2.5. Statistical Methods

Meta-analysis of summary statistics from individual studies was performed utilizing the Medical Package program of STATISTICA 11.0 software (StatSoft Poland, Krakow, Poland). For each study, we constructed separate two-by-two (2 × 2) contingency tables to calculate the odds ratios (OR) and 95% confidence intervals (CIs), cross-classifying OC users and occurrence of breast cancer. The Mantel-Haenszel test was calculated to assess the association between OC use and cancer. An OR of one indicates that the odds of having breast cancer are the same in the case group and the control group; an OR greater than one indicates that the odds of having breast cancer in the case group are greater than in the control group; an OR of less than one indicates that the odds of having breast cancer in the control group are greater than in the case group. Because about one-third of the studies did not present adjusted ORs, only crude ORs were used in the primary meta-analysis. Meta-analyses combining the ORs across studies were conducted using the DerSimonian–Laird random effects model [28]. The random effects meta-analysis model was applied due to the diversity of research in terms of, for example, design and population. In the random effects model, it is assumed that there is no common effect size for independent studies. Instead, each study is assumed to have a different population effect size, which is a random variable and has a normal distribution. Therefore, there is a difference between the size of the effects of individual studies. Thus, the variance of the effect size in the random effects model is the sum of the variance within and between studies. As suggested by DerSimonian and Laird [28], the variance ‘between studies’ is estimated using the moments method. Weighting of the studies in the meta-analysis was calculated on the basis of the inverse of the sum of ‘within study’ and between studies variances. Heterogeneity was assessed graphically using forest plots and statistically using the Q test and I2 index. I2 values of 25%, 50%, and 75% were regarded as respectively representing low, moderate, and high heterogeneity between studies [29,30].

Due to the high heterogeneity of the studies, the following subgroups were also analyzed: case-control studies of the period of recruitment into the study before 1986/case-control studies of the period of recruitment into the study after 1986; premenopausal women or women younger than 50 years/postmenopausal women or women older than 50 years/women under 25 years old; nulliparous women; ever use of oral contraceptives before first pregnancy; and ever use of oral contraceptives for longer than 5 years.

For all the analyses, forest plots were generated to display results, whereby diamonds represent study-specific odds ratios and 95% CIs for individual studies are represented by horizontal lines.

To analyze the publication bias, the Begg and Mazumdar and Egger tests were performed. The meta-analysis methodology was based on the guidelines of DerSimonian and Laird [28] and Higgins et al. [29,30]. The method used is globally recognized as the primary method for the evaluation of case-control studies.