2.1 Data sources, search strategy, and selection criteria
This review was performed according to the Meta-analysis of Observational Studies in Epidemiology guidelines issued in 2000 [25] (PROSPERO Registration Number: CRD42019128195). Two authors systematically searched electronic databases to identify all studies related to the impact of rs1045642 and rs1128503 on the overall survival (OS) and progression-free survival (PFS) of patients receiving taxane-based chemotherapy. The electronic databases PubMed, Embase, Web of Science, Cochrane library, Chinese National Knowledge Infrastructure, Wanfang, and Weipu were searched from the time of library construction to August 2019 with the following core search terms: (taxane or taxol or paclitaxel or docetaxel or Albumin-Bound or Protein-Bound) and (outcomes or “overall survival” or OS or "Progression Free Survival" or PFS) and (“Multidrug Resistance Gene” or ABCB1 or MDR1). We conducted a manual search of the reference lists of retrieved studies or relevant reviews to identify additional published studies. To identify unpublished literature, databases of abstracts were searched, and the authors were contacted. Only the most recent study was included when overlap occurred between studies.
Studies were considered eligible if they fulfilled all of the following criteria: participants: all included patients were diagnosed with Solid tumor; intervention: all included patients were administered taxane-based chemotherapy; control: all studies included comparisons of different categories of the polymorphic genotypes ABCB1 C3435T and C1236T; outcomes: all studies reported the OS and PFS of patients with different genotypes, and studies with enough genotype data to estimate the hazard ratio (HR) and 95% confidence interval (CI) in at least one genetic comparison model. The study selection process was performed by 2 authors independently; a third author determined the final criteria for any inconsistencies.
2.2. Data collection and quality assessment
Two authors were responsible for extracting the data from eligible studies using a standardized data extraction table. Disagreements were resolved by group discussion or a third author if a consensus could not be reached. Information on the first author’s name, publication year, country, patient mean age, sample size, tumor type, regimen, detection, Hardy-Weinberg equilibrium, PFS, and OS for each category of genotypes was collected. Bias caused by individual studies was examined by 2 authors independently according to the Newcastle–Ottawa Scale score, which is useful for comprehensively evaluating the quality of observational studies in meta-analysis. The bias of selection (4 items), comparability (1 item), and outcome (3 items) were assessed during this process.
2.3. Data synthesis and analysis
Three genetic models were analyzed for each gene site in this meta-analysis: homozygote model (TT vs. CC), heterozygote model (CT vs. CC), and dominant model (CT + TT vs. CC). For time to event survival analysis, we assessed the effect of ABCB1 status on patient prognosis by calculating the hazard ratio (HR). For each study, the HR and its 95% confidence interval (CI) were retrieved. If these parameters were not available in the studies, we used the software Engauge Digitizer 4.1 to extract specific survival rates according to Kaplan-Meier curves to calculate the HR as described by Tierney et al. [26]. Heterogeneity in the pooled analyses was determined by statistical analyses using the Q statistic for homogeneity and the I2 statistic. A P value of < 0.10 or I2 > 50% was considered to indicate significant heterogeneity [27]. When statistical heterogeneity existed, the random-effects model was used; otherwise the fixed effects model was applied. Sensitivity analyses were performed to evaluate the influence of single studies on overall analysis. Subgroup analyses, including area (Asia, Europe, and others), tumor types (ovary, breast, lung, and others), and regimen (paclitaxel plus cisplatin (TP) and others) were conducted. Publication biases were estimated by using Egger tests [28]. A two-sided P value of less than 0.05 was regarded as statistically significant. All statistical analyses were performed using R version 3.6.1 software.