Date created: | Last Updated:

Category: Project

Description: Standardized mean difference (SMD) is used to demonstrate the effect size of a continuous outcome in randomized controlled trials (RCTs). SMD is particularly useful in comparing effect sizes between RCTs, where different instruments are utilized regarding a common outcome. Besides, Cohen’s rule of thumb is frequently referenced when interpreting SMD (0.2, small; 0.5, medium; 0.8, large). However, there are multiple approaches to generate SMD, because either endpoint or change score mean can be used as the numerator, while either endpoint, change score or baseline standard deviation (SD) can be used as the denominator, when calculating SMD. If SMD computed from different approaches differs substantially from each other, it would be dangerous to compare the reported values from different RCTs directly, because they may compute SMD differently. It would be even worse if the original RCT does not clarify the method clearly. Additionally, applying Cohen’s rule of thumb as a common standard to interpret SMD based on various approaches may become inappropriate. Therefore, we aim to investigate how original RCTs have calculated, reported and interpreted SMD, and how different the values would be if using different approaches. We will conduct a survey of RCTs published in high-impact journals that report SMD for any results. We will explore how they compute SMD and report the method. We will also calculate SMDs using different approaches based on different sets of reported means and SDs, and compare them with the reported SMD. Then, we will conduct a simulation study based on empirical data, to calculate the difference between several prespecified SMD approaches for each study. We will run a meta-analysis of the difference to examine its pattern. Our hypothesis is different SMD approaches may vary, thus comparing it among different studies or to a common standard should be with caution. And RCT authors should provide adequate details about the method they use as well as different means and SDs so that readers or meta-analysts can calculate the particular SMD that fits their own research hypothesis.