Risk Factors for Fluoroquinolone Resistance in Nosocomial Escherichia coli and Klebsiella pneumoniae Infections FREE

THE INTRODUCTION of the fluorinated quinolones in the mid-1980s was heralded not only because of their high potency and broad spectrum of activity, but also because the potential for development of resistance to these agents was predicted to be very low, particularly among the Enterobacteriaceae.¹ In the 5 years after their initial introduction, there were few reports of resistance despite widespread use of these agents. Early descriptions of fluoroquinolone (FQ) resistance were largely limited to Staphylococcus aureus and Pseudomonas aeruginosa, organisms with intrinsically marginal susceptibility to these agents.²

Because of the intrinsically low minimum inhibitory concentrations (MICs) of the FQs to Escherichia coli and Klebsiella pneumoniae, it was believed that emergence of FQ resistance in these organisms was especially unlikely. In one study it was noted that after 5 years of increasing FQ use, the percentage of P aeruginosa resistant to these agents rose from 6% to 30%, while the percentage of FQ-resistant Enterobacteriaceae, including E coli and K pneumoniae, did not change.³ However, increasing reports of FQ resistance in E coli and K pneumoniae have appeared in recent years,^4- 5 although mostly limited to European centers with specific patient populations (eg, cancer patients).

While early reports of FQ resistance in E coli and K pneumoniae suggested a correlation between FQ resistance and increasing FQ use,^4- 5 there remains little consensus on which factors place patients at increased risk of FQ-resistant infection. While some studies have demonstrated an association between prior FQ use and FQ-resistant infection,^6- 11 others have not.^12- 13 Although there has been little consistency across studies, other risk factors purported to be associated with FQ resistance include immunosuppression,⁷ previous infection,⁷ older age,⁸ presence of a urinary catheter,⁸ aminoglycoside use,^6,12 receipt of another antibiotic,¹¹ and presence of a decubitus ulcer.¹¹

The emergence of FQ resistance in E coli and K pneumoniae is of great concern because these pathogens account for over 20% of all hospital-acquired infections and over 25% of all bloodstream infections.^14- 15 While most E coli and K pneumoniae isolates remain FQ susceptible, recent reports suggest the prevalence of FQ resistance is increasing. In a 1997 survey of bloodstream isolates, FQ resistance rates in E coli and K pneumoniae were from 3% to 4% in North America and from 3% to 8% in Europe^14- 15 compared with rates of less than 1% seen only several years before.¹⁶ While not yet as ominous as the 15% to 28% resistance rates seen for P aeruginosa,^14- 15 these findings suggest that if current trends continue, FQ resistance among even the most common of gram-negative pathogens will become increasingly problematic.

There are several important limitations in the studies that have to date examined risk factors for FQ resistance. First, some studies^7,11 included patients with infections due to Enterobacteriaceae (eg, E coli and K pneumoniae) and non-Enterobacteriaceae (eg, P aeruginosa and Acinetobacter species) despite the fact that the likelihood of developing resistance when exposed to an FQ may be very different in these groups.¹⁷ Second, other studies failed to perform multivariable analysis to account for other variables (eg, use of other antibiotics and species of infecting organism) that may function as confounders or effect modifiers in the association between FQ use and FQ resistance.^9,12 Third, many studies were conducted in Spain,^8- 10 a country in which availability of FQs and FQ use patterns may differ significantly from other countries.^8,18 Fourth, many studies were limited to certain patient populations (eg, patients with neutropenia and/or cancer)^9- 10,12 or to patients infected with certain types of organisms (eg, extended-spectrum β-lactamase [ESBL]–producing Enterobacteriaceae).⁶ Finally, studies have often not differentiated between isolates that reflect colonization vs infection,^8,11 although it has been suggested that risk factors for colonization and infection may be different.¹⁹ For example, patients with severe underlying diseases may have altered host defenses and are therefore at greater risk of infection. On the other hand, host defenses are of little importance in determining whether a patient becomes colonized by an organism that naturally inhabits the gastrointestinal tract. We undertook the present epidemiologic investigation of nosocomial FQ resistance to better understand the clinical forces driving the emergence of FQ resistance and to better identify successful strategies to curb the further dissemination of resistance and preserve the utility of these agents.

We conducted a hospital-based case-control investigation to identify risk factors for FQ resistance in infections due to E coli and K pneumoniae. The present study was reviewed and approved by the Committee on Studies Involving Human Beings of the University of Pennsylvania School of Medicine, Philadelphia. The study was conducted at 2 hospitals within the University of Pennsylvania Health System (UPHS): the Hospital of the University of Pennsylvania (HUP), an academic tertiary care medical center with 725 patient beds, and Presbyterian Medical Center (PMC), a 324-bed urban community hospital. The investigation was conducted over a 1.5-year period (January 1, 1998, through June 30, 1999).

All cultures demonstrating an E coli and/or K pneumoniae isolate were identified through records of the clinical microbiology laboratory at HUP, which processes and cultures all specimens obtained at the participating hospitals. Polymicrobial cultures (ie, cultures demonstrating growth of ≥1 organism(s) in addition to E coli and/or K pneumoniae) were also considered eligible for inclusion. Designation as a potential case or control was based solely on FQ susceptibility testing. Identification and susceptibility testing of E coli and K pneumoniae was performed using standard techniques. This was performed using either a semiautomated system (MicroScan WalkAway System, NC16 panel, Dade Behring, Inc, Deerfield, Ill) or disk diffusion susceptibility testing. Results were interpreted as described by the National Committee for Clinical Laboratory Standards.²⁰ Resistance to levofloxacin was used as a marker for resistance to the FQ antibiotics. An isolate was considered resistant if it demonstrated a minimum inhibitory concentration (MIC) of 8 µg/mL or greater to levofloxacin.

Escherichia coli and K pneumoniae isolates with a ceftazidime and/or aztreonam MIC greater than 2 µg/mL were suspected of producing an ESBL or AmpC-type β-lactamase. Such isolates were subjected to the double disk diffusion test as described by Thomson and Sanders²¹ but with the ceftazidime and amoxicillin-clavulanic acid disks placed 15 mm apart (edge to edge). Any distortion of the zone around the ceftazidime disk toward the amoxicillin-clavulanic acid disk was considered to be positive for ESBL production. AmpC-type β-lactamase production was suspected in isolates that had an elevated ceftazidime MIC (>2 µg/mL) and were resistant to both amoxicillin-clavulanic acid and cefoxitin. Isolates demonstrating an AmpC-type β-lactamase phenotype were also classified as ESBLs. Of note, approximately 10% of ESBLs at our institution are AmpC-like (P.H.E., oral communication, October 5, 2001).

All E coli and K pneumoniae isolates demonstrating FQ resistance (ie, potential cases), identified from the microbiology laboratory database, were considered eligible for inclusion. All FQ-susceptible E coli and K pneumoniae isolates (ie, potential controls) from the same period were identified from the same database. Of all potential controls, a number equal to the total number of potential cases was randomly selected. Following this selection of eligible cases and controls, the medical records of these patients were reviewed to determine whether each patient met inclusion criteria for the study. Of the eligible cases and controls, only those patients whose isolates represented true infection, defined by criteria established by the Centers for Disease Control and Prevention,²² were included. Furthermore, eligible cases and controls were included in the study only if their isolates represented nosocomial acquisition. Infection was considered nosocomial if one or more of the following was true: (1) infection occurred more than 48 hours after hospital admission; (2) the patient was admitted from another medical center or long-term care facility (LTCF), having spent more than 48 hours in the other facility; and/or (3) the patient had been hospitalized within the past 2 weeks. Matching cases and controls for infecting pathogen or site of infection was not undertaken because preliminary data suggested resistance rates were similar across these covariates and would permit appropriate adjustment for these factors in the final analysis.

Each patient was included as a case or control only once. If E coli or K pneumoniae was isolated on multiple occasions, only the first episode of infection was reviewed. For patients with an infection occurring during the first month of the study, the medical record was reviewed to ensure that another infection with the same organism did not occur in the preceding month. If such an infection had occurred in the preceding month, the patient was excluded.

Data regarding potential risk factors for FQ resistance were ascertained through review of inpatient medical records. Data obtained included age, sex, race, hospital location, number of hospital days prior to infection, number of days in an intensive care unit in the 30 days prior to infection, and severity of illness calculated by the APACHE (Acute Physiology and Chronic Health Evaluation II score.²³ We ascertained the presence of an indwelling device (ie, central venous catheter, urinary catheter, or mechanical ventilation) and diarrhea, as well as whether infection with Clostridium difficile was diagnosed. Finally, the species of infecting organism, site of infection, presence of bloodstream involvement, and the presence of a coinfecting organism were documented.

The presence of each of the following comorbid conditions was assessed: hepatic dysfunction, malignancy, diabetes mellitus, renal insufficiency, human immunodeficiency virus infection, neutropenia, corticosteroid use, prior transplantation, use of an immunosuppressive agent in the preceding 30 days, and surgical procedure or trauma in the preceding 30 days. Hepatic dysfunction was defined by 2 or more of the following at the time of infection: a bilirubin concentration greater than 2.5 mg/dL (42.75 µmol/L); an aspartate aminotransferase or alanine aminotransferase level more than twice the upper limit of normal; or known liver disease. Neutropenia was classified as an absolute neutrophil count less than 500/µL at the time of infection. Renal insufficiency was indicated by a creatinine level greater than 2.0 mg/dL (34.2 µmol/L). Corticosteroid use was defined as the receipt of prednisone at a dosage of 20 mg/d (or equivalent) for at least 2 weeks within the preceding 30 days.

All inpatient antimicrobial therapy in the preceding 30 days was documented through review of the in-hospital medical record. If another UPHS hospitalization occurred within the preceding 30 days, the inpatient medical record from that hospitalization was reviewed. If a patient was admitted from a non-UPHS medical center or LTCF, antibiotic use in that setting was investigated through review of documents obtained at the time of transfer. Of patients admitted from an outside hospital or LTCF, approximately 80% had records detailing recent medication use forwarded at the time of admission. Finally, any outpatient antibiotic use in the 30 days prior to infection was documented if it was recorded in the hospital record.

The specific antibacterial and the antibacterial class prescribed were documented. The total number of antibiotics used and the total number of antibiotic days (ie, the sum of days exposed to each antibiotic during the study period) were calculated. In assessing prior FQ use specifically, we examined the type of FQ received, total duration of FQ use, and the number of separate courses of FQ therapy (defined by a gap of >48 hours between appropriately scheduled doses). Finally, we documented the concurrent use of antacids, sucralfate, magnesium, calcium, iron, and zinc, since these medications may interfere with oral absorption of FQs and result in subinhibitory levels of these agents.⁸

Cases and controls were first characterized by all of the potential risk factors of interest. Bivariable analysis was subsequently conducted to determine the association between potential risk factors and FQ-resistant infection. Categorical variables were compared using either χ² analysis or the Fisher exact test when appropriate. An odds ratio (OR) and 95% confidence interval (CI) was calculated to evaluate both the strength of any association as well as the precision of the estimate of the effect. Continuous variables were compared using the Wilcoxon rank sum test.²⁴

Stratified analyses were then conducted to help identify where data were sparse and to elucidate where confounding and interactions were likely to exist in the multivariable analysis of the relationship between antimicrobial drug exposure and drug-resistant infections. We were primarily interested in investigating the effect of stratification by several variables on the association between prior FQ use and FQ-resistant infection. Specifically, we examined the effect of stratifying by the following variables: species of infecting organism, hospital, duration of hospitalization (divided into quartiles), LTCF residence, and calendar date of infection. The variable "calendar date of infection" was described in 2 ways: (1) divided into tertiles based on the distribution of this variable in the study population and (2) divided into 3 specific calendar periods (ie, January-June 1998, July-December 1998, and January-June 1999). The Mantel-Haenszel test for summary statistics was used to evaluate possible confounding,²⁵ while interaction was assumed to be present when the test for heterogeneity between the OR for different strata was significant (P<.05).

Adjusted ORs were calculated using multiple logistic regression analysis.²⁶ Building of the multivariable model began with the inclusion of the variable for prior FQ use, based on our a priori hypothesis of an association between FQ use and FQ-resistant infection. Variables were considered for inclusion in a multivariable model if they were found to be associated with FQ-resistant infection on bivariable analysis (P≤.15). In addition, risk factors were considered for inclusion in the model if they were involved in confounding or interaction on stratified analysis. These variables remained in the final multivariable model if their inclusion in the model resulted in a 15% or greater change in the effect size of the primary association of interest (ie, the association between FQ use and FQ-resistant infection). Potential interactions between risk factor variables in the final multivariable model were also investigated.

It has been suggested that certain characteristics of FQ therapy (eg, multiple courses and monotherapy) may increase the likelihood of a subsequent FQ-resistant infection.²⁷ In a separate subanalysis, we identified all patients who received an FQ antibiotic during the preceding 30 days. We investigated whether certain characteristics of FQ use (eg, agent, duration, number of courses, and combination therapy) were associated with subsequent FQ-resistant infection. We also investigated other factors (ie, demographics and comorbidities) as potential confounders of interest in the association of characteristics of prior FQ use and FQ-resistant infection.

In our sample size calculations, we wanted to be able to identify an effect size for the association between FQ use and FQ-resistant infection as small as an OR of 2.0. Given our projected sample size, we would have had 90% power to detect such a difference, with α = .05. For all calculations, a 2-tailed P value of less than .05 was considered significant. All statistical calculations were performed using standard programs in STATA v6.0 (Stata Corp, College Station, Tex).

During the study period, 178 potential cases (FQ-resistant isolates) were identified from the database of the clinical microbiology laboratory. Of 3278 potential controls (FQ-susceptible isolates), 178 were randomly selected to equal the number of potential cases. The medical records of these patients were then reviewed to determine if patients met eligibility criteria previously described. Of the 178 potential cases, 42 were ineligible because their isolates represented colonization and/or community acquisition. Of 178 potential controls, 96 were ineligible for the same reasons. Of 136 eligible cases, 123 (90.4%) had complete medical records available for review. Of 82 eligible controls, 70 (85.3%) had complete records available.

The median age of all 193 patients was 72 years (95% CI, 69-75 years; range, 22-100 years). One hundred twenty-one (62.7%) patients were women. Of 183 patients for whom race was noted, 67 (36.6%) were white, 109 (59.6%) were African American, 5 (2.7%) were Latin American, and 2 (1.1%) were Asian American. One hundred twenty-five (64.8%) patients were hospitalized at HUP compared with 68 (35.2%) at PMC.

Of 193 patients, 149 (77.2%) had infection with E coli, while 44 (22.8%) had K pneumoniae infection. The location of infection was as follows: urinary, 131 (67.9%); bloodstream, 25 (13.0%); wound, 19 (9.8%); respiratory, 15 (7.8%); central venous catheter, 2 (1.0%), and abdominal, 1 (0.5%). When comparing patients with infection due to an FQ-susceptible organism (controls) with those with infection due to an FQ-resistant organism (cases), cases were older, more likely to have been African American, admitted from a LTCF, and hospitalized at PMC (Table 1). There were borderline associations between FQ-resistant infection and APACHE II score, male sex, and diarrhea (Table 1). Control patients were more likely to have been admitted from an outside hospital, while there was a borderline significant association between hospitalization in the past 30 days and infection due to an FQ-susceptible organism (Table 1). Finally, patients hospitalized on medicine floors were significantly more likely to have FQ-resistant infection, while location in the cardiac intensive care unit was significantly associated with FQ-susceptible infection.

In comparing the comorbidities of the cases and controls, there was only a borderline association between malignancy and FQ resistance, with malignant disease being more common in patients with FQ-susceptible infection (Table 2). There was no association between the infecting pathogen and FQ resistance (Table 2). While a primary bloodstream infection was more common in FQ-susceptible infections, there was no association between bloodstream involvement, in general, and FQ resistance (Table 2).

Fluoroquinolone-resistant isolates were significantly more likely to demonstrate resistance to other antibiotics (Figure 1). With the exception of nitrofurantoin and imipenem, a significantly greater percentage of FQ-resistant isolates were resistant to other antibiotics for which at least half of all isolates were tested. Of note, 31 (25.2%) cases and 3 (4.3%) controls were infected with isolates demonstrating ESBL production.

Patients with FQ-resistant infections had significantly greater antimicrobial exposure in the 30 days prior to infection than patients with FQ-susceptible infection (Table 3). This was manifested both in greater number of antibiotics as well as greater median antibiotic days. Cases were also significantly more likely to have been exposed to an FQ or an aminoglycoside.

On stratified analyses, the following variables were noted to be confounders in the association between prior FQ use and FQ-resistant infection: sex, LTCF residence, admitting hospital, and infecting pathogen. Duration of hospitalization prior to infection and calendar year of infection (categorized as described in the "Methods" section) were not confounders in this association. None of these variables were noted to be involved in effect modification.

On multivariable analysis, the following variables were noted to be independent risk factors for FQ resistant infection: (1) prior FQ use; (2) LTCF residence; (3) prior aminoglycoside use; and (4) older age (Table 4). Sex, admitting hospital, and infecting organism remained in the final multivariable model but were not independently associated with FQ-resistant infection.

Because the accuracy and comprehensiveness of data regarding outpatient antibiotic use may have been less than that of inpatient use (ie, in-hospital, LTCF, and outside hospital), we repeated the above analyses excluding all data regarding outpatient antibiotic use. This reanalysis did not result in a substantive change in the final adjusted OR for the association between FQ use and FQ-resistant infection. In addition, the final variables included in the multivariable model were unchanged.

In our primary analysis, we combined E coli and K pneumoniae isolates, given their epidemiological and microbiological similarities. Although the infecting organism (eg, E coli or K pneumoniae) was not significantly associated with FQ resistance in our final multivariable model, we further explored the impact of the infecting organism on the epidemiologic factors of FQ resistance by repeating the above analyses, limiting the data set to only those patients with E coli infections. This final multivariable model did not differ substantially from our primary model except that age was no longer significantly associated with FQ-resistant infection.

To explore whether certain characteristics of FQ use might be associated with FQ-resistant infection, we subsequently conducted a subanalysis on all patients who received an FQ during the 30 days prior to infection. There were 41 patients who received an FQ during this time. Thirty-nine (95.1%) received levofloxacin, while 2 (4.9%) received ciprofloxacin. Of the 41 patients, 35 (85.4%) had an FQ-resistant infection. In comparing demographic variables and comorbidities of cases and controls within this subgroup, there were no significant differences. Of the 35 cases, 7 (20.0%) had been exposed to an aminoglycoside compared with 0 of the 6 controls (P = .56, Fisher exact test). Fifteen (42.9%) cases resided in a LTCF compared with 2 (33.3%) controls (P>.99, Fisher exact test). The median number of antibiotics to which cases and controls had been exposed in the preceding month was 4.5 and 3.0, respectively (P = .14, Wilcoxon rank sum test).

Of the 35 cases, 3 (8.6%) had received multiple courses of FQ in the prior month, compared with 1 (16.7%) of the controls (P = .48, Fisher exact text). The median days of prior FQ use in the cases and controls was 5 and 4.5 days, respectively (P = .54, Wilcoxon rank sum test). Of the 35 cases, 14 (40.0%) had received an agent that could have interfered with oral FQ absorption compared with 4 (66.7%) of the controls (P = .22, Fisher exact test).

The median time from start of prior FQ therapy to infection in cases and controls was 11 and 10 days, respectively (P = .52, Wilcoxon rank sum test). The median time from completion of prior FQ therapy to infection in cases and controls was 8.5 and 4.0 days, respectively (P = .27, Wilcoxon rank sum test).

Of the 35 cases, 20 (57.1%) had received another antibacterial in combination with an FQ compared with 4 (66.7%) of 6 controls (P>.99, Fisher exact test). When comparing concurrent use of an antibacterial active against gram-negative pathogens (ie, cephalosporins, penicillins, aminoglycosides, carbapenems, and sulfamethoxazole-trimethoprim), 9 (25.7%) cases had received such combination therapy compared with 3 (50.0%) controls (P = .33, Fisher exact test).

This study investigated the epidemiology of FQ resistance in the two most common gram-negative pathogens E coli and K pneumoniae, which together cause up to one quarter of all nosocomial infections.^14- 15 We found the following variables to be independent risk factors for FQ resistance: (1) recent FQ use; (2) LTCF residence; (3) recent aminoglycoside use; and (4) older age.

The high potency, broad spectrum of activity, and relative tolerability of the FQs have led to widespread use and misuse of these agents in a variety of settings.²⁸ Less than a decade after their release in the United States, FQs were reported to be the second most commonly used antibiotic class in hospitals²⁹ and the most common antibiotic class used in LTCFs.³⁰ As antimicrobial resistance to other traditional first-line agents (eg, sulfamethoxazole-trimethoprim for urinary tract infections) increases, the reliance on FQs as empiric and definitive treatment is likely to increase further.³¹

Resistance to the FQ antibiotics occurs mostly through chromosomal mutation.³² This occurs either by alteration of the two FQ target enzymes DNA gyrase and topoisomerase IV or through modification of those enzymes that affect drug permeation by altering expression of porin-diffusion channels across the gram-negative outer membrane and/or altering expression of endogenous efflux pumps located in the inner membrane.³²

Our demonstration of a significant association between FQ use and FQ-resistant infection confirms in vitro studies demonstrating the selection of high-level FQ resistance by serial passage of organisms through increasing concentrations of drug.³³ Our results confirm the findings of some smaller studies.^{6- 7,10- 11} While our main emphasis was to elucidate the relationship between any FQ exposure and subsequent FQ resistance, we also explored whether certain characteristics of FQ use might be associated with FQ resistance. In our subanalysis of all patients who received an FQ, we noted no significant difference between the cases and controls with regard to the number of FQ courses, duration of FQ treatment, or exposure to other antibiotics. Interestingly, 50.0% of controls, but only 25.7% of cases, had received an agent with gram-negative activity in combination with an FQ prior to development of an FQ-resistant infection. While this subanalysis was limited by the small sample size, these preliminary findings should be explored further in future, larger studies.

What are the implications of the association between FQ exposure and subsequent FQ-resistant infection? While FQs have great utility in a variety of clinical settings, it has been suggested that use of these agents can be improved significantly. It has recently been noted that in only 29% of in-hospital cases in which an FQ agent was used was it considered a first-line agent by accepted guidelines.³⁴ Such findings suggest that use of FQs should be closely monitored to ensure appropriateness of use, particularly when alternative agents are available. Given the widespread use of FQs in the community, emergency departments, and in LTCFs, detailed analysis of their use in these settings should also be conducted. Such information is vital if efforts directed at improving the use of FQs are to be successful.

We also found that prior use of an aminoglycoside was significantly associated with FQ-resistant infection. In vitro studies have shown that exposure of E coli to other antibiotics (eg, tetracycline or chloramphenicol) can, through outer membrane protein changes, select mutants with low-level FQ resistance.³⁵ It has also been suggested that exposure to antibiotics such as sulfamethoxazole-trimethoprim and aminoglycosides may cause alterations in membrane permeability, resulting in concurrent FQ resistance.³⁶ We have recently demonstrated that, in infections due to ESBL-producing Enterobacteriaceae, prior aminoglycoside use was an independent risk factor for FQ resistance in these organisms.⁶ Finally, a recent study found that prior aminoglycoside exposure was significantly associated with FQ-resistant P aeruginosa bloodstream infection.¹² If future studies confirm the role of aminoglycoside use as a risk factor for FQ resistance, one might include efforts to limit the overuse of aminoglycosides in an overall strategy to reduce FQ resistance.

There are several potential explanations for the association between LTCF residence and FQ-resistant infection. While patients in LTCFs have greater exposure to antibiotics than the general population,³⁷ LTCF residence remained an independent risk factor for FQ resistance in our analysis, even after controlling for prior FQ and aminoglycoside use. It is possible that the window of antibiotic exposure chosen for this study (ie, 30 days prior to infection) was too narrow and failed to account for more remote antibiotic use that may have occurred in the LTCF setting. In addition, documentation of antibiotic use in the LTCF may not have been accurate.

Another possible explanation for the association with LTCF residence is that patients residing in such facilities have greater exposure to other infected or colonized patients and health care workers because such centers often have limited infection control programs.³⁸ Persons in LTCFs may acquire resistant infections through person-to-person spread rather than through exposure to particular antimicrobial agents. Future studies should elucidate the role of horizontal spread of FQ resistance in such settings as well as whether interventions directed at improved barrier precautions would be effective in reducing FQ resistance in these facilities.

The influence of LTCF residence on the epidemiologic factors of nosocomial FQ resistance reinforces the importance of close collaboration between various types of health care facilities in working to coordinate efforts to address the spread of antimicrobial resistance. These results suggest that the success of any intervention designed to curb resistance will be limited if efforts focus on single health care institutions and related health care facilities are not considered in its implementation.

We also noted a significant association between older age and FQ-resistant infection. One explanation for this association, which remained significant even after controlling for LTCF residence, may be the immune dysregulation that has been demonstrated to occur with advancing age.³⁹ A recent study of E coli urinary tract infections also noted that an age older than 65 years is a risk factor for FQ resistance.⁸

There were several potential limitations to our study. Although the possibility of selection bias is normally of concern in a case-control study, all cases and controls were readily identified through the same microbiology laboratory that processes all bacterial cultures for the participating institutions. Of patients with E coli and K pneumoniae isolates, all patients whose isolates were FQ resistant were eligible for inclusion, while a random sample of patients with FQ-susceptible isolates was selected as eligible controls. On subsequent review of medical records to determine whether predetermined eligibility criteria were met, a greater number of controls were excluded when compared with cases. This discrepancy is, however, unlikely to have introduced a differential bias since, based on the microbiology database, all potential cases and a random sample of all potential controls were selected. Thus, the potential of selection bias was likely small except for that introduced through lost medical records, which did not differ substantially between cases and controls.

Although misclassification bias is likewise of concern in case-control studies, cases and controls were drawn from the same hospitalized patient population and were identified based solely on antimicrobial susceptibility data. Because these tests were conducted without prior knowledge of the patient's status with regard to possible exposures of interest, there is unlikely to be any differential misclassification bias.

Although retrospective medical record review studies may be limited by their availability, we found that 85% of the records were complete and available for review. Data were, however, limited to that contained in the hospital record. Although information concerning in-hospital antibiotic use was available from the medical record, the possibility exists for inaccuracies in data regarding antibiotic use at an outside hospital or LTCF. However, of patients admitted directly from an outside hospital or LTCF, approximately 80% had records detailing recent medication use forwarded at the time of admission.

It is also possible that data regarding recent outpatient antibiotic use may not be reported or documented at the time of admission and thus were not available for abstraction. Although this potential deficiency is of concern, it would likely result in a nondifferential bias given that information regarding antibiotic use would be obtained at the time of admission, prior to the physician knowing whether the patient subsequently develops an FQ-susceptible or FQ-resistant infection. Alternatively, it is possible that in patients who develop an FQ-resistant infection, additional subsequent interviewing of the patient might result in a more thorough assessment of recent outpatient antibiotic use. Although we noted no substantive difference in our final results whether or not documented outpatient antibiotic use was included, future studies should seek to more clearly evaluate the possible contribution of recent outpatient antibiotic use to nosocomial infection with FQ-resistant infections.

If one assumes that data regarding in-hospital antibiotic use are more accurate and complete when a study relies solely on information in the hospital medical record, one might suggest that the association between FQ use and FQ-resistant infection is simply a reflection of duration of hospitalization. However, duration of hospitalization was not related to probability of an FQ-resistant infection. We also limited our analysis of antibiotic use to 30 days prior to infection and were thus unable to assess possible associations between FQ-resistant infection and more remote antibiotic use.

Another potential limitation was the unavailability of isolates to permit molecular epidemiologic analysis. As such, we were unable to determine whether there existed any horizontal transmission of FQ-resistant organisms. Although past studies have suggested that horizontal spread plays a very limited role in the emergence of FQ resistance,⁴ the higher prevalence of FQ resistance in ESBL-producing organisms suggests that future studies should examine the possible impact of ESBL-mediated resistance on dissemination of FQ resistance. Finally, our study was conducted in a large academic tertiary care medical center and a smaller urban community hospital, thus the results may not be generalizable to other institutions.

We investigated the epidemiologic factors of FQ resistance in E coli and K pneumoniae infections and found the following independent risk factors for FQ resistance: (1) prior FQ use; (2) LTCF residence; (3) prior aminoglycoside use; and (4) older age. In addition to elucidating the causes of the emergence of FQ resistance, our results suggest that strategies designed to curb the rise in FQ resistance should focus on limiting the use of FQ and potentially aminoglycoside antibacterial agents. Preserving the usefulness of FQs in the treatment of E coli and K pneumoniae infections will allow these agents to remain viable therapeutic options, particularly as resistance to other antibiotics also grows. Future studies should seek to prospectively validate these findings, perhaps with a particular focus on patients residing in LTCFs, determine the impact of FQ resistance on clinical and economic outcomes, and evaluate the impact of interventions designed to decrease FQ resistance.

Accepted for publication February 13, 2002.

This work was financially supported by a postdoctoral award from the Infectious Diseases Society of America (Alexandria, Va)/Roche Laboratories (Nutley, NJ) (Dr Lautenbach).

This study was presented in part at the 38th Annual Meeting of the Infectious Diseases Society of America, New Orleans, La, September 7-10, 2000.

Corresponding author and reprints: Ebbing Lautenbach, MD, MPH, University of Pennsylvania School of Medicine, Center for Clinical Epidemiology and Biostatistics, 825 Blockley Hall, 423 Guardian Dr, Philadelphia, PA 19104-6021 (e-mail: elautenb@mail.med.upenn.edu).