Network meta‐analysis compares multiple treatments in terms of their efficacy and harm by including evidence from randomized controlled trials. Some trials use within person designs, where each patient receives more than one treatment. Data from treatment arms within these trials are no longer independent. As a result, adjusting weights for dependent treatment arms within trials within special design trials are essential to obtain accurate results. Ms. Yu-Xuan Su, a graduate from the Master degree program, and her advisor Prof. Yu-Kang Tu, in the Institute of Epidemiology and Preventive Medicine at the College of Public Health, National Taiwan University, proposed three methods, the data augmentation approach, the adjusting variance approach, and the reducing weight approach to resolve this problem. Their study has been published on July 6 in Research Synthesis Methods.
Most clinical trials adopt the parallel group design, where each patient is randomly assigned to one treatment group. For such trials, the calculation of a treatment contrast and its variance is straightforward. However, some clinical trials adopt special designs such as split‐body, split‐mouth, and crossover designs, where patients may receive more than one treatment in a randomly determined order. In such trials, the correlations between effects of treatments on the same patients must be considered, since observations are no longer independent. However, the issue of data dependency becomes more complex when multi-arm trials are included in a network meta‐analysis. Moreover, their correlations are already known and need to be taken into account in the data analysis. This not only makes the adjustment of data dependency more complicated, but also poses a challenge to preparation of data prior to the analysis.
The three methods proposed by Ms Su and Prof Tu can be easily implemented in commonly used statistical software packages such as R and STATA. They demonstrated that the proposed approaches yielded similar results. When the number of trials is large and the maximum number of treatments compared is greater than 3, the adjusting variance approach will be the most efficient way to make multiple adjustments simultaneously. All these approaches, when implemented into software packages, could be very useful tools for undertaking network meta‐analysis with special designs.