Oral Fluoroquinolones and the Risk of Uveitis: A Retrospective Cohort Study

Abstract

Importance

Fluoroquinolones (FQ) are the most commonly prescribed antibiotic in the outpatient setting. Recent reports have implicated an association between oral fluoroquinolones and an increased risk of uveitis.

Objective

To determine the hazard of uveitis with oral fluoroquinolone use.

Design

This is a retrospective cohort study using medical claims data from a large, national U.S. insurer. Cohorts were created using every new user of an oral fluoroquinolone or beta-lactam (BL) antibiotic prescription from 2000–2012.

Setting

Ambulatory care centers across the United States.

Participants

Any new user of an oral fluoroquinolone or beta-lactam antibiotic with at least 24 months of data prior to the date of the prescription was eligible. Exclusion criteria consisted of any previous diagnosis of uveitis or a uveitis-associated systemic illness. Subjects were censored for a new diagnosis of a uveitis-associated systemic illness, the end of an observation period, use of the other class of antibiotic, or removal from the plan.

Main Outcome Measure

The main outcome was the hazard of being diagnosed with uveitis after FQ prescription compared to a BL prescription using multivariate Cox regression.

Results

843,854 FQ patients and 3,543,797 BL patients were included in analysis. Univariate analysis showed an increased hazard of uveitis for FQ users at every time point (HR range: 1.34–1.64; p<0.001 for all comparisons). After controlling for age, race, and gender using multivariate analysis, no hazard for developing uveitis at the 30-, 60-, or 90-day observation windows was seen (HRs 0.96–1.05; p>0.38 for all comparisons). The 365-day observation showed a small increase in hazard for the FQ cohort (HR=1.11; 95% CI: 1.05–1.17; p<0.001). Moxifloxacin showed an increased hazard for uveitis at every time point (HR range: 1.47–1.75; p<0.001 for all comparisons). Secondary analysis demonstrated a similar hazard at 365 days for a later diagnosis of a uveitis-associated systemic illness after FQ use (HR range: 1.46–1.96; p<0.001 for all comparisons).

Conclusions

These data do not support an association between oral fluoroquinolones and uveitis. Instead, this study shows an association between oral fluoroquinolones and the risk for uveitis-associated systemic illnesses, a possible source of bias that could explain the findings of previous studies.

Introduction

Oral fluoroquinolones (FQ) have been implicated as a cause of uveitis in multiple case reports and case series.^1,2,3,4 Although the first formal study identified no association between the two⁵, further concern was raised when a second study focusing on ciprofloxacin and moxifloxacin recently demonstrated a 1.96–2.98-fold higher risk of uveitis compared with non-users.⁶ These results are clinically non-trivial since fluoroquinolones are so commonly prescribed for many outpatient diseases, including respiratory, urinary, soft tissue, and gastrointestinal infections. At present, fluoroquinolones are the most commonly prescribed class of antibiotic in the ambulatory setting.⁷

The exact mechanism by which fluoroquinolones may induce uveitis is unclear. Pharmacokinetic studies using whole-body scintigraphy have shown that moxifloxacin concentrates in melanin-containing structures in the meninges, skin, and eye.⁸ It is well established that fluoroquinolones can sensitize the skin to phototoxicity, thereby predisposing it to photodermatitis. Moreover, melanin has been shown experimentally to potentiate fluoroquinolone-induced phototoxicity to the skin.⁹ Thus, it has been postulated that moxifloxacin may have some direct toxic effect on the iris pigment.⁸ This is consistent with most previous reports of fluoroquinolone-induced uveitis, which featured few inflammatory cells in the anterior chamber (AC) but rather pigment dispersion in the AC with diffuse iris transillumination defects (TIDs) and sometimes pupil mydriasis.² Almost all reported cases of uveitis thus far have occurred two to four weeks after administration of the fluoroquinolone, and usually within two weeks.

While the finding in that case-control study was novel, the most robust results are those that are confirmed through different methodologies across different populations. We aim to evaluate the risk of uveitis after oral fluoroquinolones using a retrospective cohort study in a large, national United States database.

Methods

Dataset

Data was abstracted from the Clinformatics™ Data Mart Database (OptumInsight, Eden Prairie, MN), which contains the de-identified medical claims of all beneficiaries from a large managed care network in the United States. Included within the database are all outpatient medical claims (office visits and associated diagnoses), all outpatient pharmaceutical prescriptions filled, and demographic data for each beneficiary during their enrollment in the insurance plan. The subset of data available for this study included all patients in the database from January 1, 2000 to January 30, 2013. The University of Pennsylvania’s Institutional Review Board deemed this study exempt from review due to the de-identified nature of the data.

Cohorts

For this study, two cohorts were created for comparison, consisting of all patients who had received either a prescription for a fluoroquinolone or a beta-lactam (BL) antibiotic during their time in the plan. Beta-lactams were chosen as the comparison group because they share similar medical indications with FQs but are not thought to be associated with uveitis. The index date for each patient was considered the first date of either an FQ or BL prescription during their coverage period. Patients were required to have been enrolled in the insurance plan for at least 24 months prior to the index date. Exclusion also occurred if the patient had any previous diagnosis of uveitis or a systemic disease known to be associated with uveitis (see eTable 1 for all ICD9 codes used during this study) or received a prescription for both an FQ and BL on the index date. Due to the potentially chronic and indolent nature of uveitis, patients were only allowed to be included in a cohort once during the study regardless of multiple prescriptions.

Outcomes of interest

The primary outcome of interest was defined as having an incident diagnosis code for uveitis made by an eye-care provider. Next, due to the pathophysiologic mechanism suggested earlier, a secondary outcome using only cases of incident anterior uveitis was also examined. Due to previous work reporting a positive association between specific FQs and uveitis,⁶ a sub-analysis was performed examining individual FQ antibiotics. Lastly, a sensitivity analysis was performed in which a case was only considered to have occurred in which a steroid prescription (oral or topical) was filled by the patient within 30 days of the incident uveitis diagnosis.

Indication bias, which is any association between a drug and an outcome that is due to reasons underlying why the drug is actually prescribed, rather than a direct effect of the drug itself, is often a major limitation of these types of studies. Although using BL as a comparator group attempts to reduce this, another secondary analysis was done using as an outcome measure an incident diagnosis of a systemic disease known to be associated with uveitis. This was done because one can be more confident in a potential positive association between FQs and uveitis if there is no association between FQs and these systemic diseases, since these conditions are clearly not caused by FQ use (e.g. sarcoidosis, lupus, inflammatory bowel disease, etc.). Covariates used in multivariate analysis were age, gender, and race.

Statistical Analysis

Baseline demographic characteristics were assessed at the time of the index date and were analyzed using descriptive statistics. Mean averages and ranges were used for continuous variables and percentages for categorical variables. Cox proportional hazard models were used to analyze the time from prescription index date to the outcome of interest. The censoring date of eligible patients was defined as the earliest date among the following: diagnosis of uveitis, prescription for the alternate class of oral antibiotic filled after the index date, the patient’s exit from the insurance plan, or the end of the observation period. Hazard ratios were estimated for four observation periods: days 1–30, 1–60, 1–90, and 1–365 after the index date. STATA^® 12 (College Station, Texas) software was used for all statistical analysis.

Results

In total, 4,387,651 patients were included for analysis (843,854 FQ, 3,543,797 BL) (Fig. 1). Ciprofloxacin was the most commonly prescribed FQ, followed by levofloxacin and moxifloxacin (Table 1). 58.1% of the FQ cohort was female, while only 51.7% of the BL cohort was female. Patients ranged from 0–85 years of age. Average age in the FQ cohort was 53.2 years (IQ range 40–67 years) and 37.1 years (18–53 years) in the BL cohort. The FQ cohort also had a higher percentage of black people than the BL cohort.

Fig 1.

Table 1.

Baseline characteristics of cohorts

	Fluoroquinolone	Beta-lactam
Total patients	843,854	3,543,797
Mean age (years), IQ range	53.2, [40–67]	37.1, [18–53]
Female	58.1%	51.7%
Race (%)
White	72.1%	71.3%
Black	9.4%	8.7%
Hispanic	9.6%	9.1%
Asian	3.2%	3.5%
Other/Unknown	5.7%	7.4%
Fluoroquinolone type	N (% of FQ total)
Ciprofloxacin	602,157 (71.4%)
Levofloxacin	136,850 (16.2%)
Moxifloxacin	102,928 (12.1%)
All others	1919 (0.2%)

Table 2 shows the overall number of cases found at each time point for incident cases of uveitis, anterior uveitis, and systemic illnesses associated with uveitis. In the FQ cohort, the percentage of patients who were diagnosed with uveitis increased from 0.025% to 0.22% from the 30-day to the 365-day observation period. The BL cohort had a similar increase in diagnoses from 0.019% to 0.15%.

Table 2.

Outcomes of interest after index date

	Fluoroquinolone cohort (n=843,854)		Beta-lactam cohort (n=3,543,797)
	n	%	n	%
All Uveitis Diagnoses (Dx)
Within 30 days	208	0.025%	657	0.019%
60 days	377	0.045%	1132	0.032%
90 days	561	0.066%	1608	0.045%
365 days	1894	0.224%	5227	0.147%
Anterior Uveitis Dx
30 days	113	0.014%	361	0.010%
60 days	188	0.022%	569	0.016%
90 days	256	0.030%	782	0.022%
365 days	788	0.093%	2402	0.068%
Systemic Illness Dx
30 days	2557	0.303%	4684	0.132%
60 days	4042	0.479%	7811	0.220%
90 days	5250	0.622%	10,561	0.298%
365 days	13,080	1.550%	31,414	0.886%

Univariate analysis showed that female gender had a protective effect with regard to the hazard of developing uveitis. Older age and black race had higher hazards for developing uveitis at all time points. These associations remained even after multivariate analysis (data not shown). No confounding or effect modification by these variables was seen in the relationship between FQ use and the hazard for developing uveitis.

With multivariate analysis, an association was not identified between FQs and uveitis at the 30-, 60-, or 90-day time points (HR range: 0.96–1.05; p>0.38 for all comparisons) (Table 3). However, there was a small effect seen at 365 days with a hazard ratio of 1.11 (95% CI: 1.05–1.17; p<0.001) (Table 3). The sub-analysis of just anterior uveitis demonstrated that anterior uveitis constituted fewer of the total uveitis diagnoses in the FQ cohort (788/1894 or 41.6%) than in the BL cohort (2402/5227 or 45.9%)(Table 2). Multivariate analysis similarly showed no increased hazard at the 30-, 60-, and 90-day time points (HR range: 1.05–1.10; 95% CI: 0.84–1.30; p>0.19 for all comparisons), but similar to the all-uveitis result showed a slightly increased HR of 1.11 at 365 days (95% CI: 1.03–1.21; p=0.01) (Table 3). Sensitivity analysis with a stricter definition of a case that also required a prescription for steroids in addition to the incident diagnosis of uveitis had similar results to the main analysis with the 30-, 60-, and 90-day observation periods showing no association (HR range: 1.06–1.09; p>0.32 for all comparisons), and an increased hazard ratio (HR 1.13; 95% CI: 1.03–1.24; p=0.01) at the 365-day observation period.

Table 3.

Multivariate hazard ratios for fluoroquinolones versus beta-lactams

		Adjusted Hazard Ratio*	95% CI	P Value
All Uveitis
Within 30 days		0.96	0.82–1.13	0.61
	60 days	0.99	0.88–1.12	0.93
	90 days	1.05	0.95–1.16	0.38
	365 days	1.11	1.05–1.17	<0.001
Anterior Uveitis
	30 days	1.05	0.84–1.30	0.69
	60 days	1.10	0.92–1.30	0.30
	90 days	1.10	0.95–1.28	0.19
	365 days	1.11	1.03–1.21	0.01
Systemic Illness
	30 days	1.96	1.87, 2.07	<0.001
	60 days	1.81	1.74, 1.89	<0.001
	90 days	1.72	1.67, 1.79	<0.001
	365 days	1.46	1.42, 1.48	<0.001

^*Adjusted for age, gender, and race

Of all the FQs prescribed, only ciprofloxacin, moxifloxacin, and levofloxacin included sufficient numbers for multivariate analysis. An association was not identified between ciprofloxacin and uveitis at the 60-, 90-, or 365-day time points (HRs 0.88–1.03; 95% CI: 0.77–1.10; p>0.09 for all comparisons)(Table 4). Within 30 days, ciprofloxacin actually showed a protective effect against developing uveitis (HR 0.75; 95% CI: 0.61–0.92; p=0.005). An association was not identified between levofloxacin and uveitis at the 30-, 60-, and 90-day time points (HRs 1.03–1.29; 95% CIs: 0.80–1.72; p>0.09 for all comparison), but there was an increased hazard ratio of 1.17 seen at 365 days (95% CI: 1.04–1.31; p=0.007)(Table 4). Moxifloxacin showed an association with uveitis at every time point with HRs of 1.75 at 30 days that decreased to 1.47 at 365 days (95% CIs: 1.27–2.37; p<0.001 for all comparisons)(Table 4).

Table 4.

Multivariate hazard ratios for specific fluoroquinolones versus beta-lactams

		Adjusted Hazard Ratio*	95% CI	P Value
Ciprofloxacin
	Within 30 days	0.75	0.61, 0.92	0.005
	60 days	0.88	0.77, 1.02	0.09
	90 days	0.92	0.81, 1.03	0.14
	365 days	1.03	0.97, 1.10	0.32
Moxifloxacin
	30 days	1.75	1.28, 2.37	<0.001
	60 days	1.62	1.27, 2.06	<0.001
	90 days	1.69	1.39, 2.06	<0.001
	365 days	1.47	1.30, 1.65	<0.001
Levofloxacin
	30 days	1.29	0.96, 1.72	0.09
	60 days	1.03	0.80, 1.32	0.82
	90 days	1.14	0.94, 1.39	0.19
	365 days	1.17	1.04, 1.31	0.007

^*Adjusted for age, gender, and race

There are multiple systemic conditions that have well known associations with uveitis (eTable 1). A final secondary analysis was performed using any incident diagnosis of a known uveitis-associated systemic illness as the outcome. This analysis mirrored closely the results of the moxifloxacin-specific analysis above and revealed that any oral fluoroquinolone use was significantly associated with an incident diagnosis of one of these conditions at all time points (HR range: 1.46–1.96; 95% CIs: 1.42–2.07; p<0.001 for all comparisons)(Table 3).

Discussion

This study found no increased hazard of developing uveitis within 30, 60, or 90 days of taking an oral fluoroquinolone. There was a small increased hazard at the 365-day observation period. Similar results were found examining only anterior uveitis cases. Also, when moxifloxacin was studied alone, an increased hazard was seen at all time points. Of note, when uveitis-associated systemic illnesses were used as an outcome measure, there was an association between FQs and a later diagnosis of these illnesses that closely mirrored the association between moxifloxacin and uveitis.

Since fluoroquinolones achieve a high steady-state concentration in the serum shortly after onset of use and then quickly decay after discontinuation, one would expect adverse effects (AE) of the drug to occur within a short time frame after initiating treatment. With increasing time after use, the rate of AEs should decrease as the drug is excreted from the body, if it is in fact to blame for the effect. The findings in this study were not consistent with this pattern, as no significant association was seen for oral fluoroquinolones as a class and uveitis or anterior uveitis until the 365-day observation period.

A previous large, case-control series had found that new uveitis patients were more likely to currently be taking a fluoroquinolone than control patients. Specifically, there was a rate ratio of 2.98 for moxifloxacin and 1.98 for ciprofloxacin.⁶ That study used patients taking a non-antibiotic, finasteride, as its control group. The use of a non-antibiotic control may have introduced a source of bias known as indication bias. The medical indications for FQs are considerably different from those for finasteride. Thus, it may be the different underlying medical reasons for prescribing these drugs that differentially influence the outcome of interest, rather than the drugs themselves. Our study also evaluated specific FQs, finding an increased hazard for uveitis at all time points for moxifloxacin, but no increased hazards were noted for ciprofloxacin at any time point or levofloxacin through the first 90 days. Prima facie, the moxifloxacin data would seem to confirm the findings of the previous study; however, interestingly, these increased HRs for moxifloxacin closely mirrored the HR at all time points for FQs and an incident diagnosis of a uveitis-associated systemic disease.

We interpret these data to mean that FQs themselves are unlikely to cause uveitis, but rather a prescription for an FQ is a marker of patients with some broader immune dysregulation that predisposes them to an infection for which an FQ would be prescribed, which is then followed by uveitis. To be more specific, a patient with an inflammatory condition could have signs or symptoms that mimic infection, which is subsequently treated with antibiotics. For instance, a patient with a lung opacity on chest x-ray due to sarcoidosis might be mistakenly treated for pneumonia with moxifloxacin. Were that patient to develop uveitis following the antibiotic, one might falsely attribute a causal role to the FQ when the patient’s underlying, pre-diagnosed disease is more likely the culprit. It is highly unlikely that FQs cause inflammatory bowel disease, Behcet’s, or a myriad of other autoimmune diseases. Yet, the data show this very association. Lastly, another possibility is that infectious disease syndromes more commonly treated with FQs may be more likely to elicit a lasting immune response that could eventually manifest as uveitis, much in the way that acute infections with yersinia or salmonella are thought to trigger a lasting immune response in reactive arthritis and its associated uveitis.^10,11,12

We also suspect that indication bias is responsible for the association seen between moxifloxacin and uveitis at all time points in this study. One argument against this would be that levofloxacin (also a third-generation FQ) did not have similar results, but it is important to note that although the two medications’ indications overlap, they are not identical. Moxifloxacin is typically used for respiratory and skin infections, while levofloxacin has a broader range of indications. It is this difference in patient populations that we suspect has a higher propensity for an underlying uveitis-associated systemic disease. Hence, moxifloxacin is prescribed to this group and appears to cause uveitis when the results have actually been confounded by the indications for the drug itself.

There were several strengths to this study. First, this is a large national database, decreasing the likelihood that outlying prescribing patterns by a minority of doctors could influence the results of the study. Second, if assessing drug safety is the goal of a study, a cohort design is a somewhat more robust methodology than a case-control study because of the ability to examine all patients who received an oral FQ. Using the entire cohort causes less data loss compared to evaluating only the small fraction of uveitis patients that used oral fluoroquinolones in a case-control study. Next, the multiple sub-analyses used within this study help to shed light on potential associations seen in previous reports.

Limitations of this study must also be noted. While the FQ and BL groups in general have similar indications for use, the exact diagnosis for which each prescription was written cannot be identified in this database. Next, this study included only outpatient prescriptions. Many FQs and BLs are prescribed orally or intravenously in the inpatient setting, and none of these prescriptions was captured in the data set. Moreover, the database cannot confirm that once patients filled their antibiotic prescriptions the full course was taken. Diagnoses were made based on ICD9 codes used in billing data, and are not able to be verified with clinical charts. Lastly, iris transillumination defects may be a sign of fluoroquinolone-associated uveitis, but TIDs do not have a specific ICD9 code and thus cannot be identified in this database. It is possible that some of these cases were not coded as uveitis by eye-care providers and thus not captured in these data.

Conclusion

The findings of this study do not support the association between oral fluoroquinolone use and uveitis risk. Although there was an increased risk of uveitis one year after FQ use, this time frame is not consistent with an expected AE for a drug that is used sporadically and quickly eliminated from the body after cessation. Given the systemic illness findings in this study, it is more likely that FQ use is a marker that identifies a group of patients who are more likely to develop a disease associated with uveitis, rather than causing uveitis itself.