Article type
Year
Abstract
Background: Practicing evidence-based medicine requires succinctly summarized data, preferably in a meta-analysis if the data are appropriate. For continuous outcomes using different measurement instruments, systematic reviewers may calculate the standardized mean difference (SMD), ratio of means (RoM), or a more recent method based on mean difference between groups expressed in minimal important differences (MD-mid) units, where the MID is considered a constant. Although standardization in MID units is easily interpretable clinically, considering MID as a constant imposes important limitations.
Objectives: Our objective is to illustrate how considering the MID as a random variable, with a distribution, provides solutions to these limitations.
Methods: We calculated the variance of MD-mid using the delta method. Using sensitivity analyses, we compare results using the two methods of calculating the MD-mid variance.
Results: Considering the MD-mid as a random variable instead of a constant, 1) enables investigators to obtain estimates for questionnaires with no previous MID, and 2) makes underlying assumptions more transparent. Furthermore., considering MD-mid as a random variable allows investigators to avoid making the unrealistic assumptions that 1) the coefficient of variation for MID is independent of the measure, and 2) there is no correlation between the MID and mean difference. Using sensitivity analyses for different assumptions, we illustrate that the variance of MD-mid calculated when MID is considered a constant instead of a random variable can be under or over-estimated to significant degrees. We explore the effects on two different datasets, 1) data originally used to present the MD-mid, and 2) data from osteoarthritis studies using different pain scales and disability scales.
Conclusions: Considering the MID as a random variable instead of a constant makes underlying assumptions more transparent and accounts more appropriately for the true variance.
Objectives: Our objective is to illustrate how considering the MID as a random variable, with a distribution, provides solutions to these limitations.
Methods: We calculated the variance of MD-mid using the delta method. Using sensitivity analyses, we compare results using the two methods of calculating the MD-mid variance.
Results: Considering the MD-mid as a random variable instead of a constant, 1) enables investigators to obtain estimates for questionnaires with no previous MID, and 2) makes underlying assumptions more transparent. Furthermore., considering MD-mid as a random variable allows investigators to avoid making the unrealistic assumptions that 1) the coefficient of variation for MID is independent of the measure, and 2) there is no correlation between the MID and mean difference. Using sensitivity analyses for different assumptions, we illustrate that the variance of MD-mid calculated when MID is considered a constant instead of a random variable can be under or over-estimated to significant degrees. We explore the effects on two different datasets, 1) data originally used to present the MD-mid, and 2) data from osteoarthritis studies using different pain scales and disability scales.
Conclusions: Considering the MID as a random variable instead of a constant makes underlying assumptions more transparent and accounts more appropriately for the true variance.