The study was registered in the International Prospective Register of Systematic Reviews (PROSPERO: CRD42021267568). The PRISMA checklist for reporting the meta-analysis is shown in Supplementary Table 1.
Search strategy
PubMed and Web of Science were searched for observational studies investigating the association of red meat intake with risk of RA until June 30, 2021, without language restrictions. The subject terms were established as follow: ("meat" or "beef" or "veal" or "pork" or "lamb" or "mutton" or "horse" or "goat" or "bacon" or "ham" or "sausage" or "salami" or "hot dogs") and ("rheumatoid arthritis" or "RA"). In addition, references of related studies were also checked to identify additional publications of interest.
Inclusion and exclusion criteria
Studies meeting the following criteria were included: (1) study type was an observational study; (2) directly reported odds ratio (OR) or hazards ratio (HR) or relative risk (RR) with 95% confidence interval (CI) or indirectly provided relevant data for calculation; (3) if study populations overlapped, the one with larger sample size was included.
The exclusion criteria were as follows: (1) animal study, review, meta-analysis, letter or comment; (2) no access to full text; (3) with insufficient data to obtain risk estimates with 95% CI.
Two authors (Chen W and Liu K) independently evaluated the retrieved literature in full text, and discrepancies were resolved by a third author (Ye D).
Data extraction
Data was extracted cross-checked by two researchers (Chen W and Liu K) independently from eligible studies. The extracted information included name of first author, publication year, area, age, gender, type of study design, follow-up period, sample size, dose of red meat intake and the maximally adjusted OR, RR or HR with corresponding 95% CI.
Quality assessment
The Newcastle-Ottawa Scale (NOS) [16] was used to evaluate the quality of included studies with scores ranging from 0 to 9 points. Studies with a quality score of no less than 7 points were considered as high quality. Two reviewers (Sang X and Zhuang Y) assessed the quality, and discrepancies were resolved by consensus and discussion.
Statistical analysis
All analyses were performed within Stata version 13 (Stata Corp LP, College Station, TX). Cochran’s Q test and I² statistics were utilized to assess the heterogeneity across the included studies [17]. Random-effects model was preferred when p < 0.10 and I2 > 50%; otherwise, fixed-effect model was applied.
In the categorical meta-analysis, ever red meat intake was compared with non/occasional red meat intake, which was defined by study-specific reference ranges. If the group of ever red meat intake was set up into multiple categories, we combined the effect estimates of different categories into a single value in each study. When the study reported mean and standard deviation (SD) of red meat intake between RA patients and controls rather than OR value, we transformed standardized mean difference (SMD) to OR according to the formula of ln OR= SMD [18]. If the study provided median and range of red meat intake, we computed mean and SD by the method of Hozo et al [19]. We also pooled the risk estimates comparing the highest with the lowest red meat intake among the studies with equal or more than three different categories of red meat intake. Moreover, subgroup analysis and meta-regression analysis were further performed based on food culture (western, oriental and Arab diet), publication year (<2018 and 2018), quality score (≥7 and <7), sample size (≥1000 and <1000). Furthermore, sensitivity analysis was used to check stability of the results by omitting one study at a time and combining the effect values of the remaining studies. Begg's test and Egger's test [20, 21] were used to evaluate publication bias.
For the dose-response meta-analysis, we used the method described by Greenland and Longnecker [22] to estimate the dose-response relationship between red meat intake and RA. If one study reported red meat intake by range, we assigned the midpoint of the lower and upper bound as mean or median was not provided. If the highest dose group was open-ended, the lower limit plus the width of the previous group was supposed as the corresponding consumption of red meat. When lowest dose group was open-ended, it was deemed zero as the lower bound of the reference group. If the consumption was shown by servings per day, we transformed it into grams per day using standard units of 120 g [23].