Stata software version 12.0 se

Data analysis was done using STATA software, version 12.0 SE (Stata Corporation, TX, USA). A weighted survey analysis was done due to the fact that this was survey data. Sampling weights used were the pweights; which denote that the inverse of the probability that the observation is included due to the sampling design and or non-response. Firstly descriptive statistics were done to determine the frequencies of late antenatal care booking. This included cross tabulation to determine the overall distribution of predictor variables by early or late ANC booking. This was followed by univariate logistic regression. Significance at univariate logistic regression was set at a p-value of 0.1 and a 95 per cent confidence interval. Variables that were found to be significant at univariate logistic regression were then fitted into the multiple logistic regression to control for confounding and to come up with the final model of predictor variables. For multiple logistic regression, significance was set at a p-value of 0.05 and 95 per cent confidence interval. Variables with the largest p-values were then removed one at a time until only significant variables were left in the final model. The analysed information was summarised using tables and graphs.

We used STATA software version 12.0-SE (StataCorp, College Station, TX) to perform our analyses. We first assessed the inter-study heterogeneity using Cochran’s Q statistic and the I² test. A P-value of Cochran’s Q statistic < 0.1 or I² value > 50% was considered to show a high level of heterogeneity. We thus used the DerSimonian–Laird association test with a random-effects model. Otherwise, we used the Mantel–Haenszel association test with a fixed-effects model. The P-value of association test, summary odds ratio (OR), along with the corresponding 95% confidence interval (CI) could be obtained for the allele (T compared with C), homozygous (T/T compared with C/C), recessive (T/T compared with C/C+C/T), heterozygous (C/T compared with C/C), dominant (C/T+T/T compared with C/C), and carrier (T compared with C) models.
We performed subgroup analyses by race, cancer type, and control source. Additionally, we assessed possible publication bias by means of Begg’s and Egger’s tests and evaluated the robustness of the results through sensitivity analysis.