Statistical analysis
Meta-analyses were carried out using STATA Meta-Analysis (V2.0, Biostat, Englewood, NJ, USA). The RoB graph was drawn by Review Manager 5.3 (Cochrane Collaboration Software, RevMan). The measure of treatment effect was calculated by the pooled odds ratio (OR) with 95% confidence intervals (CI) with random effects models. Intention-to-treat (ITT) analysis would be used when available. The number of trials in meta-analysis might not always be 49, because the included studies had multiple treatment or control types and the meta-analysis was undertaken based on each comparison.
Subgroup and meta-regression analyses were predefined with respect to age of participants (<35 or ≥35 years (28) ), different acupuncture types, type of control (no treatment or sham/placebo control), acupuncture timing (two types were categorized: type A, the acupuncture was performed around the time of ET (embryo transfer), which referred to the acupuncture treatment performed before and after ET (29); type B, before the ET); geographical origin of the study (Asian or not Asian), whether or not RIF (repeated IVF failure), acupuncture sessions (1, 2 or more than 3 sessions). Otherwise, the clinical characteristic previously was found to benefit trials with lower baseline pregnancy rates (20, 30), therefore, the stratified analysis was conducted by the pregnancy rate of no treatment group. Sensitivity analyses were also conducted to explore the robustness of the pooled effects of included studies: (1) restricting CPR results to the studies which simultaneously reporting LBR; (2) limiting the studies with low risk of bias; (3) deleting the studies whose primary outcome was not to evaluate the effects of acupuncture on pregnancy rates from IVF.
Heterogeneity was assessed using the I 2 test. Publication bias were shown by a funnel plot and assessed by the Egger weighted regression methods (p < 0.05 was considered statistically significant).