Comment This investigation, which concluded that intravenous midazolam, 75 μg/kg, reduced vomiting after tonsillectomy in children, had a number of strong design features, including large (more than 100 patients) sample sizes, a power level of 0.80, and precise application of the nonparametric Fisher's exact test to assess differences between the placebo-and midazolam-treated groups in unscheduled hospital admissions for vomiting. This investigation also had a number of design weaknesses, including investigator bias, confounding by the administration of anesthetics and analgesics with emetic effects, and underrepresentation of males in both study samples. Investigator bias was introduced in the topic sentence that “hypothesized that midazolam would decrease vomiting after tonsillectomy in children.” Investigators should have remained neutral, unbiased observers, prepared to test the null hypothesis of no effect (midazolam equals placebo) with appropriate test statistics. Fentanyl, nitrous oxide, and neostigmine were administered intraoperatively to all patients and have been associated with postoperative vomiting (POV). Although both placebo and midazolam-treated groups received the same anesthetics, confounding between anesthetic adjuvants and POV could have been eliminated by avoiding known emetics and by administering air/oxygen, nonnarcotics, and short-acting muscle relaxants that do not require reversal. The postoperative administration of codeine, another emetic, may have also confounded assessment of POV in the treatment groups. The authors correctly emphasized that POV is age-related and that their study groups were similar for age, eliminating confounding by age. Unfortunately, POV is also sex-related, more common in adolescent females than males, especially at menarche and during menses. There were more females than males in both groups (placebo, 52% female; midazolam, 53% female) again confounding proper assessment of pure treatment effects, not sex-related effects, on POV. Finally, the study groups were separate and distinct (unpaired) and were assessed for effects and outcome differences appropriately with the unpaired t-test. Where was the paired t-test (to assess differences in the same group) applied? A possible application would have been to evaluate hospital admission rates for vomiting in each group separately, with each group serving as its own control, but this was not reported.
Read full abstract