This meta-analysis adhered to the Meta-analysis of Observational Studies in Epidemiology (MOOSE) [9] and Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) [10] guidelines. The protocol was registered at PROSPERO and its number was CRD42019129056.
2.1 Literature Search
A systematic electronic in PubMed, Embase and Web of Science were performed from inception to March 2019. Electronic searches were used by exploded Medical Subject Headings (MeSH) terms and related key words. The search terms used were (MeSH exp “aorta” and key words “aorta”) and (MeSH exp “dissection” and key words “dissection”) and (MeSH exp “aortic aneurysm” and key words “aortic aneurysm” and “aneurysm”) and (key word ti “fluoroquinolone*”). The detailed search strategy for each of the database was provided in PROSPERO protocol. We did not make language restriction. Because of ensuring literature saturation, we reran the search on July 6, 2019.
2.2 Selection Criteria
Two authors (X.-C. D. and X.-X. Y.) independently conducted the original search, struck out duplicate articles, read the relevant titles and abstracts and classified records as included, excluded or uncertain. Due to ambiguity, the full-text article was acquired to identify eligibility. Discussion and consensus were used to resolve discrepancy.
Published cohort or case-control studies that met the following criteria were included:
- Participants: Adults no younger than 18 years treated for any condition with any fluoroquinolone.
- Exposures: patients received any of fluoroquinolones: ciprofloxacin, levofloxacin, enoxacin, sparfloxacin, norfloxacin, lomefloxacin, moxifloxacin, pefloxacin, ofloxacin, besifloxacin, gatifloxacin and gemifloxacin.
- Control: The placebos included either patients with none antibiotic or exposed to other antibiotics.
- Outcomes: the major outcome of interest was the first occurrence of aortic diseases.
- Data: studies reported on the unadjusted or the adjusted Relative Risk (RR) or Hazard Ratio (HR) or Odds Ratio (OR) or Risk Difference (RD) and their confidence intervals (CIs) or provided sufficient data to estimate these data.
2.3 Data extraction
Data extraction was performed by X.-C. D. and checked separately by other authors (X.-X. Y. and L. M.). Gathered data including the following: first author, year of publication, participants, interventions, primary outcomes, study design, controls, covariates adjustment, exposure category, crude and adjusted relative risk (RR), hazard ratio (HR) or odds ratio (OR) and 95% confidence intervals. Extracted data were collected into a Word (Microsoft Corporation) file. coauthors resolved discrepancies by discussion. Different follow-up and durations of fluoroquinolone use were provided in these studies; Although recording all follow-up data, we chose 60-day period from treatment start or follow-up end point because 60-day duration after current fluoroquinolone use had a higher RR, HR or OR.
2.4 Risk of Bias Assessment
Two author (X.-C. D. and G.-M. T.) independently assessed risk of bias using the Newcastle Ottawa Scale (NOS) for Quality Assessment for cohort and case-control studies.[11] NOS rates case-control studies on case definition, representativeness of cases, selection and definition of controls, comparability of controls, Ascertainment of exposure, same method for ascertainment of cases and controls and non-response rate. NOS rates cohort studies on representativeness of the exposed cohort, representativeness of the non-exposed cohort, ascertainment of exposure, outcome of interest was not present at the beginning of study, comparability of cohorts because of the design or analysis, assessment of outcome, follow-up long enough for outcomes to occur and adequacy of follow up of cohorts. Studies were considered as low quality (below 5 stars), moderate quality (5-7 stars’) and high quality (above 7 stars’).
2.5 Accessing Quality of Evidence
The quality of evidence classified as very low, low, moderate, or high for primary outcome according to Grading of Recommendations Assessment, Development, and Evaluation (GRADE) [12] methodology for risk of bias, inconsistency, indirectness, imprecision, and publication bias was independently evaluated by two authors (X.-C. D. and Y.-Y. P.). The GRADE Profiler (Windows-only tool GRADEpro) was used to construct summary tables.
2.6 Statistical Analysis
Random effects model with inverse variance method was used to pool studies’ adjusted RR, HR and OR and we supposed resemblance between OR and other relative measures such as RR or HR because aortic diseases were rare events.[13] So, using OR, RR or HR based on propensity score matching or adjusting whenever available or from adjusted multivariate analysis when propensity score matching was not obtainable. We calculated ORs with 95% confidence intervals (CIs) for overall effect estimate. Heterogeneity across studies was quantified using the I2 statistic; I2 >50% was considered as considerable heterogeneity.[14] P < .05 was considered statistically meaning for total included analyses, except where otherwise specified. All statistical analyses were conduct in STATA (Stata Version 14.0; Stata Corporation, College Station, TX, USA).
We conducted the number needed to treat to harm (NNTH) [15] and its 95% CI to estimate an absolute measure of effect because the OR represented a relative measure of effect. The NNTH means the number of patients needed to be treated with fluoroquinolones for one additional patient to have an adverse event. We used the pooled ORs to calculate the NNTH using the Rx software to convert the ORs to the NNTH.[16] The baseline risk was acquired from the no antibiotic group or exposure to other antibiotics in the population-based studies.