Abstract:
|
N-of-1 trials compare the effects of different treatments for a single individual. The individual may experience a pair of treatments in sequence, with the treatment order randomized in each pair of the sequence (e.g., a multiple cross-over design). Alternatively, a sequence of one treatment may precede a sequence of the other treatment (e.g., a bi-phasic or pre-post design). In both cases, responses from the treatments exhibit serial correlation. However, statistical methods often used to analyze data from these trials, such as paired or 2-sample t-tests, fail to account for serial correlation. Consequently, Type I error rates tend to be inflated over nominal levels. To address this, we developed some t-tests that account for serial correlation, and their Type I error rates were lower than those from usual t-tests. For N-of-1 trials with small sequence sizes, though, Type I error rates were still inflated. In this work, we present correction factors for these serial t-tests that bring Type I error rates closer to nominal levels. The serial t-tests corrected for small-sized trials provide accurate and accessible tests for making decisions in N-of-1 trials.
|