The heterogeneity of the combined HRs was evaluated with Cochran’s Q test and theHiggins Isquared
statistic. The HRs were considered statistically heterogeneous if they displayed P < 0.05
and/or I2 > 50% [30]. Significant heterogeneities among the studies were resolved with the random-
effectsmodel (DerSimonian-Laird method) [31]. Otherwise, the fixed-effects model (Mantel-
Haenszel method) was applied [32]. Subgroup analysis was conducted to determine the source of
existing heterogeneity. HR values> 1 indicated significant associations with poor prognosis. We
also examined the correlation between miR-21 expression and the clinical variables in breast cancer
through odds ratio (OR) [33]. Publication bias was analyzed using the funnel plot in combination
with Egger's test (P > 0.05 suggested a lack of publication bias). Finally, the influence of a single
study on overall HR was assessed for sensitivity analysis. All analyses were conducted with “Stata:
Data Analysis and Statistical Software” V12.0 (Stata Corporation, College Station, TX, U.S.).