A physician is conducting a double-blind randomized controlled trial to determine the effect of a new cream in reducing the risk of relapse in chronic recurrent atopic dermatitis. A total of 30 patients with moderate to severe atopic dermatitis who were experiencing a flare are randomly divided into 2 groups: 15 patients will receive the new cream, and 15 patients will receive emollient alone. The rate of relapse after 2 weeks of treatment is 25% in the group who received the new cream and 50% in the group who received emollient alone. However, the difference is found to be not statistically significant (p = 0.14). The physician concludes that use of the new cream does not reduce the risk of relapse in chronic recurrent atopic dermatitis. Which of the following is most likely to explain the results of the study?
Statistical power represents a study's strength to detect a difference (ie, effect size) between treatment groups when one truly exists. It depends on sample size (among other factors): studies with greater sample sizes have greater power than studies with smaller sample sizes. An excessively large sample size may determine that a clinically irrelevant difference between groups is statistically significant (ie, p < 0.05), while an inappropriately small sample may fail to determine that a clinically relevant difference between groups is statistically significant (ie, p > 0.05).
In this case, the rate of relapse in the group receiving the new cream is 25% and the rate in the group receiving the emollient alone is 50%. This represents a relative risk reduction (RRR) of 50% (ie, [50% − 25%] / 50%); therefore, the new cream reduced the risk of relapse by 50% (ie, effect size) compared to emollient alone. A 50% risk reduction may be considered clinically relevant; however, it was not statistically significant (ie, p = 0.14 > 0.05) in the study.
The most likely explanation is that the sample size (ie, 15 per group) provided insufficient power to detect the observed difference between groups (ie, effect size, RRR = 50%). A larger sample size would increase the power of the study (ie, its ability to detect the difference), and the p-value would reach statistical significance (ie, p < 0.05).
(Choice A) Ascertainment bias occurs when the results of a study are distorted by awareness of treatment assignment, as in an unblind study. This study is double-blinded, thereby minimizing potential ascertainment bias.
(Choice B) Confounding distorts the relationship between exposures (eg, treatments) and the outcome (eg, disease) of interest, and can wholly or partially account for observed effects. This study is a randomized trial, thereby minimizing potential confounding bias by generating groups that are comparable with respect to known and unknown confounding variables.
(Choice C) Ecological fallacy occurs when conclusions are made about individuals based on studies where the unit of analysis is a group (ie, the conclusions of a study assessing population groups do not necessarily apply to individuals). In this study, the unit of analysis is the individual, not the group.
(Choice E) Recall bias results from the inaccurate recollection of past exposure status. It is a potential problem for case-control studies, particularly when questionnaires are used to inquire about distant past exposure. However, this study is an experiment in which patients are exposed to treatments and then assessed for the outcome.
Educational objective:
An inappropriately small sample will fail to identify important clinically significant differences as statistically significant because of a lack of sufficient statistical power.