3.1 Trial selection
The PRISMA study flowchart is shown in Figure 1. Our initial search yielded 101 articles. A total of 16 records were thought to be preliminarily eligible for inclusion after screening the titles and abstracts and removing duplicates. After reviewing the full-text articles, our meta-analysis finally included 5 observational studies. [3,4,18-20] No randomized controlled trials were included because the incidence of aortic diseases is low and cannot be captured in randomized controlled trials with adequate power. [21, 22]
3.2 Trials Characteristics
The study design and main characteristics of the included studies are summarized in Supplementary Table 1 (Additional File 1). The publication years of the studies were between 2015 and 2019. The group sizes ranged from 2,426 to 1,744,360, with 2,829,385 patients in total. All included studies were observational in nature, including two cohort studies, [19, 20] one nested case-control study,[18] one case-crossover design study [4] and one case/no case study [3]. The follow-up ranged from 60 days to 17 years; one article [3] did not report the follow-up duration. All studies adjusted the RRs, HRs or ORs, including 3 trials that used propensity score matching, [18-20] 1 trial that used a conditional logistic regression model [4] and 1 trial that used multivariable logistic regression [3]. All studies evaluated different fluoroquinolones, and the trials used a combination of administrative diagnostic codes to identify new-onset aortic diseases.
3.3 Risk of Bias Assessment
We summarized the detailed information about selection, comparability and exposure in the five trials, and the risk of bias according to the NOS assessment is shown in Supplementary Table 2 (Additional File 2). Overall, the five studies were considered to be at low risk for bias, which meant that they were of high quality (8 stars for each study). Publication bias was assessed by funnel plots and the Egger and Begg test for asymmetry when it was available.
3.4 Users of Fluoroquinolones vs. Nonusers or Users of Other Antibiotics
3.4.1 Primary Outcome: The First Occurrence of Aortic Dissection or Aortic Aneurysm: Four trials provided data on the first occurrence of aortic diseases. [3, 4, 18, 20] Compared with nonusers or users of other antibiotics, users of fluoroquinolones had a substantially higher risk of aortic diseases (adjusted OR, 2.23; 95% CI, 1.80-2.77; P=.000) (Fig. 2), with low heterogeneity (I2=17.9%).
3.4.2 The First Occurrence of Aortic Dissection and Aortic Aneurysm: Compared with nonusers or users of other antibiotics, users of fluoroquinolones also had a higher risk of aortic dissection (adjusted OR, 2.07; 95% CI, 1.33-3.21; P=.001), with very a high degree of statistical heterogeneity (I2=74.6%), and a higher risk of aortic aneurysm (adjusted OR, 1.74; 95% CI, 1.13-2.66; P=.011) (Fig. 2), also with a very high degree of statistical heterogeneity (I2=86.2%).
3.4.3 The NNH for Aortic Aneurysm or Dissection
The NNH for aortic aneurysm or dissection with current fluoroquinolone use was 1301 (95% CI 852-2201) treatment courses of fluoroquinolones. One study with no follow-up was excluded, the effect estimate from our meta-analysis (adjusted OR=2.10) was used, and a baseline risk (0.7/1,000 person-years) was obtained from a Swedish nationwide cohort [20].
3.4.4 Sensitivity Analysis
The heterogeneity among studies reporting on aortic aneurysms changed to 0 after eliminating the study by Lee et al. [18] (adjusted OR, 2.22; 95% CI, 2.01-2.46; P=.000). The heterogeneity index ranged from 82.8% to 67.9% after excluding the study by Pasternak et al. [20] (adjusted OR, 2.44; 95% CI, 1.73-3.43; P=.000) (Fig. 3).
3.4.5 GRADE Evidence and Publication Bias
The GRADE level of evidence was moderate for the first occurrence of aortic diseases (fluoroquinolone users vs. nonusers or users of other antibiotics) because of the study design (observational study) and other considerations (large weight of the effect, no dose response gradient and no plausible confounding).
In this meta-analysis of the first occurrence of aortic dissection or aneurysm, we did not use tests such as Egger’s test of Begg’s test to assess publication bias, because the number of trials was low.