3.6. Assessment of inconsistency

Due to the large number of interventions involved in this study, in the evidence network of each outcome indicator, the closed loop formed by studies with direct evidence and indirect evidence needs to be tested for inconsistency through Stata software. The inconsistency factor (if) was calculated, and the inconsistency was judged by the size of if value and P value.^[14] If the if is close to 0, 95% CI starts at 0, and P > .05, the results of direct comparison and indirect comparison are consistent. At the same time, the node split model is used to determine whether there is local inconsistency in each node.^[15] If P > .05, consistency model is used; otherwise, inconsistency model is used. For the consistency model analysis results, the stability of the results can be tested by the inconsistent model, when the inconsistent model factors include 0 and the inconsistency standard deviation. When the random standard deviation of the consistent effect model is approximately equal to the standard deviation of the inconsistent model, the consistency model results are more stable and reliable.^[16,17]