Influence of Fluoroquinolone Susceptibility on the Therapeutic Response of Fluoroquinolone-Treated Bacterial Keratitis

Objective  To estimate how a corneal isolate's minimal inhibitory concentration for a fluoroquinolone agent affects the rate of clinical response of bacterial keratitis to fluoroquinolone therapy.

Design  Prospective cohort study.

Methods  Six hundred sixty-three individuals with suspected bacterial keratitis underwent diagnostic corneal scraping and were treated with topical 0.3% ciprofloxacin solution or ointment. Of 407 patients with culture-confirmed bacterial keratitis, improvement and cure rates with ciprofloxacin monotherapy were estimated for 391 who had in vitro ciprofloxacin susceptibility of the principal corneal isolate.

Main Outcome Measures  Slitlamp biomicroscopic assessment for clinical improvement of corneal inflammation and for clinical cure with complete reepithelialization.

Results  Adjusted rates of improvement and of cure were reduced, respectively, by 43% (95% confidence limits, 8%, 64%) and by 29% (95% confidence limits, 0%, 49%) among corneal infections having a ciprofloxacin minimal inhibitory concentration above 1.0 µg/mL compared with those with more sensitive isolates.

Conclusions  Corneal infection by relatively ciprofloxacin-resistant bacteria responds more slowly to ciprofloxacin therapy. Antibacterial susceptibility testing of corneal cultures may predict the fluoroquinolone therapeutic response rate of bacterial keratitis.

MICROBIAL INFECTIONS of the cornea affect 1 in 10 000 Americans per year, ¹ and the incidence in developing countries is 10 times higher.² Antimicrobial therapy aims to remedy this costly and sight-threatening disease.

Ciprofloxacin³ is a topical ophthalmic fluoroquinolone used to treat bacterial keratitis.^4-6 By disrupting bacterial DNA synthesis, ⁷ fluoroquinolones are bactericidal against many corneal pathogens.⁸ Clinical trials show that ciprofloxacin therapy is effective, ^9-12 although resistant keratitis isolates are emerging.^13-25

The diagnosis of corneal infection can be assisted by the microbiology laboratory, ²⁶ but how to select appropriate treatment remains controversial.^27,28 Consensus is lacking on what ways antimicrobial therapy could be guided by cultures and susceptibility testing for microbial keratitis.^29-33 Whether an antibiotic's laboratory sensitivity relates to its efficacy in ophthalmic practice has not been established.³⁴ This prospective cohort study examined how the in vitro susceptibility of corneal isolates to ciprofloxacin influenced the rate of clinical resolution among patients treated with ciprofloxacin.

This prospective cohort study was nested in a set of multicenter clinical trials of topical ophthalmic ciprofloxacin hydrochloride, a 0.3% solution or ointment (Ciloxan, Alcon Laboratories, Inc, Fort Worth, Tex), conducted at sites in the United States, the Caribbean, France, Belgium, Portugal, Germany, and India.^9-11 Protocols were approved by local institutional review boards, and study participants gave written informed consent. Eligibility required suppurative, ulcerative keratitis without bilateral involvement, infection of the only functional eye, keratoscleritis, or endophthalmitis. Children younger than 2 years, women who were pregnant or of childbearing age but not practicing birth control, and individuals with allergy to ciprofloxacin were ineligible. Between July 28, 1989, and March 14, 1995, 663 patients were enrolled by 51 investigators in 40 cities.^9-11 Patients enrolled during 1989 to 1990 were instructed to instill ciprofloxacin hydrochloride solution, 3 mg/mL, every 15 minutes for 6 hours, every 30 minutes for 18 hours, every hour for 24 hours, then every 4 hours for 12 days.⁹ During 1990 to 1992, patients treated with ciprofloxacin hydrochlorideointment, 3 mg/g, were told to apply a 1-cm ribbon every 1 to 2 hours for 2 days, then every 4 hours for 12 days.¹⁰ Trial participants during 1992 to 1995 were directed to use ciprofloxacin eyedrops every 30 minutes for 6 hours, every hour for 3 days, every 2 hours for 2 days, then every 4 hours for 9 days.¹¹ In all trials dosing modifications after 14 days were adjusted individually, and investigators decided when to end therapy.

Data were collected prospectively onto preprinted study forms. Information on clinical history included the duration of symptoms and any use of a topical antibiotic or topical corticosteroid within the preceding 24 hours. Slitlamp biomicroscopy estimated the location of the infiltrate, the longest diameter of the epithelial defect, the maximal depth of stromal infiltration, and the amount of aqueous humor cells. Improvement was judged to occur when signs of corneal inflammation first diminished and infection appeared controlled. Criteria for cure were resolved corneal inflammation, no evidence of active bacterial infection, and complete corneal reepithelialization. Clinical examinations were scheduled at 1, 2, 7, and 14 days of ciprofloxacin treatment; on the day that ciprofloxacin treatment ceased; and at 1 week later. Ciprofloxacin therapy was stopped for worsening, sustained lack of improvement over any 1-week interval for uncured patients, or a clinically important adverse event. Censoring occurred on the day that ciprofloxacin therapy stopped, on report of isolation of a fungus or amoeba from corneal scrapings, or at the last examination day for patients lost to follow-up.

Enrolled patients who had bacterial but no fungal or amoebic growth on any medium were considered to have culture-confirmed bacterial keratitis and to be eligible for analysis. Subcultures of bacterial isolates were numerically labeled and sent to a reference laboratory for agar-dilution or broth-dilution susceptibility testing.^35,36 The minimal inhibitory concentration (MIC) was the lowest ciprofloxacin concentration without visible growth for 16 to 20 hours at 35°C, and 1.00 µg/m Lwas selected as the cutoff point for defining resistance.³⁷ The principal corneal isolate for patients with polybacterial keratitis was defined as the microorganism having the highest ciprofloxacin MIC and then by the following ordered sequence: mycobacteria or actinomycetes, gram-negative rods or cocci, gram-positive cocci other than Staphylococcus epidermidis or Micrococcus species, gram-positive rods other than Propionibacterium species and coryneforms, S epidermidis or Micrococcus species, then Propionibacterium species or coryneforms. For patients with more than 1 corneal isolate having identical ciprofloxacin MIC values within one of these groups, the principal isolate was chosen by the greatest amount of growth.

Estimates of cumulative probability for improvement and for cure were plotted by MIC category. Rate ratios with 95% confidence limits (CLs) were estimated by maximum-likelihood Cox proportional hazards regression to evaluate the effect of MIC and 13 other independent predictors, including demographic characteristics, recent medications, severity, bacterial species, and ciprofloxacin formulation, on the rate of clinical improvement during ciprofloxacin therapy. A conditional-risk model for both improvement and cure was stratified by end point order, measuring time to cure from study enrollment.³⁸ Martingale residuals were used to examine the functional form of covariates. Effect modification was assessed by testing first-order product terms that included the categorized MIC variable. Beginning with covariates with an unadjusted Wald P<.25 and any interaction term with P<.10, a backward selection process was based on the likelihood ratio test and on the precision of the hazard ratio associated with the categorized MIC variable. Multiple event-time Cox modeling similarly proceeded by stepwise backward selection using the likelihood ratio test. The assumption of invariant relative risk during follow-up was examined using scaled Schoenfeld residuals. A stratified Cox model was fit to account for any predictor not satisfying the local proportional hazards assumption. The extent of bias from informative censoring of ciprofloxacin-treated patients who were switched to an alternative agent was examined by empirical bounds from zero to infinity on improvement or cure probabilities.³⁹ Attrition bias was assessed by estimating bounds that assumed improvement time varied from zero to infinity for individuals not completing follow-up through 14 days.⁴⁰

Analyses were performed using statistical software.⁴¹ Two-tailed P values in the Cox models are based on the Wald test using Efron's method for handling ties. The sample size of the study was a function of trial enrollment, so power was calculated for univariate survival analysis.⁴² A 60% power was estimated a priori assuming a 2-sided type I error rate of 5% to detect an improved rate ratio of 0.5 or less for fluoroquinolone-resistant bacterial keratitis, using an anticipated 80% cumulative incidence of clinical improvement among patients with fluoroquinolone-sensitive bacterial keratitis⁴³ and a 10% prevalence of ciprofloxacin-resistant microorganisms.^{15,18-20,22,23}

One or more microorganisms were recovered from 420 (63.3%) of 663 treated patients. Thirteen culture-positive patients (3.1%) were excluded because a fungus (12 patients) or Acanthamoeba (1 patient) was isolated; 8 (61.5%) of these cases had bacterial co-isolates that were not considered in the analysis. Sixteen (3.9%) of 407 patients with culture-confirmed bacterial keratitis did not have a ciprofloxacin MIC measured; these untested primary isolates included staphylococci (4 patients), streptococci (4 patients), gram-negative rods (3 patients), and gram-positive rods (5 patients). Of 391 patients with culture-confirmed, susceptibility-tested bacterial keratitis, 93 (23.8%) had more than 1 bacterial isolate, including 22 patients with 3 or more isolates. The proportion of patients with each baseline clinical characteristic was similar among categories of ciprofloxacin susceptibility, but differences in ciprofloxacin susceptibility were present among bacterial groups (Table 1).

Using a ciprofloxacin MIC breakpoint of 1.00 µg/mL, 333 (91.2%) of 365 patients whose principal corneal isolate was sensitive improved compared with 19 (73.1%) of 26 individuals with a ciprofloxacin-resistant species (Figure 1). Compared with a ciprofloxacin MIC less than 0.20 µg/mL, the unadjusted improvement rate ratio was 1.11 (95% CL, 0.79, 1.56) for those whose principal corneal isolate's ciprofloxacin MIC was 0.20 to 0.49 µg/mL, 1.00 (95% CL, 0.70, 1.45) for an MIC of 0.50 µg/mL, 1.30 (95% CL, 0.90, 1.88) for an MIC of 0.51 to 1.00 µg/mL, and 0.61 (95% CL, 0.35, 1.05) for an MIC greater than 1.00 µg/mL. A parsimonious improvement-rate model included ciprofloxacin MIC, age, hypopyon status, and corneal ulcer diameter, but the Cox proportional-hazards assumption was not met (P = .06) owing to a covariate-specific violation associated with corneal epithelial defect diameter (P = .004). Therefore, a model stratified by categorized corneal ulcer size was fit and found to satisfy the Cox proportional hazards assumption(P = .56). Ciprofloxacin-treated patients whose principal corneal isolate had a ciprofloxacin MIC exceeding 1.00 µg/mL improved significantly more slowly than those with a more susceptible isolate (Table 2).

Cure with ciprofloxacin monotherapy occurred in 272 (74.5%) of those with bacterial keratitis with a ciprofloxacin MIC less than 1.00 µg/m Land in 15 (57.7%) of those with a less sensitive isolate (Figure 2). In a multivariate model for the ordered events of improvement and cure, the proportional hazards assumption was not met (P = .01) owing to an effect by corneal epithelial defect diameter (P<.001). A stratified Cox model satisfying the proportional hazards assumption (P = .43) showed that the combined improvement and cure rate of ciprofloxacin-resistant bacterial keratitis was significantly slower than that of ciprofloxacin-sensitive keratitis (Table 2).

Twelve culture-positive patients were censored before achieving clinical improvement because of an adverse event (1 patient), noncompliance with ciprofloxacin dosing (2 patients), patient-initiated use of another antibacterial agent(4 patients), or voluntary withdrawal (5 patients). Eleven additional patients were censored after achieving clinical improvement but before achieving clinical cure because of noncompliance (1 patient), use of another antibacterial agent(5 patients), or withdrawal from the study (5 patients). Thirty patients were informatively censored because of worsening or lack of improvement and were switched to another antibacterial agent by the treating ophthalmologist. The bounds on how informative censoring could affect the improvement rate ratio associated with the ciprofloxacin MIC variable were 0.50 (95% CLs, 0.31, 0.80) and 0.74 (95% CLs, 0.46, 1.19), and corresponding bounds for the cure rate ratio were 0.66 (95% CLs, 0.47, 0.94) and 0.78 (95% CLs, 0.54, 1.12). The bounds on how early attrition could affect the improvement rate ratio were 0.47 (95% CLs, 0.29, 0.75) and 0.79 (95% CLs, 0.49, 1.27), and the bounds for the cure rate ratio were 0.72 (95% CLs, 0.51, 1.03) and 0.74 (95% CLs, 0.52, 1.05).

Antimicrobial resistance is a growing problem that requires ongoing efforts in surveillance, control, and prevention.^44,45 The use of antibiotics to which the responsible pathogen is not susceptible increases the complications and costs of systemic^46,47 and ocular⁴⁸ infections. Susceptibility assessment aims to predict the outcome of treatment with the antimicrobial agent tested, ⁴⁹ but few studies have critically evaluated how laboratory measurements relate to clinical effectiveness.^50,51 When antimicrobial levels at the site of infection are considered, the MIC of the infecting bacterial species affects the treatment outcome of some respiratory tract and soft tissue infections.⁵² Resistant infections have a higher incidence of therapeutic failure, ⁵³ but whether susceptibility testing can direct optimal antimicrobial selection remains unresolved.

Criteria that identify resistant ocular isolates need to be established. An experimental method for inferring resistance is to compare an antimicrobial agent's achievable corneal concentration with its in vitro MIC for common ocular pathogens.^26,54 A clinical approach is to determine how the rate of therapeutic recovery with an antimicrobial agent differs by MIC level of the same agent.⁵⁵ Despite long-standing uncertainty, ⁵⁶ interpretative standards for systemic infections are often applied to ocular isolates when assessing relative susceptibility. Although this study's susceptibility cutoff point derived from nonocular infection guidelines, improvement rates of bacterial keratitis were similar among lower MIC categories. This study supports a shared breakpoint for systemic and corneal infections when judging fluoroquinolone susceptibility.⁵⁷

The emergence of resistance is monitored by in vitro testing, but the importance of laboratory evaluation in the customary care of ocular infection is disputed.^4,58 Retrospective studies of bacterial keratitis have not shown a correlation between susceptibility and the effects of initial antibacterial therapy.^30,59-61 Interpreting these reports is limited by use of multiple antimicrobial agents and potential confounding by clinical severity. Relating antimicrobial susceptibility with outcome would help to establish evidence-based practice recommendations in ophthalmology.⁶²

Ciprofloxacin treatment failure of bacterial keratitis can occur when the corneal bacterial isolate has a high MIC to ciprofloxacin.^{13,14,20,24,25} A minority of our patients' principal corneal isolates, 7% of gram-positive cocci and 3% of gram-negative rods, had a ciprofloxacin MIC exceeding 1.0µg/mL. Ciprofloxacin-treated patients with corneal isolates having a higher ciprofloxacin MIC improved 43% more slowly and achieved combined improvement and cure 33% more slowly than those with more sensitive strains. The effect of ciprofloxacin resistance on reducing the rate of resolution with ciprofloxacin was most apparent in staphylococcal keratitis. In vitro resistance can portend a delayed response of bacterial keratitis, and the MIC may be useful in ophthalmic practice to indicate a less effective antibacterial agent. Further clinicomicrobiologic correlations are indicated for multiple bacterial isolates and different antibacterial agents.

This study was limited by informative censoring and early loss to follow-up that could overestimate the relative effect of ciprofloxacin susceptibility on response rates. Prospective data collection and performance of susceptibility testing at a reference laboratory that did not receive information on treatment effect reduced the possibility of differential misclassification, but measurement error of the ciprofloxacin MIC could have biased relative risk estimates. While the premise that ciprofloxacin-resistant bacterial keratitis responds more slowly to ciprofloxacin is biologically plausible, causality cannot be inferred from this observational study.

Whether in vitro antibacterial susceptibility predicts therapeutic response and is a cost-effective strategy are important issues. Current recommendations for antimicrobial therapy depend on clinical experience and recent susceptibility profiles of ocular isolates. A targeted approach of correct-spectrum coverage based on laboratory testing offers the prospect of guiding individualized management of bacterial keratitis.

Corresponding author and reprints: Kirk R. Wilhelmus, MD, PhD, Department of Ophthalmology, 6565 Fannin St, Houston, TX 77030 (e-mail: kirkw@bcm.tmc.edu).

Submitted for publication August 29, 2002; final revision received March 27, 2003; accepted April 10, 2003.

This study was supported by grant EY013782 from the National Eye Institute, Bethesda, Md (Dr Wilhemus), and a senior scientific investigator award from the Research to Prevent Blindness Inc, New York, NY (Dr Wilhemus).

We thank the investigators, listed in previous reports of this study group, ^9-11 who provided patient data through contractual agreements with Alcon Laboratories, Inc.