Oral fluoroquinolone use and the risk of aortic and mitral valve regurgitation: a nationwide cohort study

Abstract

Objective

This study aims to investigate whether oral fluoroquinolone use is associated with an increased risk of aortic and mitral valve regurgitation.

Design

A nationwide cohort study based on linked healthcare register data encompassing patient characteristics, prescription data and cases of aortic and mitral valve regurgitation.

Setting

Sweden, 2006–2018.

Participants

Study participants were adults aged 18–75 years. 794 588 courses of fluoroquinolones (88% ciprofloxacin) propensity score-matched (1:1) to an equal number of courses of penicillin V.

Main outcome measures

Aortic or mitral valve regurgitation leading to ED visit, hospitalisation or death. Cox regression was used to estimate HRs for the first incident of valve regurgitation within 120 days of treatment initiation. Subgroup and supplementary analyses included sex, age, first episode, surgery or death only, 356-day follow-up and analysis using amoxicillin as comparator drug.

Results

There were 29 (incidence rate 0.5 per 1000 person-years) cases in the fluoroquinolone group compared with 43 (0.7 per 1000 person-years) cases in the penicillin V group during the main (1–30 days) follow-up period; corresponding to a HR 0.70 (95% CI 0.43 to 1.11). Analyses of subsequent time periods yielded results consistent with the main period: HR 1.06 (0.63–1.78) for days 31–60 and 0.90 (0.59–1.36) days 61–120. No significant differences were found in any of the subgroup- or supplementary analyses. The absolute risk difference was −17 (95% CI −33 to 6) cases of valve regurgitation per 1 000 000 courses of fluoroquinolones.

Conclusion

In a nationwide cohort study, there was no statistical support for an increase in the risk of mitral or aortic valve regurgitation associated with oral fluoroquinolone use.

STRENGTHS AND LIMITATIONS OF THIS STUDY.

• Based on the nationwide register data (Swedish healthcare registers) with near-complete coverage.

• Active-comparator design (fluoroquinolones vs penicillin V) to minimise the risk of confounding by indication.

• Propensity score matching on a wide range of demographic, clinical and healthcare utilisation variables to enhance comparability between exposure groups.

• Strict inclusion and exclusion criteria to isolate incident antibiotic use may limit generalisability.

• As in all observational studies, residual confounding cannot be fully excluded.

Introduction

Fluoroquinolones are a group of broad-spectrum antibiotics used to treat a variety of both gram-positive and gram-negative bacterial infections. Due to its versatility in treating infections across a wide range of clinical scenarios, it is one of the most frequently prescribed antibiotics with annual prescription rates of almost 15 million per year in the US alone.¹

While generally regarded as well tolerated, studies in recent years have reported an increased risk of aortic aneurysm, aortic dissection and heart valve regurgitation.2,4 The proposed mechanism behind these adverse reactions is the drug interacting with collagen and extracellular matrix degradation, resulting in an increased risk of connective tissue disorders. Therefore, both European and US drug administration agencies have issued warnings regarding the potential association.^{5 6} The recommendations pertaining to valve regurgitation rely primarily on a case-control study based on data from a US health claims database that reported an increased risk of heart valve insults compared with amoxicillin and azithromycin.⁴ In contrast, a recent case-control study based on the registry data from Denmark found no significant increase in the risk of valve regurgitation among individuals who used fluoroquinolones compared with users of penicillin V.⁷

Given these controversies, we conducted a Swedish nationwide cohort study investigating whether oral fluoroquinolone use was associated with an increased risk of heart valve regurgitation.

Methods

Study design

This nationwide cohort study was based on individual data from July 2006 to December 2018 collected from the national health and administrative registries, encompassing all Swedish residents between 18 and 75 years of age. We used individual information on underlying health status and healthcare use and data on filled prescriptions to match users of oral fluoroquinolones to users of penicillin V to estimate the risk of seeking care for or dying of mitral and aortic valve regurgitation within 120 days from start of drug use. We employed an active comparator design with penicillin V as the reference. Penicillin V is an antibiotic without any known side effects on connective tissue or collagen synthesis. We used propensity score matching on a large range of covariates (see online supplemental sTable 2) to balance differences in baseline health status. Each Swedish citizen is assigned a unique personal identification number, which enables linking between the registries.

Data sources

We collected individual data (between 2006 and 2019) from the National Prescription Registry (study drugs and concomitant medications), the National Patient Registry (outcomes and health status), the National Cause of Death Registry (outcomes), and the Total Population Registry (vital status, demographics such as: age, sex, county and municipality). No values were missing. We used a 5-year lookback from filled prescription to collect data on drug use and underlying diseases. Diagnoses and drug codes in the registries are classified using the International Classification of Diseases (ICD-10) and the Anatomical Therapeutic Chemical Classification system (ATC) (see online supplemental sTable 3 for details on the codes used).

Study cohort

A treatment episode was characterised as a filled prescription by an individual aged 18–75 years, of any of the study drugs (fluoroquinolones or penicillin V) between 1 January 2006 and 31 December 2018. Episodes were excluded if they were preceded by: any filled prescription of a study drug past 120 days, hospitalisation in the past 120 days, multiple filled prescriptions of study drugs on index date or lack of activity in any of the nationwide registers past 365 days (to ensure capture of ascertained covariates). Several diagnoses served as exclusion criteria (see online supplemental sTable 1) in attempting to isolate the incident cases and to reduce confounding from participants with a high pre-treatment risk of valve regurgitation. The study allowed individuals to contribute multiple treatment episodes, provided that each episode was precluded by at least 120 days without study-drug exposure or hospitalisation, ensuring non-overlapping courses.

Patient and public involvement

None.

Propensity score model

To mitigate possible confounding caused by disparities in baseline health status (demographic characteristics, concomitant drug use, healthcare use and medical history), the propensity for undergoing fluoroquinolone treatment was estimated using logistic regression on 78 covariates as predictors (see online supplemental sTable 2 for a full list). We used the greedy 5->1-digit propensity score algorithm to match users of fluoroquinolones with users of penicillin V on a 1:1 ratio.⁸ Courses were matched without replacement, using a calliper width of 0.2 SD of the logit of the propensity score. Treated episodes (fluoroquinolones) were prioritised during sequential matching according to the 5->1-digit principle. Covariate balancing was estimated using standardised differences with values less than or equal to 0.10 considered well balanced.⁹ Missing values were detected in the ‘region of residence’ category (0.2%) and addressed by adding a missing value category.¹⁰

Follow-up and outcome

The time from start of fluoroquinolone treatment to onset of connective tissue side-effects has been reported to range from hours to weeks in previous studies.^{11 12} Based on this, we defined the main risk period as days 1–30 after the date when the prescription was filled (index date) with separate sub-analyses for subsequent time periods (31–60 and 61–120 days, respectively). A follow-up started on the index date and continued until either of the following occurred: end of study (31 December 2018), participant reaching 76 years of age, hospitalisation or death due to any of the primary outcome diagnoses, reaching 120 days from filled prescription, death or emigration.

The primary outcome mitral and aortic valve regurgitation was defined as: mitral valve insufficiency (ICD-10: I34.0), aortic valve insufficiency (ICD-10: I35.1), rupture of chordae tendinae (ICD-10: I51.1A), recorded in either the National Cause of Death Register or the National Patient Register. The collection of ICD codes used for this study is listed in online supplemental sTable 3.

Statistical analyses

The association between fluoroquinolone use and risk of aortic and mitral valve regurgitation was estimated using Cox regression with treatment status as the sole variable. The exclusion criteria ensured that there were no overlapping courses, and patients who experienced any of the outcomes were similarly not included in future analyses. Non-independence due to multiple observations was accounted for by applying sandwich variance estimation clustered by individual. The proportional hazards assumption was estimated using a Wald test which tested the interaction between treatment status and time scale.¹³

Separate subgroup analyses were performed by estimating HRs for participants classified according to sex and age (<65 years of age). We used likelihood ratio tests to estimate homogeneity between the subgroup estimates.

We performed sensitivity analyses using a more restrictive outcome definition, including only cases who either had cardiac surgery due to the aortic or mitral valve regurgitation or who died and in whom aortic or mitral valve regurgitation was the underlying cause of death. Furthermore, we performed a sensitivity analysis restricted to the first episode included in the study for each individual participant.

We also estimated all-cause mortality for the main time period in order to evaluate potential unmeasured confounding, since an observed difference in all-cause mortality could reflect an imbalance in baseline health status or severity of infection, between the two treatment groups. A Fine-Grey subdistribution hazards model was used as a sensitivity analysis to account for all-cause mortality as a competing risk. Finally, two supplementary analyses were conducted: one where we varied the comparator antibiotic, using amoxicillin as comparator, and another in which we extended the follow-up period to 365 days.

All statistical tests were two-sided where a 95% CI not overlapping 1.0, or equivalently had a p-value <0.05, was considered statistically significant. Statistical analyses were performed using SAS V.9.4 (SAS Institute).

Results

Cohort

We collected information on all filled prescriptions of oral fluoroquinolones (2 228 668) and penicillin V (7 601 340) from the source population encompassing all Swedish adults aged 18–75 years of age during the 2006–2018 time period. After exclusions were applied, there remained 854 498 courses of fluoroquinolone and 3 826 096 courses of penicillin V which were subsequently propensity score matched at a 1:1 ratio (figure 1). The matched cohorts consisted of 794 588 courses of fluoroquinolone use and penicillin V use, respectively. The matched cohorts were determined as well balanced on all covariates (online supplemental sTable 9). figure 2 shows the propensity score distribution in the matched and unmatched cohorts. The most common fluoroquinolone was ciprofloxacin (88.1%), followed by norfloxacin (8.54%), moxifloxacin (1.79%), levofloxacin (1.31%) and ofloxacin (0.27%).

Figure 1. Cohort flowchart. *Including HIV/AIDS, other end-stage illness, dementia, drug misuse, psychosis and replaced valve. ^†Valve disease: ICD-10 M05–M09, M45, 100–102, 105, 133, 138, 1398, 1340, 1351, Q874. ICD: International Classification of Diseases.

Figure 2. Distribution of propensity scores prior to and post-matching.

Main results

During the main time period (1–30 days), there were 29 events of aortic or mitral valve regurgitation among courses of fluoroquinolones (60 942 person-years; incidence rate 0.5 per 1000 person years) and 43 events among the courses of penicillin V (62 757 person-years; incidence rate 0.7 per 1000 person years), corresponding to a HR of 0.70 (95% CI 0.43 to 1.11) for oral fluoroquinolones compared with penicillin V. The absolute risk difference for oral fluoroquinolone use compared with penicillin V in the main time period, expressed as number of cases of valve regurgitation per 1 000 000 courses of oral fluoroquinolones, was −17 (95% CI −33 to 6). There was no increased risk associated with oral fluoroquinolone compared with penicillin V in the later time periods, HR for days 31–60: 1.06 (0.63–1.78) and for days 61–120: 0.90 (0.59–1.36). See table 1 for results from all time periods.

Table 1. Risk of valve event in users of fluoroquinolones and penicillin V.

Analysis period	Fluoroquinolones	Penicillin V	Hazard ratio (95% CI)
	Events (n)
Days 1–30 (main period)	29	43	0.70 (0.43 to 1.11)
Days 31–60	28	28	1.06 (0.63 to 1.78)
Days 61–120	40	48	0.90 (0.59 to 1.36)

Subgroup analyses

Table 2 contains a summary of the HRs for aortic and mitral valve regurgitation for oral fluoroquinolone users compared with penicillin V according to subgroups. We found no noticeable difference in HRs according to sex (test of homogeneity, p value=0.83) or age group (p value=0.45).

Table 2. Subgroup analyses of the risk of valve event in users of fluoroquinolones and penicillin V (days 1–30).

Analysis	Fluoroquinolones	Penicillin V	Hazard ratio (95% CI)	P value	Interaction hazard ratio (95% CI)
	Events (n)
Sex
Male	20	29	0.72 (0.41 to 1.28)	0.83	1.12 (0.41 to 3.08)
Female	9	14	0.65 (0.28 to 1.50)
Age (year)
<70	15	26	0.62 (0.36 to 1.09)	0.45	1.50 (0.52 to 4.33)
≥70	14	17	0.94 (0.38 to 2.31)

Supplementary and sensitivity analyses

We observed a small, statistically uncertain, elevation in all-cause mortality among users of fluoroquinolones in the matched cohorts during the main time period (HR=1.14, 0.93–1.40). See online supplemental sTable 4. However, the causes of death in the two cohorts remained congruent. figure 3 contains a visual summary of the supplementary analyses also listed in detail in online supplemental sTables 4–8. The Fine-Grey competing risk model yielded an estimate closely aligned with the cause-specific Cox model.

Figure 3. Forest plot of supplementary analyses.

In the analysis using a restricted outcome definition, HR for death from, or who had cardiac surgery due to aortic or mitral valve regurgitation (days 1–30) was 0.90 (0.33–2.48). The sensitivity analysis, where only the first antibiotic prescription was included, yielded similar results as the main analysis (HR=0.72, 0.37–1.41). We performed a supplementary analysis with amoxicillin as the reference drug. Amoxicillin is an antibiotic with a broader spectrum of activity than penicillin V but with similar indications (primarily upper and lower respiratory tract infections). In accordance with the main analysis, we did not find a difference in the risk of aortic and mitral valve regurgitation between courses of fluoroquinolones and courses of amoxicillin (HR=1.10, 0.58–2.09). Finally, we estimated the risk of valve regurgitation using a longer time period of (HR for days 1–365: 0.94, 0.80–1.10).

Discussion

In this nationwide, propensity score-matched study, we did not find an increased risk of aortic and mitral valve regurgitation among users of fluoroquinolones during a follow-up period of 30 days. The results were consistent in the subsequent time periods 31–60 and 61–120 days following filled prescription.

In recent years, there have been several reports of serious side effects of fluoroquinolones on connective tissue. The earliest reports, shortly after the fluoroquinolones were released to the market, were mainly regarding tendinitis and achilles tendon rupture.^{14 15} Recently, however, more alarming reports have described serious adverse events such as aortic rupture and aortic dissection.^{2 3} The suggested rationale behind these effects is the fluoroquinolones’ ability to up-regulate metalloproteinases which in turn leads to the degradation of collagens, in particular type I collagen which is an integral part of the blood vessels of the vascular tree as well as the inner lining of the heart including the valves.^{16 17} In 2020, the EMA published a Direct Healthcare Professional Communication with regard to the risk of heart valve regurgitation and incompetence associated with use of systemic fluoroquinolones.⁵ The recommendation primarily stemmed from a North American study from 2019 in which the authors reported an increased risk of aortic and mitral valve regurgitation associated with current use of fluoroquinolones compared with amoxicillin and azithromycin (RR 2.40 and 1.75, respectively).⁴ This study had a case-control design, based on a health claims database combined with the FDA’s FAERS (FDA Adverse Event Reporting System) database. Although methods were applied to balance cases and controls, there was a considerable imbalance among several of the included background factors. Although health claim databases have the power of numbers, there are weaknesses such as: incomplete coverage (not all healthcare encounters are included), no data on non-reimbursable drugs, and selection bias. Therefore, it cannot be excluded that the observed increased risk in the study was attributable to factors associated with the acute infection itself or to confounding because of poorer baseline health of the fluoroquinolone users.

As opposed to this, a study based on national registry data from Denmark found no association between the use of fluoroquinolones and mitral and aortic valve regurgitation when compared with penicillin V (HR 1.0, 95% CI 0.95 to 1.23).⁷ Scandinavian registries have near complete national coverage due to their design combined with universal healthcare. However, the Danish study had a case-control design wherein the distribution of background factors varied considerably between cases and controls.

Our study has both strengths and weaknesses. The nationwide and complete coverage of the healthcare registries minimises the risk of selection bias and assures generalisability to the general adult population. In addition, we used an active comparator design and propensity score matching on a wide number of covariates to reduce differences in baseline health status between the two cohorts. We excluded individuals with severe underlying diseases as well as individuals with any history of valve disorders to avoid including those with a presumed increased pre-treatment risk of valve regurgitation irrespective of fluoroquinolone use. An intrinsic weakness of registry-based studies is the risk of misclassification bias; both of exposure and outcome. To address the lack of data on in-hospital drug use, we excluded courses that were preceded by hospitalisation the past 120 days. Although there has been no formal validation of the diagnoses used to identify the outcomes in our study, the overall validity of the Swedish registry data has been studied, reporting an estimated PPV of >85% for all diagnoses with an estimated completeness of >99%.¹⁸ To refine specificity, we performed a supplementary analysis restricted to surgery and death due to aortic and mitral valve regurgitation, which did not change the main conclusion. The impact of death as a competing risk was evaluated using a Fine-Grey subdistribution hazards model, which yielded an estimate closely aligned with the cause-specific Cox model. This suggests that accounting for competing death did not meaningfully change the effect estimate. A large proportion of the original cohort was excluded due to restrictions made to enhance the internal validity of the study, ensuring incident, non-overlapping courses and reducing bias from prior study drug exposure or hospitalisation. This approach may limit the generalisability of the results to more stable patients treated in the outpatient setting. The risk of protopathic bias, where a patient’s symptoms of the studied outcome (ie, valve regurgitation) would be misinterpreted as an infection that would prompt the prescribing of a fluoroquinolone is possible but unlikely, given that the indication for the most used fluoroquinolone (ciprofloxacin) is infections of the urinary tract, abdomen and infected prosthetic joints.

In summary, in this large nationwide study, we did not find any statistical support for an increased risk of mitral and aortic valve regurgitation associated with oral fluoroquinolones compared with users of penicillin V. However, given the uncertainty of the estimates, both modestly increased and decreased risk remain plausible. Since most prescribed fluoroquinolones in Sweden are ciprofloxacin, the results are primarily applicable to this drug, and we cannot exclude that intraclass differences may influence the risk of valve regurgitation associated with other fluoroquinolones.

What is already known on this topic

Fluoroquinolones have pharmacodynamic properties that allow them to interact with human connective tissues, such as the extracellular matrix and collagen. The inner lining of the heart and heart valves may be vulnerable to these unwanted effects due to the abundance of connective tissues.

Recent studies have reported an increased risk of heart valve regurgitation associated with fluoroquinolone use, prompting formal warnings from government agencies such as the Food and Drug Administration (FDA) and European Medicines Agency (EMA).

What this study adds

Oral fluoroquinolone use was not associated with an increased risk of aortic or mitral valve regurgitation in this large nationwide (Sweden) cohort study. The complete coverage and scope of Swedish healthcare registers allows for significant statistical power. Furthermore, the study employed propensity score matching on a large number of covariates as well as used penicillin V as a comparator drug, reducing the potential for confounding.

Data availability statement

Data are available upon reasonable request.