In this nationwide, retrospective case-control study of 312,404 cases of new-onset AMD and 312,376 matched controls from the general population, we demonstrated an association between antibiotic exposure and increased odds of a new-onset ICD code diagnosis of AMD. This association persisted after adjustment for well-established AMD risk factors. Fluoroquinolones and aminoglycosides increased the odds of a new-onset ICD code diagnosis of AMD the most, as compared to other antibiotic classes. Notably, there appeared to be a frequency- and duration-dependence to the association, with greater cumulative number of prescriptions and day supply of antibiotics linked with increasing odds. Additionally, exposure to broad-spectrum antibiotics was associated with greater odds of new-onset AMD compared to narrow-spectrum antibiotics.
These findings support the hypothesis that larger perturbations to the gut microbiome, triggered by either broader antibiotic coverage or greater cumulative exposure, may increase the odds of developing AMD. Although outside the scope of this study, there are several biologically plausible explanations for this finding. Antibiotic exposure has been shown to cause persistent disruptions to the composition of the gut microbiome, even when administered in short courses.
20,21 Perturbations to the diversity of the gut microbiome are also more severe for broad-spectrum as compared to narrow-spectrum antibiotics.
22 Disruption to the gut microbiome is termed
dysbiosis, a state linked with greater intestinal wall permeability that enables translocation of microbial endotoxins, which may be possible sources and triggers of systemic inflammation.
23 In a mouse model of neovascular AMD, Andriessen et al.
24 found that feeding mice a high-fat diet triggered gut dysbiosis, thereby augmenting intestinal permeability to microbial products that induced systemic inflammation and exacerbated choroidal neovascularization. In AMD, inflammation causes breaks in the retinal pigment epithelium and Bruch's membrane, leading to drusen, progressive degeneration of the photoreceptors, and neovascular development.
25 Hence, we suggest that the observed frequency- and duration-dependent association may reflect a trend of sustained guy dysbiosis with inflammatory spikes that eventually disseminate along the gut-retina axis. Additionally, the three-times increased odds of a new-onset ICD code diagnosis of AMD with broad- as compared to narrow-spectrum antibiotics may be indicative that greater deviations from the gut microbiome's homeostatic baseline further exacerbate inflammatory signals along the gut-retina axis.
The relationship between antibiotics and chronic disease development does not appear to be confined to AMD. Antibiotic exposure in early childhood has been linked with the development of celiac disease, type 1 diabetes, and juvenile idiopathic arthritis.
26–28 Antibiotic use in adults also appears to increase the risk of IBD and rheumatoid arthritis.
6,29 Hence, alterations to the gut microbiome due to widespread antibiotic use may underlie or influence a variety of immune-mediated or inflammatory diseases.
It is also possible that the association between antibiotic exposure and a new-onset ICD code diagnosis of AMD is secondary to, or at least partially due to, a direct effect of antibiotics on retinal tissue. Studies have identified that fluoroquinolones are linked with connective tissue complications, including Achilles tendon rupture and retinal detachment, the latter of which is speculated to be through damage to the structurally integral collagen in the vitreous body.
30,31 Interestingly, cultured human retinal pigment epithelium (RPE) cells treated with ciprofloxacin, a fluoroquinolone, displayed decreased viability with corresponding upregulation of inflammatory and pro-apoptotic genes.
32 This raises the possibility that fluoroquinolones exert a cytotoxic effect on RPE cells, thereby raising AMD risk. In a small prospective study of 76 patients who received oral fluoroquinolones and 50 sex-matched controls, Ozcelik-Kose et al.
33 found that fluoroquinolones had no effect on retinal degeneration at one week and one month after treatment. The patient sample used in that study was small, limited to a one-month follow-up, and younger in age (mean age, 49.8 years). Comparably, we used a large sample of patients aged 55 years or older with a 24-month lookback period. Our older population better captures those at risk of AMD, whereas a 24-month lookback period extends the time that may be required for antibiotics to exert a degenerative effect on the retina.
Aminoglycosides and fluoroquinolones were associated with the greatest odds of a new-onset ICD code diagnosis of AMD among the antibiotic classes. For fluoroquinolones, this finding may be partially due to the excellent oral bioavailability of newer agents that allows for greater systemic absorption.
34 In addition, fluoroquinolones persistently deplete the abundance of anaerobic gut bacteria such as
Bacteroides spp. and
Bifidobacterium spp.
35 These genera produce short-chain fatty acids, anti-inflammatory mediators that can also reduce gut permeability.
36,37 Hence, depletion of these beneficial bacteria by fluoroquinolones may be associated with systemic circulation of inflammatory products that eventually localize to the retina. For aminoglycosides, this finding is less clear, because they generally do not cover gram-positive or anaerobic bacteria. Furthermore, literature detailing the impact of aminoglycosides on the gut microbiome are lacking, compared to other antibiotic classes.
35 Additional studies are required to precisely define antibiotic-mediated alterations to the gut microbiome and how these may propagate to the retina.
Interestingly, fluoroquinolones and aminoglycosides were associated with a greater odds of new-onset coding of wet AMD compared to dry AMD. Although these disease classifications share several similarities, they are also unique in their risk factors and pathogenesis.
38 These results suggest that fluoroquinolones and aminoglycosides may interact differently with the pathways that drive development and progression of dry and wet AMD. In support of this, we also found an increasing odds ratio of a new-onset ICD code diagnosis of wet AMD, but not dry AMD, as the number of prescriptions for these two classes of antibiotics increased (Aminoglycosides: one prescription: OR = 1.52; 95% CI, 1.41–1.63; two prescriptions: OR = 1.56; 95% CI, 1.33–1.81; three or more prescriptions: OR = 1.63; 95% CI, 1.34–1.99; Quinolones: one prescription: OR = 1.25; 95% CI, 1.19–1.32; two prescriptions: OR = 1.32; 95% CI, 1.23–1.41; three or more prescriptions: OR = 1.51; 95% CI, 1.41–1.61). Further research is required to understand how these medications may modulate AMD development and progression
Given that outpatient oral antibiotics comprise more than 200 million prescriptions per year in the United States, we identified that these antibiotic prescriptions may pose a significant burden on AMD development.
18 We identified three commonly prescribed antibiotics, including amoxicillin, azithromycin, and trimethoprim/sulfamethoxazole, that failed to increase or decrease the odds of AMD when administered as a single prescription. This finding may help to guide judicious prescription patterns when multiple antibacterial agents are available and appropriate for use. For other commonly prescribed antibiotics that included amoxicillin, azithromycin, cephalexin, ciprofloxacin, or doxycycline, however, we saw evidence of a frequency-dependent association. Interestingly, doxycycline is currently being studied in a clinical trial to identify if it can attenuate the progression of geographic atrophy through its anti-inflammatory effects at low doses.
39 Future observational studies should examine the impact of antibiotic dosing on AMD development.
We found that exposure to fluoroquinolones in the zero to six months preceding AMD diagnosis was associated with the greatest odds of a new-onset ICD code diagnosis of AMD compared to exposure at six to 12 months or one to two years. These data may support the notion that fluoroquinolones mediate an acute inflammatory or cytotoxic effect which accelerates AMD development in patients with a heightened baseline risk. For other antibiotic classes, however, exposure at one to two years generally increased the odds of a new-onset ICD code diagnosis of AMD more than exposure at zero to one year preceding diagnosis. This finding may be indicative of sustained gut dysbiosis and chronic inflammation from long-term antibiotic use. We also investigated the interaction terms between female sex and antibiotic classification, which revealed that the odds of a new-onset ICD code diagnosis of AMD associated with fluoroquinolones was mildly attenuated in females. This finding is counterintuitive, because female sex has been proposed as a risk factor for AMD development. Hence, there is likely an unaccounted variable, such as lifestyle factors, that may help to explain this modest attenuation. For the other antibiotic classes, except for penicillins, female sex did not significantly alter odds of a new-onset ICD code diagnosis of AMD when interaction terms were included, suggesting that sex does not further modulate AMD risk in patients exposed to antibiotics.
Strengths of this study include the use of MarketScan databases, which offer access to a large, nationwide sample of Americans with employer-sponsored health insurance or Medicare supplemental coverage. This allowed for a precise estimation of the odds of a new-onset ICD code diagnosis of onset AMD following exposure to antibiotics. Second, we showed evidence of a strong frequency-response relationship. The presence of this relationship strongly supports an association between antibiotic exposure and AMD diagnosis. Additionally, our multivariable analyses corroborated and adjusted for well-supported AMD risk factors. Thus our study design minimized possible confounding that would otherwise have compromised internal validity. Finally, this study is likely not susceptible to reverse causation. This is to say that antibiotics are unlikely to be prescribed for symptoms related to AMD in its undiagnosed stages. For this reason, we did not include a lead-in period where antibiotic prescriptions were excluded from cumulative tabulations.
We acknowledge several limitations to the study design. First, records of pharmacological dispensations may not capture actual antibiotic usage. Second, cases of AMD were identified using diagnosis codes and were not verified with retinal imaging. Such codes may be incorrect or underrepresent AMD, but with a large study population, the effect of aberrant coding is likely small, and it should not differ between cases and controls. With that said, the use of use of a claims database may still give way to misclassification bias in this study. Diagnosis of AMD with ICD coding does not necessarily correspond to the development of new-onset AMD. There may be a delay between development of disease and clinic diagnosis as determined by ICD coding. We did not account for the body mass index of subjects, which is a limitation as obesity has been linked to a greater risk of AMD, though this association is inconsistently characterized.
40 Furthermore, we were unable to delineate between oral, topical, intravenous, intravitreal, or other routes of administration—an opportunity for future research to determine whether the effects of antibiotics on AMD development are reserved to certain routes of administration. Additionally, lifetime or early-life exposure to antibiotics were not assessed, which may better capture the sustained gut dysbiosis and chronic inflammation that we hypothesize may contribute to AMD pathogenesis. We were unable to capture the indication for antibiotic prescriptions, which could introduce confounding by indication. For example, it may be that repeated infections are associated with AMD, as opposed to the antibiotics prescribed to treat such infections. Alternatively, individuals with recurrent infections who require more frequent and longer courses of antibiotic therapy could have weakened or dysfunctional immune responses, leading to systemic inflammation that raises AMD risk. However, the mean Charlson Comorbidity Index score for both cases and matched controls was 1.1. This indicates that our study population was afflicted with relatively few comorbidities that might introduce this source of confounding. In addition, with the large sample size used in this study, it is unclear whether statistical significance would also carry clinical significance. However, given the abundance with which antibiotics are prescribed in clinical practice, we believe that there is potential clinical relevance to this finding that merits further exploration. Finally, the retrospective nature of the study only allows for us to draw conclusions regarding the association of antibiotics and AMD. Long-term, prospective epidemiological studies are required to determine whether antibiotics increase risk of AMD development.
Although it is too early to draw definitive conclusions and to substantially alter clinical practice due to these findings, several results merit consideration and can guide future studies. Our results suggest that use of fewer prescriptions, limiting total days of exposure, and choosing agents with appropriately narrow coverage may be important components of antibiotic stewardship to prevent AMD development. For example, avoidance of aminoglycosides and quinolones when an alternative option is available and clinically efficacious should be considered. Further population-level studies of multiple independent databases are needed to confirm the results herein, given that we carried out the first analysis of antibiotic exposure and a new-onset ICD code diagnosis of AMD.