ABSTRACT
Several studies have indicated that fluoroquinolone use may be associated with an increased risk of aortic aneurysm or dissection (AAD). Because patients with AAD or Marfan syndrome are at increased risk for aortic rupture, we performed a retrospective cohort study to determine the prevalence of systemic fluoroquinolone exposure and predictors of fluoroquinolone use in these patients. Data were obtained from the advisory board billing and administrative database, which contained information on 22 million adult hospitalizations in the United States for the study period (2009 to 2015). International Classification of Diseases (9/10) and Current Procedural Terminology codes were used to identify patients who had AAD or Marfan syndrome or underwent aortic repair. We identified 136,789 admissions for AAD, which involved 99,818 unique patients, 20% of whom received fluoroquinolone during a hospital admission. Of the 7,045 patients with dissection, 18% were exposed to fluoroquinolone. Of the 27,876 AAD patients who underwent aortic repair, 19% received fluoroquinolone during a hospitalization before the repair. In the AAD patients, having a diagnosis of pneumonia or urinary tract infection increased the likelihood of receiving fluoroquinolone during admission by 46% and 40%, respectively (P < 0.001). Additionally, we identified 2,871 admissions for Marfan syndrome, which involved 1,872 patients, 14% of whom received fluoroquinolone during an admission. In these patients, pneumonia and urinary tract infections also increased the risk of fluoroquinolone exposure. If the deleterious effects of fluoroquinolone on aortic integrity are substantiated, reducing fluoroquinolone use in hospitalized patients with aortic disorders will become an urgent safety issue for antibiotic stewardship programs.
INTRODUCTION
Aortic aneurysms are characterized by weakening and subsequent focal dilation of the aortic wall. Often, aortic aneurysms cause no symptoms until they rupture, a catastrophic event associated with an estimated mortality rate of over 80% (1). In addition, patients with an aortic aneurysm are at risk of developing aortic dissection, which is characterized by separation of the layers of the aorta. Aortic dissection also has a grave prognosis because it commonly results in rupture (2, 3). The literature on risk factors associated with aortic aneurysm and/or dissection (AAD) is robust. The main nonmodifiable risk factors associated with AAD development are older age, male sex (men have a 4-fold increased risk compared with women), and family history of aortic disease, whereas the strongest modifiable risk factor is smoking (4). Another well-described risk factor for AAD development is Marfan syndrome, which is a genetic disorder caused by an inherited mutation in the gene for fibrillin, resulting in abnormal connective tissue. An estimated one-third of patients with Marfan syndrome have an aortic event during their lifetime (5). Given the devastating effect of aortic disease, it is critical to avoid factors that promote aortic damage, disease progression, and eventual aortic dissection or rupture when managing patients at high risk for AAD (6, 7).
Fluoroquinolones (FQs) are among the most commonly prescribed antibiotics worldwide (8–10). They are used to treat a variety of common bacterial infections, including urinary tract infections (UTIs) (11–14) and pneumonia (15, 16). Fluoroquinolones are also used to treat infected aortic aneurysms (17–19). Postmarketing surveillance and clinical studies have indicated that FQs are associated with several adverse effects (20), including tendonitis and tendon rupture (21–23), QT prolongation (24), dysglycemia (25, 26), and neuropathy (27–30). Because of the adverse effects associated with FQ use, the United States Food and Drug Administration (FDA) issued a black box warning for FQs in 2016 and has provided updated warnings several times since then (31–33). Moreover, FQ use is one of the factors that has led to the recent emergence and global dissemination of the multidrug-resistant Escherichia coli strain sequence type 131 (34), and FQ exposure puts patients at high risk of subsequently developing a Clostridium difficile infection (35).
Recently, the results of 4 observational clinical studies have suggested that FQ use may also be associated with an increased risk of AAD (36–39). Although the underlying mechanisms have yet to be fully determined, this alarming association has raised concern that AAD patients may progress to aortic rupture while being treated with FQs. In a recent laboratory-based study, we found that exposure to the FQ ciprofloxacin increased the incidence of aortic dissection and rupture in a mouse model of AAD (40), further substantiating that caution should be used when considering the use of FQ in patients who have aortic dilatation or are otherwise at high risk for AAD. In response to this additional evidence, the European Medicines Agency (EMA) Pharmacovigilance Risk Assessment Committee released an official warning against the use of FQ in patients with diagnosed AAD or conditions that predispose an individual to aneurysm development (41).
In light of these escalating concerns, we decided to explore how often patients with AAD or Marfan syndrome are actually receiving FQ antibiotics, given that these patients are at increased risk of aortic rupture. In this retrospective cohort study, we examined the prevalence of FQ exposure in patients with AAD or Marfan syndrome in the acute care setting. Our secondary aim was to determine the clinical and demographic factors most predictive of FQ exposure in these patients, with the eventual goal of targeting antibiotic stewardship interventions to the patients most at risk.
This work was presented, in part, at the Society for Healthcare Epidemiology of America 2018 spring meeting, 20 April 2018 in Portland, Oregon.
RESULTS
Demographics.For this study, we analyzed hospital admissions data from the advisory board billing and administrative database for the years 2009 to 2015. Using this database, we identified 134,469 unique admissions for AAD, which involved 99,818 unique patients (Table 1). Six percent of these admissions (n = 8,035) were for dissection. We also identified 2,871 unique admissions for Marfan syndrome, which involved 1,872 unique patients. Of these, 338 had AAD and, thus, were also included in the AAD group. For the AAD admissions, the patients were predominantly male (67%). As expected, for the Marfan syndrome admissions, the patients were younger, were more likely to be female, had a lower comorbidity index, were less likely to smoke, and were less likely to be admitted for a UTI or chronic obstructive pulmonary disease than those admitted for AAD (P < 0.001 for all comparisons).
Characteristics of the cohorts studieda
Fluoroquinolone exposure.Of the 99,818 patients with AAD, 19,889 (20%) were administered an FQ antibiotic during a hospital admission (Table 2). For the subset of 7,045 patients with dissection, the proportion exposed to FQ was 18%. For the AAD patients who underwent aortic repair, the overall FQ exposure was 12% (3,301/27,806), which was significantly lower than the 20% exposure for the entire AAD group (P < 0.001). Furthermore, for the AAD patients who underwent aortic repair, the proportion exposed to FQ was significantly higher before the repair than during or after the repair (19% versus 11%, P < 0.001). Concerningly, 23% of the patients with diagnosed AAD who did not undergo repair were exposed to FQ during a hospitalization; this exposure was significantly higher than the 12% for the aortic repair patients (P < 0.001). For the Marfan syndrome patients, the proportion exposed to FQ was 14%; the exposure proportions did not differ significantly among the Marfan patients who did or did not have AAD (P = 0.75), did or did not have dissection (P = 0.78), or did or did not undergo repair (P = 0.13).
Fluoroquinolone exposure during hospital admission in patients at risk for aortic rupture
Predictors of fluoroquinolone exposure for patients with AAD.A multilevel logistic regression analysis revealed that several clinical factors were predictive of FQ use in the patients with AAD (Table 3). Factors found to be independently associated with a higher likelihood of FQ use included having a diagnosis of pneumonia (odds ratio [OR], 6.22; 95% confidence interval [CI], 2.32 to 16.69) and having a diagnosis of UTI (OR, 4.97; 95% CI, 2.09 to 11.82). In the patients with AAD, having a diagnosis of pneumonia or urinary tract infection increased the likelihood of receiving FQ during an admission by 46% and 40%, respectively (P < 0.001). Factors associated with being less likely to receive FQ during an admission included being male (OR, 0.79; 95% CI, 0.70 to 0.90) and undergoing aortic repair (OR, 0.62; 95% CI 0.48 to 0.80). Of the two-way interactions between variables that were assessed, only “pneumonia × UTI” achieved statistical significance. Although pneumonia and UTIs were both associated with an increased likelihood of FQ exposure, having both did not have an additive effect. Rather, the patients with both pneumonia and a UTI had only a slightly greater chance of FQ exposure than patients with only pneumonia.
Predictors of exposure to fluoroquinolones in patients with aortic aneurysm and/or dissectiona
Predictors of fluoroquinolone exposure for patients with Marfan syndrome.A multilevel logistic regression analysis revealed that a narrower set of clinical factors were predictive of FQ use in the patients with Marfan syndrome (Table 4). Factors independently associated with a higher likelihood of FQ use included having a diagnosis of pneumonia (OR, 10.53; 95% CI 6.57 to 16.87) and having a diagnosis of UTI (OR, 7.90; 95% CI, 4.99 to 12.52).
Predictors of exposure to fluoroquinolones in patients with Marfan syndromea
DISCUSSION
Given the grave prognoses associated with aortic rupture, it is important to reduce the modifiable risk factors associated with the development or progression of AAD. In the present study, we found that 20% of the patients with diagnosed AAD, 19% of the patients hospitalized before aortic repair, and 14% of the patients with diagnosed Marfan syndrome were exposed to FQ during a hospital admission. For both the AAD patients and the Marfan syndrome patients, a diagnosis of UTI or pneumonia was associated with an increased risk of FQ exposure during hospitalization. Therefore, a substantial proportion of patients with AAD were exposed to FQ while in the hospital, and having infections that are commonly treated with FQ increased the risk of exposure.
Because of the risks described as being associated with FQ use, there has been a decrease in FQ use in recent years; however, FQs remain the most commonly prescribed class of antibiotics in inpatient and outpatient settings (42, 43). Fluoroquinolones are frequently prescribed for the treatment of UTIs (43) and are a key component of the empirical treatment of community-acquired pneumonia (15, 16). Consistent with this, we found that one-fifth of the patients with AAD received FQ, particularly if the patient also had pneumonia or a UTI.
Our finding that having pneumonia or a UTI increased the likelihood of FQ exposure in both AAD and Marfan syndrome patients is not surprising given the frequency with which FQ drugs are used (appropriately or inappropriately) for these indications. We also found that within the group of patients with AAD, women were more likely to receive a FQ antibiotic than men. A possible explanation for this finding is that women are more likely to be diagnosed with and treated for a UTI during hospitalization (44). The lower risk of FQ exposure in hospitalizations that occurred during or after aortic repair than in those that occurred before repair may reflect a shorter time available to be hospitalized after repair in this population comprising mostly older adults. We observed the inverse relationship in the population of Marfan syndrome patients (i.e., more patients were exposed to FQs during hospitalizations that occurred after repair).
All database studies, including ours, have limitations that must be considered when interpreting the results. Although it would be of interest to examine FQ exposure in separate aneurysm and dissection cohorts, this was not possible because of limitations related to coding practices that precluded reliable identification of patients who had aneurysm without dissection or who had dissection without aneurysm. In our analyses, we were unable to include admissions or repair procedures that occurred at hospitals not reporting to the advisory board billing and administrative database. The database did not contain any information about the indications for antibiotic use or why a specific antibiotic was chosen, so we cannot specify the prescribers’ motives for the increased use of FQ in women and decreased use in Marfan syndrome patients. Furthermore, we were unable to determine the prevalence of FQ exposure during hospital admissions involving the general population, preventing us from including this comparison group. Our results only show the incidence of FQ exposure in the inpatient setting. Because FQs are frequently prescribed in the outpatient setting (10, 45), our results very likely underestimate the overall incidence of exposure to these drugs in patients with AAD and Marfan syndrome. The purpose of our study was to describe the frequency of FQ exposure in patients with AAD, but the results do not imply a causal relationship. There is still an immediate need for additional translational research to expand our understanding of the biologic mechanisms underlying the association between the use of FQ antibiotics and the occurrence of adverse aortic events.
Our study revealed that a large number of patients with AAD, including those with unrepaired lesions, are receiving FQ antibiotics. If the suspected deleterious effect of FQs on aortic integrity is substantiated, reducing FQ use in these patients should be a high priority for antibiotic stewardship and patient safety programs. Our results help identify the patients who are most at risk of FQ exposure during hospitalization and, thus, provide a starting place for stewardship interventions.
MATERIALS AND METHODS
Data sources and study population.The protocol was reviewed by the Baylor College of Medicine institutional review board. The board determined that this project did not constitute human subjects research. The data were gathered from the advisory board billing and administrative database, a comprehensive reporting system that collects data from acute-care hospitals across the United States. For the years 2009 through 2015, the database included patient-specific data on 22 million hospitalizations involving adults. The database provides a cross-payer perspective; it includes hospital billing and administrative data for inpatient and outpatient encounters covered by commercial insurance, government insurance, self-pay, and charity care. The data in the database were deidentified in accordance with the Health Insurance Portability and Accountability Act (HIPAA) privacy rule. The hospitals contributing to this database are generally representative of hospitals across the United States, except that there are fewer small facilities (i.e., <50 beds) and fewer facilities in the western United States; these differences were ≤5% from the national averages determined by the American Hospital Association. We have used this database for prior studies of antibiotic use (43, 46).
Measures.To create a cohort of patients at risk for aortic rupture, we used International Classification of Diseases (ICD-9/10) codes to identify records regarding patients with an aortic aneurysm (thoracic or abdominal), an aortic dissection, or Marfan syndrome or who underwent aortic repair (see Table S1 in the supplemental material). We also used Current Procedural Terminology (CPT) codes to determine if a procedure had been performed. We included patients with a diagnosis of AAD or Marfan syndrome anywhere among the admission diagnoses. We included patients with ruptured aneurysm because those who undergo surgical repair of the ruptured segment (especially those with a ruptured thoracic aorta) often have other unrepaired segments with disease that remain at risk for progression. We also identified the following diagnoses among the admission diagnoses by using ICD-9/10 codes: pneumonia, UTI, emphysema, and chronic obstructive pulmonary disease (see Table S1). These additional diagnoses were studied because they are the leading causes of antibiotic use in inpatient hospitalizations. We focused on admission diagnoses in order to capture data regarding the clinical scenario that will need to be addressed by antibiotic stewardship interventions if FQ is proven to increase risk in patients with AAD. Stewardship programs will need to identify all patients with a primary or secondary diagnosis of AAD at the time of admission and discourage prescription of FQ in such patients. Patient demographic data extracted from the database included sex, age (continuous), length of stay (continuous), tobacco use (yes or no), and Elixhauser score. Aortic repair procedures were identified by using both CPT and ICD-9/10 codes.
Fluoroquinolone exposure was defined as any instance of a systemic FQ being ordered during a hospitalization, including when such medications were given in the emergency department on the first day of admission. We queried the Charge Description Master codes for the names of all the FQ drugs approved in the United States (ciprofloxacin, gatifloxacin, levofloxacin, moxifloxacin, ofloxacin, and their corresponding brand names) to determine which patients were exposed to FQ during their hospital admission. Topical, otic, and ophthalmic formulations of FQ were excluded from the list. Complete details of the codes used to generate our cohort are included in the supplemental materials (Table S1). We identified 2 patient groups, namely the AAD group, which included all patients who had an aneurysm and/or dissection involving the thoracic or abdominal aorta and were 18 years old or older; and the Marfan syndrome group, which included all patients with Marfan syndrome, regardless of their age or whether they had an aneurysm or dissection. Marfan syndrome patients with AAD were included in the AAD group and the Marfan syndrome group for analyses. Because dissection patients are at particularly high risk for rapid aortic expansion and rupture, we also identified a subgroup of the AAD group that comprised adult patients with confirmed aortic dissection (rather than aneurysm without dissection). Because of the inherent limitations of the coding structure, we were not able to reliably identify patients with dissection who did not have aneurysm or vice versa. Patients who had a length of stay over 31 days were excluded to ensure that our study focused on acute, inpatient hospitalizations. Each patient could have had multiple admissions.
Statistical analysis.First, we identified all the AAD admissions and the Marfan syndrome admissions separately using the corresponding ICD-9/10 codes. Then, we separated our cohort of patients with AAD on the basis of more granular diagnoses (i.e., thoracic AAD, abdominal AAD, dissection, Marfan syndrome, and aortic repair procedures). Next, for the patients fitting each diagnostic profile, we determined what percentage of the patients was exposed to FQs. Summary statistics were compared by using the chi-square test, Kruskal-Wallis test, Wilcoxon-Mann-Whitney test, 2-sample proportion test, or independent sample t test, depending on the nature of the variables.
Because of the hierarchical nature of the data, we used multilevel logistic regression to analyze the predictive value of various admission characteristics on FQ use in both the AAD group and the Marfan syndrome group. We have used this approach in prior studies to examine the effects of urine culture at admission on antibiotic selection (46). The levels were facilities and attending physicians, which means we accounted for clustering of observations by facility and attending level. To mitigate the problem of multiple comparisons and ensure model parsimony, we chose the predictors of our statistical model not through one regression but through repeated bootstrapped samples. This procedure mitigates multiple comparison problems by insisting on predictor significance across many random samples. The model for all AAD patients was developed by using repeated random samples of 30,000 until the model’s estimates were stable. With each sample, we noted which predictors influenced the model (P < 0.25) and excluded predictors that had a P value of ≥0.25 in the previous step. The predictors evaluated were sex, age, length of stay, aortic repair, pneumonia, UTI, aneurysm, dissection, Elixhauser score, tobacco use, diagnosis of emphysema, diagnosis of chronic obstructive pulmonary disease, diagnosis of Marfan syndrome, and all two-way interactions (such as pneumonia and UTI). We repeated the procedure 50 times. Because each random sample was large (n = 30,000), there was no substantive statistical power lost compared with using the full data set. Because effects are not uniform over all values of a predictor in logistic regression, we included maximum effect estimates in addition to traditional odds ratios. The same analysis was performed with the Marfan syndrome group by using repeated random samples of 1,000. All analyses were performed in R 3.4. (R Core Team 2017) (47).
ACKNOWLEDGMENTS
B.W.T. was supported by the Center for Innovations in Quality, Effectiveness and Safety (CIN 13-413) at the Michael E. DeBakey VA Medical Center, Houston, TX. S.A.L. was supported, in part, by the Jimmy and Roberta Howell Professorship in Cardiovascular Surgery at Baylor College of Medicine.
We declare no conflicts of interest with the contents of this article.
FOOTNOTES
- Received 14 August 2018.
- Returned for modification 5 September 2018.
- Accepted 20 November 2018.
- Accepted manuscript posted online 26 November 2018.
Supplemental material for this article may be found at https://doi.org/10.1128/AAC.01712-18.
- Copyright © 2019 American Society for Microbiology.