Our protocol is registered in The International Database to Register your Systematic Reviews (INPLASY). And the register number is INPLASY202160001.
2.1.1 Study Inclusion and Data Extraction
We launched the meta-analysis using the keyword “Vitamin C” and “prostate cancer” from Pubmed, Cochrane Library, and EMBASE. The articles up to 2021 Jan were all included. The terms were selected referring to the previous study, and the prebious study’s reference was also reviewed to identify additional relevant study . Only RCT and prospective cohort researches were selected.
The criteria were as follows: 1) prospective cohort research or RCT; 2) studied the correlation between prostate cancer incidence and vitamin C intake (because no RCT or cohort studies have studied the effect of plasma vitamin C and vitamin C intake can also release the plasma vitamin C level); 3) reported risk estimates (odd ration (OR) or hazard rate (HR)). The excluded articles included:1) non-English articles; 2) articles of basic experiments; 3) reviews or retrospective studies; 4) no accessible data.
2.1.3 Data Extraction
Participants’ baseline data, including age, sex, basic diseases, outcomes, etc., were extracted to record. The outcome was defined as the first diagnosis of prostate cancer, while the previous diagnosis and benign prostate hyperplasia (BPH) were excluded. We tried to contact the authors by email if essential data are not reported.
2.1.2 Quality Assessment
Two researchers reviewed the existing articles and extracted the data independently. The quality study assessment was processed according to the allocation of patients to the study’s aims, the concealment of allocation procedures, blinding methods, and the data loss by attrition. Quality assessment was performed according to Cochrane Handbook for Systematic Reviews of Interventions 5.1.0. The detection bias (assessment of outcome blinding), selection bias (random sequence generation and allocation concealment), attrition bias outcome data incompleteness), reporting bias (selective report), performance bias (blinding of participants and personnel) and other bias were evaluated in RevMan 5.3. The quality of articles was categorized by the following standard: A. low risk of bias: the article can meet all quality criteria were adequately; B. moderate risk of bias: the article can only partially meet the quality criteria or is unclear; or C. high risk of bias: if one or more of the criteria were not met, or not included.
2.1.3 Statistical Method
The data were analyzed by STATA 16.0. Included studies’ effect was pooled by fixed-effect meta-analysis, the pooled effect was showed by the pooled OR and 95% confidential intervals (CI). Cochran Q test was applied as the investigator of heterogeneity from the extracted meta-analysis estimates. The significance of heterogeneity was evaluated by value (considered significant when >35%) and p-value (considered significant when <0.05) in advance. When the heterogeneity is insignificant, we use the fixed-effect statistical model to process the meta-analysis. Otherwise, the random-effects statistical model was chosen. Sensitivity analyses were conducted to assure all included studies have a low risk of bias.
2.2.1 Tool variables
The association between single-nucleotide polymorphisms (SNPs) and plasma vitamin C level was estimated through a large-scale GWAS meta-analysis with 11 SNPs found (n = 52,018), including the Fenland study (10,771 cases), European Prospective Investigation into Cancer and Nutrition (EPIC)-InterAct study (16,841 cases), EPIC Norfolk study (16,756 cases), and EPIC-CVD study (7,650 cases), with all duplicated samples excluded . Though specific pathways or enriched tissues were not found, the correlation and quantity of these SNPs were still considerable progress among other studies [9-12]. The Fenland study is a population-based study which has an ongoing cohort . The EPIC-InterAct study and the EPIC-CVD study are both case-cohort studies, while the EPIC Norfolk study is a prospective cohort study [14-16]. All genotypes were investigated by Haplotype Reference Consortium reference panel with IMPUTE v4 or the Sanger imputation server. The tool genetic instruments used in the 2-sample Mendelian Randomization were identified as single-nucleotide polymorphisms correlated with citrates (P<5× ), and the pair-wise linkage disequilibrium was significant with <0.001. The Two Sample MR package of R is applied on clumping analysis. F statistic was also calculated by the EPIC Norfolk study to estimate the instrument’s strength, and F>10 was considered a vital instrument.
2.2.2 Outcome sources
The outcome data of prostate cancer patients were obtained from two studies, the PRACTICAL (79,194 cases and 61,112 controls), and the FinnGen project (4,754 cases and 63,465 controls). Prostate cancer was defined as the diagnosis of invasive prostate cancer in the Prostate Cancer Association Group to Investigate Caner-Associated Alterations in the Genome (PRACTICAL) and the diagnosis of prostate cancers according to the Finnish Cancer Registry with the standard of International Classification of Diseases for Oncology -3 (ICD-O-3) codes FinnGen. The following covariates were included: age, family history and tumor characteristics in the PRACTICAL; age, sex, 10 principal components and genotype batch in the FinnGen consortium. Logistic regression was applied to calculate genetic association estimates by comparing cases and controls and was adjusted for principal genetic components.
2.2.3 Statistical Method
Instrument strength means the level and accuracy of association of the genetic instruments with the risk factor, and if the corresponding F-statistic is >10 we consider is as an instrument strong enough. Two-sample Mendelian randomization analysis used plasma vitamin C concentration-rising SNPs and the effect of those SNPs on the risk of prostate cancer. The correlation of incidence and vitamin C-associated genes was tested using the IVW model. Scatter plots were applied to find the influential outlier. The result of MR-Egger can still exhibit estimates without bias even in the condition that all SNPs disobey the exclusion restriction assumption. The weighted median requires at least 50% of the weight in the analysis to be vital, and the weighted model needs the largest subset of instruments to identify the causal effect to be a vital instrument. Therefore, the MR egger intercept was applied to assess pleiotropy by comparing the results of the IVW model, MR Egger model, weighed median and weighted model. Furthermore, the heterogeneity was tested by Cochrane’s Q statistic.