Selection of studies
We searched MEDLINE using PubMed, Google Scholar, Scopus, Mednar, Wanfang and CNKI (China National Knowledge Infrastructure) databases for association studies as of August 11, 2020. The terms used were “Lysyl oxidase”, “protein-lysine 6-oxidase”, “LOX”, “polymorphism” and “cancer” as medical subject headings and text. References cited in the retrieved articles were also screened manually to identify additional eligible studies. In case of duplicates, the article with the most recent date was selected. Inclusion criteria were (i) case-control studies evaluating the association between the LOX polymorphisms and cancer risk and (ii) sufficient genotype frequency data presented to calculate the odds ratios (ORs) and 95% confidence intervals (CIs). The exclusion criteria were as follows: (i) reviews; (ii); (ii) articles that were not case-control studies; and (iii) studies with genotype data that could not be used to calculate ORs and 95% CIs.
Data extraction
Two investigators (RM and NP) independently extracted data and arrived at consensus. The following information was obtained from each publication: cancer group, family name of the first author, year of publication, the country of origin, ethnicity, LOX SNP, primary tumor site, study-specific association of the LOX SNP with cancer from each publication with their respective 95% CIs and P-values, status of the controls, genotyping platform, basis for matching the controls with cases, and study features needed to tally scores for the Newcastle-Ottawa Scale (NOS).
LOX polymorphisms and cancer groups
We examined two LOX polymorphisms in five cancer groups: -22G/C in (i) osteosarcoma (bone cancer) and 473G/A in the other four cancer groups that included (ii) breast, (iii) lung, (iv) gastrointestinal cancers (GIC) and (v) gynecological cancers (GYC). Three and two cancer types comprised GIC (oral, gastric and colorectal) and GYC (cervical and ovarian), respectively.
Quality of the studies
The NOS [20] was used to assess quality of the included studies. NOS scoring is based on three broad perspectives: selection, comparability, and exposure in case–control studies. The star rating system has scores ranging from zero (worst) to 9 (best). Scores of 5–6 and ≥7 stars indicate moderate and high quality, respectively.
Statistical power and Hardy-Weinberg equilibrium (HWE)
Using the G*Power program [21], we evaluated statistical power. Meta-analyses in cancer genetics have used the ORs of 1.2 and 1.5 to assess statistical power [22]. Thus, at these OR levels with a genotypic risk level of α = 0.05 (two-sided) and 5% minor allele frequency (maf), power was considered adequate at ≥ 80%. HWE was assessed with the application in https://ihg.gsf.de/cgi-bin/hw/hwa1.pl. A P-value of < 0.05 indicated deviation from the HWE.
Data synthesis
Examining two LOX polymorphisms (473G/A and -22G/C) warranted the use of a common notation indicating var and wild-type (wt) alleles. Supplementary Table S2 includes a column for the minor (var) allele in both polymorphisms. After estimating cancer risk (OR) for each study, pooled ORs with 95% CIs were calculated for each of the five cancer groups in the following genetic models: (i) homozygous: (var–var and wt–wt) genotypes compared with wt–wt, (ii) recessive: (var–var versus wt–var + wt–wt), (iii) dominant: (wt–wt versus wt–var + var–var), and (iv) codominant: (var versus wt). Three indicators were used for strength of evidence (SOE): First, highly significant P-values (Pa = 0.00001) most likely to survive the Bonferroni correction, which was performed with Microsoft Excel (Microsoft, Redmond, WA, USA). Second, highly precise effects were assessed with the confidence interval difference (CID = upper CI-lower CI). High (> 1.0) and low (< 1.0) CID values indicate low and high precision, respectively [23]. Third, homogeneity was assessed with the I2 metric, expressed as 0% [24]. In meta-analysis, however, studies differ from each other [25]. This heterogeneity was estimated with the c2-based Q test [26] where significance was set at PHET < 0.10. The random-effects model (DerSimonian–Laird) [27] was used in the presence of heterogeneity [24] and the fixed-effects model (Mantel–Haenszel) [28] in its absence. Summary effects that met the SOE criteria were tested for robustness, with use of sensitivity analysis, which involves serial omission of the studies followed by recalculation of the pooled OR. Significant outcomes (Pa < 0.05) with ≥ 10 studies warranted assessment for publication bias. Except for heterogeneity estimation [26] two-sided P-values of £ 0.05 were considered significant. Data for the meta-analysis were analyzed using Review Manager 5.3 (Cochrane Collaboration, Oxford, England), SIGMASTAT 2.03, and SIGMAPLOT 11.0 (Systat Software, San Jose, CA).