Stata statistical software package version 17

HRs and 95% CIs were obtained directly from each piece of literature. If the univariate and multivariate survival analysis results were simultaneously presented, we would choose the latter. In the absence of reported HR, we estimated the data according to the methods previously proposed by Tierney et al. or Parmaret et al. [28 (link),29 (link)]. The combined HRs and corresponding 95% CIs were used to evaluate the association between CPCs and prognosis parameters. A pooled HR > 1 without a 95% CI overlapping 1 indicated a significant association with poor outcomes.
Heterogeneity between studies was assessed by Cochran’s Q-test and I² statistics. A random-effects model was used when high heterogeneity was determined based on I² > 50% and Q-test p < 0.10; otherwise, the fixed-effect model was chosen. Subgroup analyses were carried out according to predefined variables that might affect the findings: region, sample size, cut-off value, detection time, initial treatment, and data type. Sensitivity analysis using the leave-one-out test was conducted to examine the robustness of the overall pooled results. Publication bias was evaluated by using Begg’s test. A p > 0.05 was defined as no obvious publication bias. All analyses were carried out using STATA statistical software package version 17.0 (STATA, College Station, TX, USA).