Dr. Cohen raises a multitude of interesting points that may apply to many studies of pain.
Is a verbal report of the value represented on a visual analog scale (VAS) equivalent to pointing to the same number drawn on a line or marking a line mechanically? Studies have found absolute differences but good correlation between different methods in these rating scales. 1In our study, we accepted verbal responses because of the ease of obtaining such a verbal response at the necessary time, which was immediately after anesthesia and surgery. The potential difficulty in accurately pointing to or physically marking the intended number when postsurgical pain and residual anesthetic effects were present may have made this maneuver impractical.
The rationale for accepting a positive response when a patient could not respond with a number, either verbally or by pointing, is logical because some patients were in too much pain to care about the study while just emerging from anesthesia or were still too groggy to give anything except a positive response regarding the presence of pain. For this reason, we performed a sensitivity analysis substituting midrange (5), high (7), or low (3) values for positive responses, and recalculated the statistical analysis. For each of these values, the acetazolamide group had lower initial referred pain scores than the placebo group, demonstrating that the actual value of the positive responses was not important.
The question of how to statistically handle the VAS in a general sense, i.e. , whether to treat it as an ordinal value or a continuous variable, is important. Fortunately, this question of whether VAS scores can be evaluated using parametric statistical analyses, such as a t test or analysis of variance test, or should only be subjected to nonparametric analyses, such as the Wilcoxon rank sum test (nonparametric two sample t test) or Kruskal-Wallis test (nonparametric analysis of variance), has already been studied. Parametric tests compare means of data, and these data should ideally be normally distributed. The methods of analysis have been controversial, and early manuscripts from the 1970s and 1980s emphasized using nonparametric tests to avoid erroneous claims of difference when none existed. 2However, more recently, Dexter and Chestnut 3determined that the t or analysis of variance tests had slightly greater power than nonparametric tests to detect differences between groups and are good choices to compare VAS scores. Using actual data and computer simulations, Dexter and Chestnut further demonstrated that parametric analysis of as few as five categories of VAS data had power similar to that of a continuous VAS, although the use of fewer than five categories had reduced statistical power to identify a real difference. However, if large numbers of patients (> 16%) cluster at the extremes of the data range, nonparametric analyses may be less prone to error. 3
Applying this general question to our study, 445% (placebo group) and 78% (acetazolamide group) of patients had no referred pain at the initial postoperative measurement, and thus had a VAS of 0, representing no pain and hence a true, absolute zero. The VAS extended to 10, producing 11 categories, which exceeds the minimum of five found to be necessary by Dexter and Chestnut. 3In our acetazolamide study, the data clustered around the lower extreme but not both extremes, and the SDs were small. In addition to the t tests, we simultaneously performed a nonparametric analysis, and both methodologies had similar P values. Space limitations encouraged us to neither publish nor describe the details of the nonparametric analysis because the results were redundant. Table 1in this letter is a revision of table 2 from the original study. It includes the P values derived nonparametrically for comparison with the parametric results originally published. 4