Data analysis
A meta-analysis was performed to synthesize community-based global
prevalence of infertility. To account for differences in study
participants and diagnosed infertility among the included studies, we
used random-effects models to calculate pooled estimates and 95% CI.
For studies reporting prevalence of primary and secondary infertility,
we also performed random-effects models to calculate pooled prevalence
and 95%CI between the two groups. Higgins’ I² statistic and
Q-test were used to detect heterogeneity across studies. An I ²
value greater than 50% or a p-value less than 0.05 indicated
significant heterogeneity. We used funnel plot to assess potential
publication bias, for which p <0.1 was regarded as
significant.
Subgroup analyses were conducted to estimate the infertility prevalence
for participants or studies with different characteristics. Subgroup
analyses were performed by study region (Africa, Asia, North Americas,
Oceania, and Europe), year of investigation (1940-1990, 1991-2000,
2001-2010, and 2011-2022), female age (before 35 years versus above 35
years), research type (cross-sectional study and cohort or prospective
study), and risk of bias (high, medium and low). Unordered
multi-categorical information was compared two by two using the
Bonferroni method, and grade information was analyzed for correlation
using chi-square trend test.
Univariate meta-regression analysis is used in the text to explore
possible sources of heterogeneity (Table S4). The dependent variable was
infertility prevalence and the independent variables were year of
investigation (dummy variable: 1940-1990), female age (dummy variable:
before 35 years), study region (dummy variable: Africa), research type
(dummy variable: cohort/prospective study), risk of bias (dummy
variable: high), or sample size (defined as a continuous variable). We
used a random effects meta-regression model with a restricted maximum
likelihood approach. The proportion of prevalence estimates explained by
any meta-regression model was estimated by the R² statistic.
To assess the stability of the results, we performed sensitivity
analysis using leave-one-out method to assess the dependence of the
findings on any individual study.