Fluoroquinolone Antibiotics and Tendon Injury in Adolescents

We implemented an active-comparator, new-user^29,30  cohort study using IBM Watson Health MarketScan Commercial Claims and Encounters data (IBM Watson Health; Armonk, NY).³¹  MarketScan includes beneficiaries enrolled in employer-sponsored private insurance across the United States and with adjudicated claims data on outpatient and inpatient health services and outpatient pharmacy dispensing. The University of North Carolina Institutional Review Board reviewed and exempted this study (19-2483).

We included adolescents (aged 12–18 years) with an outpatient pharmacy claim for a dispensed prescription (henceforth, “index fill”) for oral formulations of fluoroquinolone or comparator antibiotics between 2000 and 2018. The cohort was restricted to this age group to improve comparability across treatment groups because fluoroquinolone dispensing is rare for children aged <12 years (1 fill per 1000 person-years in this database) as compared with comparator antibiotic dispensing (319 fills per 1000 person-years).³² 

In Fig 1 and Supplemental Fig 5,³³  we illustrate cohort inclusion and exclusion criteria. To restrict to new users of these antibiotics, we required ≥100 days of continuous prescription drug coverage before their index fill. We excluded adolescents with a systemic fluoroquinolone or comparator fill in that 100-day washout period on the basis of our hypothesis that prophylactic antibiotic prescriptions would not exceed a 90-day supply. To assess baseline covariates, we required continuous fee-for-service insurance coverage for ≥180 days before the index fill. We also excluded adolescents who experienced an outcome before 2 days post index fill (explained below).

Treatment was defined by using outpatient pharmacy dispensing data for oral antibiotics (fluoroquinolones and comparators). Fluoroquinolones included ciprofloxacin, levofloxacin, moxifloxacin, and gatifloxacin; comparator antibiotics included amoxicillin-clavulanate, azithromycin, cefalexin, cefixime, cefdinir, nitrofurantoin, and sulfamethoxazole-trimethoprim. Comparators were chosen because they are broad-spectrum antibiotics used for similar indications, such as urinary tract infections (UTIs), respiratory tract infections, and gastrointestinal infections. To classify medication type, we identified National Drug Codes using generic names in the National Drug Data File Plus (First Databank, South San Francisco, CA; www.firstdatabank.com) and Redbook (IBM Watson Health).

Our primary outcome was tendon rupture of the Achilles, quadricep, patellar, or a tibial tendon. Tendinitis was a secondary outcome. Additionally, we leveraged clavicle fracture as a negative control outcome to assess potential uncontrolled confounding; in the absence of confounding, the analysis of this outcome should produce a null result.³⁴ 

To define outcomes, we used International Classification of Diseases, Ninth Revision, Clinical Modification and International Classification of Diseases, 10th Revision, Clinical Modification diagnosis codes and, for surgical repair of rupture or clavicle fracture, Current Procedural Terminology (CPT) codes (Supplemental Table 2). Outcome codes were assessed starting 2 days after the index fill (1) because we hypothesized that antibiotic treatment could not plausibly affect risk of the outcome in less than a day and (2) to reduce the potential for reverse causality, in which the antibiotic is received prophylactically for a procedure to treat the injury.

We used a causal diagram to inform our selection of confounders.³⁵  Demographic factors included age, sex, geographic region of residence, and calendar time of index fill. Indication for the index fill was defined by using diagnosis codes from claims in the 3 days before and on the index fill date. Codes were summarized by using categories developed by Fleming-Dutra et al.³⁶ 

Other baseline covariates included comorbid conditions (complex chronic conditions^37,38 ), preindex inpatient and outpatient visit frequency (ie, health care use), preindex fills for other systemic antibiotics (Anatomical Therapeutic Chemical group J01³⁹ ) or systemic corticosteroids (budesonide, dexamethasone, hydrocortisone, methylprednisolone, prednisone, prednisolone, and triamcinolone hexacetonide), and systemic corticosteroid fills that occurred on the index fill date (Supplemental Figs 5 and 6, Supplemental Table 3)

We estimated the average treatment effect in the treated, which answers the question, “How much would risk of the outcome change among adolescents who received a fluoroquinolone if these adolescents had instead received a comparator antibiotic?” To estimate the average treatment effect in the treated and control for measured confounding,⁴⁰  we used standardized mortality ratio (SMR) weighting.⁴¹  In this approach, adolescents treated with a fluoroquinolone have a weight of 1 because they are the target population; adolescents treated with a comparator have a weight equal to their propensity score divided by its complement. The propensity score model parameterization is shown in Supplemental Table 3. We examined balance of measured covariates between treatment groups using standardized mean differences before and after weighting.^42,43 

We used Kaplan-Meier–based methods to estimate the cumulative incidence (ie, risk) of the outcome in each treatment group during follow-up.⁴⁴  We implemented 2 primary analyses: (1) an intention-to-treat (ITT) analysis and (2) a per-protocol analysis.⁴⁵  In the ITT analysis, patients were managed until they experienced the outcome, until disenrollment from insurance coverage, until 180 days after the index fill, or until the end of the study period (December 31, 2018), whichever came first. In the per-protocol analysis, the same criteria for follow-up were used, but follow-up was also stopped when the other treatment was dispensed, if that occurred.

We estimated the risk difference for every day of follow-up with 95% confidence intervals (CIs) using nonparametric bootstrap methods. We also calculated the inverse of the risk difference to estimate the number needed to treat to harm (NNTH),⁴⁶  which is the number of adolescents who would need to be treated with fluoroquinolones instead of a comparator antibiotic for 1 additional adolescent to experience the outcome, in expectation. Analyses were performed by using SAS version 9.3 (SAS Institute, Inc, Cary, NC).

To assess the robustness of our results to assumptions, we conducted a number of sensitivity analyses: (1) To reduce confounding by indication, we restricted to adolescents with codes for UTI. (2) To minimize the influence of adolescents with frequent antibiotic use, we restricted to their first eligible follow-up. (3) To mitigate channeling bias⁴⁷  and outcome measurement error,¹⁵  we restricted to the time period before the 2008 FDA boxed warning. (4) To reduce the potential obscuring of an association by short-course antibiotic treatment, we restricted to adolescents with index fills with at least a 3-day supply (98.8% of fluoroquinolone group and 98.7% of comparator group). (5) To increase comparability between fluoroquinolone and comparator groups,^48,49  before constructing SMR weights, we implemented asymmetric trimming of the propensity score⁵⁰  at the first and 99th percentiles. (6) To reduce potential bias due to missing data on indication (20.5% of the cohort), we used inverse probability-of-missingness weights⁵¹  in combination with refit SMR weights estimated in adolescents with complete data. (7) To examine associations with specific fluoroquinolones, we restricted the fluoroquinolone-treated group to ciprofloxacin (rupture events were too rare to examine other fluoroquinolones alone). Finally, we also conducted a bias analysis to assess the impact of possible differential outcome misclassification (Supplemental Information).

There were 15.9 million outpatient fills for an oral fluoroquinolone or comparator antibiotic among adolescents (aged 12–18 years) between 2000 and 2018. In the final study population (Fig 1), 4.4 million unique adolescents experienced a total of 7.6 million eligible antibiotic treatments, including 275 767 (3.6%) for fluoroquinolones. Ciprofloxacin was the most common fluoroquinolone (73%), and azithromycin (45%) and amoxicillin-clavulanate (19%) were the most common comparators; day supply was similar between fluoroquinolones and comparators (Supplemental Table 4).

Table 1 includes the cohort characteristics stratified by treatment. Before weighting (crude), adolescents treated with fluoroquinolones were more likely to be female (69%) and older (median age 17) compared with adolescents treated with the comparator antibiotics (female 54%, median age 15). A greater proportion of fluoroquinolone treatments (32%) than comparator treatments (8%) included a diagnosis that almost always requires an antibiotic (tier 1). As shown in Supplemental Table 5, UTI was the most common indication for fluoroquinolone treatment (29%), followed by gastrointestinal infections (8%) and sinusitis (7%); for comparators, the most common indications were sinusitis (18%), pharyngitis (17%), and skin, cutaneous, and mucosal infections (8%).

We identified 1478 tendon ruptures in the 180 days of follow-up for a crude rate of 0.43 per 1000 person-years (Supplemental Table 6). There were 32 335 tendinitis diagnoses (rate of 9.45 per 1000 person-years). For the negative control outcome, there were 7366 clavicle fractures for a crude rate of 2.15 per 1000 person-years.

SMR weights effectively balanced the treatment groups on measured covariates (Table 1, Supplemental Fig 6, Supplemental Table 5). The weighted risk difference per 100 000 adolescents for clavicle fracture and the negative control outcomes from the ITT analysis and the per-protocol analysis are depicted in Supplemental Figure 7. The difference was close to zero (null) before 90 days of follow-up. After 90 days, the difference became negative. Given that it is not biologically plausible that fluoroquinolones are protective against clavicle fracture >90 days after treatment initiation, we believe there is residual bias in this period and restrict presentation of the remaining results to 90 days of follow-up.

Figures 2 and 3 present the crude and weighted risk curves by treatment group and the weighted risk difference for tendon rupture and tendinitis, respectively, from the ITT analysis. The numeric results at 15, 30, and 90 days are presented in Supplemental Table 7. The weighted 90-day risk of tendon rupture was higher among the fluoroquinolone-treated adolescents (13.6 per 100 000 adolescents; 95% CI 9.4 to 17.8) compared with those treated with the comparator antibiotics (11.6 per 100 000 adolescents; 95% CI 9.8 to 13.5) for a difference of 1.9 per 100 000 adolescents (95% CI −2.6 to 6.4). The NNTH indicated that 52 632 adolescents would need to be treated with fluoroquinolones instead of a comparator antibiotic for 1 additional adolescent to experience tendon rupture within 90 days. For tendinitis, the 90-day ITT risk difference was 22.7 per 100 000 adolescents (95% CI 4.1 to 41.3) for an NNTH of 4405. The per-protocol analysis results were similar (Supplemental Table 8, Supplemental Fig 8).

The 30- and 90-day weighted risk differences per 100 000 adolescents for tendon rupture and tendinitis from each sensitivity analysis and the ITT analysis are presented in Figure 4. The tendon rupture sensitivity analysis estimates for 90-day risk difference per 100 000 adolescents ranged from −1.8 (95% CI −8.6 to 5.0) from the analysis restricted to the first eligible new-user period to 4.2 (95% CI −0.8 to 9.1) from the analysis in which missing data weights were used. The tendinitis sensitivity analysis 90-day estimates ranged from 8.3 (95% CI −14.9 to 31.4) from the first new-user period analysis to 38.7 (95% CI 7.5 to 70.0) from the analysis restricted to UTI indications. The bias analysis indicated that if differential misclassification of the outcome were present, our primary analysis results would have overestimated the fluoroquinolone-associated excess risk of tendon rupture and tendinitis (Supplemental Table 9).

In a cohort of 4.4 million 12 to 18-year-olds with 7.6 million antibiotic fills, we observed an elevated risk of tendon rupture associated with fluoroquinolone antibiotics compared with other broad-spectrum antibiotics. However, tendon rupture was rare, the risk difference was extremely small (even absent in some sensitivity analyses), and our primary results would overestimate the fluoroquinolone-associated excess risk in the presence of differential outcome misclassification. According to these data, on average, >50 000 adolescents would need to be treated with a fluoroquinolone for 1 additional tendon rupture to occur. The fluoroquinolone-associated excess risk was larger for tendinitis but was still small (1 additional event for every 4400 adolescents exposed) and was potentially inflated by differential outcome misclassification.

Few epidemiological studies have been focused on tendon-related adverse effects of fluoroquinolones in children. The first and only large comparative epidemiological study in children was conducted in 2002, and the 60-day risk of a composite outcome of tendon or joint injury was compared between fluoroquinolone and azithromycin treatment in nearly 21 500 children ≤18 years old.²⁸  The age- and sex-adjusted relative risk was 1.08 (95% CI 0.77 to 1.52), and the authors concluded that fluoroquinolone treatment was not associated with an increased risk. The authors validated outcomes via review of linked medical records, reducing potential bias from outcome misclassification⁵² ; however, the sample size was small (given the rarity of the outcome), the analysis did not robustly address confounding, and azithromycin alone may not be an appropriate comparator because it is not used for UTI, a common fluoroquinolone indication. In addition, by using a composite outcome, tendon-related adverse effects alone were not examined.

Investigators of multiple large epidemiological studies^12–21  have observed an association between fluoroquinolones and tendon injury or rupture in adults. Because the authors of these studies typically reported odds ratios or risk ratios without presenting adjusted risks, comparison with our results is challenging. Although investigators of some adult studies have observed large associations (eg, relative risk of 6.29¹⁸ ), it can be difficult to translate this association into potential harms for treated patients. Of the 3 adult studies in which rate differences were estimated, the excess rate difference for tendon rupture in 60 days of follow-up was 29 cases per 100 000 person-years,²¹  and rate differences for tendon injury were 320 cases per 100 000 person-years in one study¹²  and 1461 cases per 100 000 person-years in another.¹³  Although direct comparison of rates and risks requires assumptions (eg, constant rates over follow-up, valid published estimates), it appears that previous published estimates were higher in adults than what we observed in adolescents in this study.

In one adult study, the authors did not observe an association between fluoroquinolones and Achilles tendon rupture and hypothesized that observed associations in other studies may be the result of bias from differential misclassification of the outcome.¹⁵  The authors hypothesized that increased awareness of potential tendon-related adverse effects made providers more likely to include a tendon-related code for patients with recent fluoroquinolone exposure. To address this potential source of bias, we conducted a sensitivity analysis restricted to the time period before the 2008 FDA warning and a bias analysis. Although awareness of tendon-related adverse effects was likely present before the FDA warning, we hypothesize that it would have been lower before the warning than after the warning, which thus provides information on the likely direction of bias. These analyses suggest that our primary analysis, assuming no misclassification, is likely an overestimate of the excess risk due to fluoroquinolones.

Our analysis has a number of strengths. Our cohort includes >7.6 million antibiotic fills for >4 million privately-insured adolescents across the United States over 2 decades. We implemented both an ITT analysis and a per-protocol analysis, which accounts for treatment switching and can elucidate potential bias toward the null inherent to ITT designs.⁴⁵  We conducted a number of sensitivity analyses and a bias analysis to understand how our study design and analytic decisions might impact results. Importantly, the results of these analyses did not change our interpretations of the findings. Use of comparator broad-spectrum antibiotics that cover the range of potential indications of fluoroquinolones was also a strength, as was our implementation of a weighted analysis to control for measured confounding and estimate marginal risk differences. Lastly, we used a negative control outcome to assess residual confounding.

This study has limitations. Despite the size and scope afforded by MarketScan, including antibiotic prescriptions and injury care across the full spectrum of health care settings, our results may not be generalizable to adolescents outside this database, particularly those on Medicaid or without insurance. Our outcome definition relied on diagnosis and CPT codes, which are susceptible to misclassification. In particular, tendinitis identification may have a low positive predictive value (ie, many false-positives) because of “rule-out” diagnoses and subjectivity of this diagnosis. Results from our bias analysis indicate that differential outcome misclassification likely resulted in upward bias, which would overestimate the excess harm associated with fluoroquinolones. Adolescents who filled a prescription but did not actually take the medication would have been inappropriately included in the analysis, although this would likely be nondifferential with respect to the outcome. In our analysis, we assumed that censoring due to disenrollment or treatment switching was random, and if incorrect, our results could be biased.⁵³  Finally, although our negative control outcome analysis suggested good control of confounding, there could be residual bias related to differences in the confounding structure of tendon injury and clavicle fracture.

Fluoroquinolone treatment was associated with excess risk of tendon rupture and tendinitis compared with alternate antibiotic therapy. However, these differences were extremely small, and tendon-related adverse events were rare. This analysis provides important context for understanding the risk/benefit ratio of fluoroquinolones for adolescents. Given the rarity of tendon injury and the small excess risks associated with fluoroquinolones, more common adverse drug effects may be more important to consider for treatment decision-making, particularly in adolescents without other risk factors for tendon injury.

CI

confidence interval

CPT

Current Procedural Terminology

FDA

US Food and Drug Administration

ITT

intention-to-treat

NNTH

number needed to treat to harm

SMR

standardized mortality ratio

UTI

urinary tract infection