The Impact of Significant Digits in Bioequivalence Studies

In biomedical research, various statistical methods are used to analyze data collected from studies to provide statistical inferences of the unknown parameters of interest. The analysis results up to a few decimal places are usually obtained for scientific validity. The use of different number of decimal places, however, may alter the conclusion of the analysis results. A typical example is the assessment of bioequivalence for generic approval in bioequivalence studies. In practice, there is no uniformly accepted cut-off standard for the numbers of decimal places in biomedical research. In this paper, following the concept of signal-noise in quality control and assurance, we propose a simple procedure in determining the number of decimal places for providing a valid statistical inference and assurance in biomedical research. Based on a non-central t distribution, the proposed procedure is able to control the largest rounding error when considering rounding up the data to certain decimal places. In other words, the proposed method could control the worst possible error adjusted for standard deviation by selecting an appropriate digit that meets the criteria that the non-central parameter is less than a pre- specified critical value. Statistical justification of the proposed procedure was provided and illustrated through a simulated dataset.

Keywords: Analytical Research; Decimal Places; Non-Central t Distribution; Transformation