medical research.