2.4 Data synthesis
We chose the ischemic heart disease as the main disease to analysis. The studies included were summarized according to their RRs with corresponding 95% CI . The study-specific log RRs were weighted by the inverse of their variance, and the weighted values were used to assess the association between hysterectomy and risk of CVD. Furthermore, pooling was also performed for studies on different ovaries status (hysterectomy with oophorectomy or hysterectomy with ovarian preservation) and studies on age at hysterectomy (divided by 50 years old). In our analysis, hysterectomy with unilateral oophorectomy was classified as the category of hysterectomy with ovarian preservation, and hysterectomy with bilateral oophorectomy was classified as hysterectomy with oophorectomy. Overall RRs with 95% CI in each study, instead of stratum-specific values (eg, separate estimates for hysterectomy with oophorectomy and hysterectomy with ovarian preservation), were used to assess total pooled RR of this meta-analysis. If an overall value was not presented in the article, the stratum-specific values were used following the principle that the populations in the different strata did not overlap.
Assessment for among-study heterogeneity was performed by calculatingI2 which represents no, low, moderate and high heterogeneity when its values are 0, 25, 50 and 75% respectively.33 The fixed-effect model was used if moderate or lower heterogeneity (I2 ≤50%) was found. If I2 >50%, the random-effect model was adopted. Pre-specified characteristics including publication year, continent and study design for exploring potential sources of heterogeneity were used in meta-regressions and subgroup analyses.34 The sensitivity analysis which useI2 >50% as the criteria was applied to evaluate the excluded studies that have substantial impact on between-study heterogeneity.35 We also assessed the relationship between hysterectomy with or without oophorectomy and CVD, age at operation and CVD. The influence analysis was performed with one study removed at a time to assess whether the results could have been affected markedly by a single study.36 The small-study effects were detected by Egger test37 and funnel plot. All statistical analyses were carried out with STATA version 15.0 (Stata Corporation, College Station, TX, USA). All reported 2-tailed P values were considered statistically significant when P≤0.05.