Streptococcus pneumoniae Bacteremia: Duration of Previous Antibiotic Use and Association with Penicillin Resistance

Transcription

1 MAJOR ARTICLE Streptococcus pneumoniae Bacteremia: Duration of Previous Antibiotic Use and Association with Penicillin Resistance Jörg J. Ruhe and Rodrigo Hasbun Department of Medicine, Infectious Diseases Section, Tulane University School of Medicine, New Orleans, Louisiana Previous antibiotic exposure is one of the most important predictors for acquisition of penicillin-nonsusceptible Streptococcus pneumoniae (PNSP) infection. To determine the impact of duration of exposure to different antibiotic classes, a study of 303 patients with S. pneumoniae was undertaken. Ninety-eight cases of (32%) were caused by a penicillin-nonsusceptible isolate. Bivariate analysis revealed that use of b-lactams, sulfonamides, and macrolides within the last 1 and 6 months before presentation was associated with PNSP ( P!.05). Fluoroquinolone consumption was not related to due to PNSP ( P 1.1). Both short- and long-term b-lactam use significantly increased the risk for PNSP infection. Logistic regression analysis revealed that use of b-lactams and macrolides in the 6 months before the first positive blood culture result were independent risk factors ( P!.05). Risk for acquiring PNSP infection depends on both the class of antibiotic to which the patient was exposed and the duration of therapy. An alarming worldwide increase in the rates of penicillin-nonsusceptible Streptococcus pneumoniae (PNSP) infection has been observed in the past decade [1, 2]. Previous antibiotic exposure has been found to be one of the most important risk factors for PNSP infection in several studies [3]. Associations between use of different antibiotics (b-lactams, macrolides, and sulfonamides) and PNSP carriage or infection have been demonstrated in case-control studies involving children and in cross-sectional studies [4 7]. Analytical studies of risk factors for infection with PNSP in adult populations primarily report on previous b-lactam use or have small numbers of resistant isolates [8 10]. The association of individual antibiotic classes and duration of treatment with PNSP infections has not been studied Received 10 September 2002; accepted 22 January 2003; electronically published 22 April Reprints or correspondence: Dr. Rodrigo Hasbun, Infectious Diseases Section, Tulane University School of Medicine, 1430 Tulane Ave., SL-87, New Orleans, LA (rhasbun@tulane.edu). Clinical Infectious Diseases 2003; 36: by the Infectious Diseases Society of America. All rights reserved /2003/ $15.00 in detail in this patient population [3]. Therefore, we performed this study to evaluate the impact of the length of treatment with all major antibiotic classes on the risk of acquiring PNSP in a mostly adult population. PATIENTS AND METHODS Setting. A retrospective cohort study of 303 pediatric and adult patients with S. pneumoniae was conducted at 4 tertiary care hospitals in New Orleans, Louisiana, from January 1996 through October Tulane University Hospital and Clinic and the New Orleans Veterans Affairs Medical Center mainly serve the insured and veteran populations, respectively. Charity Hospital and University Hospital of the Medical Center of Louisiana at New Orleans provide medical care to the urban indigent population. Study design and definition of terms. S. pneumoniae was defined as isolation of the organism from 1 set of blood cultures for patients with clinical symptoms consistent with infection or sepsis [11]. The zero time of the study was defined as the 1132 CID 2003:36 (1 May) Ruhe and Hasbun

2 date of the first positive blood culture result. All blood cultures positive within 14 days after zero time were counted as 1 bacteremic episode. Any subsequent episodes in the same patient were excluded. Nosocomial acquisition was defined as first occurring 148 h after admission. Treatment course was defined as the generally recommended number of antibiotic treatment days for a certain type of infection in a given population (7 14 days). An application of an antibiotic was considered to be short-term if the treatment duration was shorter than a treatment course (i.e.,!5 days for azithromycin or!7 days for other antibiotics). receiving long-term antibiotic prophylaxis during the study period were those who received 2 treatment courses, although the daily doses might have differed from those received by the treatment group. Blood cultures positive for S. pneumoniae were identified by review of the clinical microbiology records. We used a standardized extraction form to collect information on drug class and treatment duration for any antibiotic prescribed within the previous 1 and 6 months before zero time by review of all inpatient, emergency department, and outpatient medical records at the 4 hospitals. Furthermore, we analyzed all computerized inpatient and outpatient pharmacy records at all hospitals. Antibiotics were classified as penicillin or penicillin derivatives, cephalosporins, carbapenems, aztreonam, sulfonamides, macrolides, vancomycin, rifampin, aminoglycosides, clindamycin, fluoroquinolones, and tetracyclines. Antibiotics were also classified as b-lactams (penicillin or penicillin derivatives, cephalosporins, carbapenems, and aztreonam). Singledose applications were also considered significant and were recorded as 1 day of treatment. We excluded patients if any existing medical or pharmacy records for the 6 months before zero time were unavailable. The study was approved by the Tulane University Committee on Use of Human Subjects. Microbiological methods. S. pneumoniae was isolated, identified, and tested for penicillin sensitivity according to the guidelines of the NCCLS [12]. Isolates were screened for penicillin resistance by use of the oxacillin disk test and classified as penicillin susceptible (MIC,!0.12 mg/ml), intermediate-level penicillin nonsusceptible (MIC, mg/ml), or high-level penicillin nonsusceptible (MIC, 2 mg/ml) by use of the Etest methodology (AB Biodisk). Statistical analysis. The outcome variable was PNSP (MIC, 0.12 mg/ml). Bivariate analysis was conducted by Pearson s x 2 test or Fisher s exact test to identify antibiotic classes significantly associated with the outcome and to test for associations between consumption of different antibiotic classes ( P!.05). The Wilcoxon rank sum test was used to detect differences in patients median age and durations of antibiotic treatment. Multivariate analysis was performed with forward and backward stepwise logistic regression. All statistical analyses were conducted with SPSS for Windows software, version 11.0 (SPSS). RESULTS We identified a total of 447 episodes of S. pneumoniae at the 4 hospitals from January 1996 through October A total of 144 bacteremic episodes were excluded for the following reasons: the patient s medical records were incomplete ( n p 121), susceptibility of the isolate to penicillin was not determined ( n p 17), and the same patient had 11 episode of pneumococcal during the study period ( n p 6). A total of 303 patients with S. pneumoniae were included; 204 patients (67%) were enrolled from the Medical Center of Louisiana, 59 (19%) were enrolled from Tulane University Hospital and Clinic, and 40 (13%) were enrolled from the Veterans Affairs Medical Center. The median age of the total cohort was 40 years; 76 patients (25%) were!6 years old, and only 6 patients (2%) were aged 6 17 years. The pneumococcal isolate was penicillin susceptible (penicillin-susceptible S. pneumoniae [PSSP]) in 205 bacteremic episodes (68%), and it was penicillin nonsusceptible in 98 bacteremic episodes (32%). Sixty-five isolates (21%) were intermediate-level penicillin nonsusceptible, and 33 (11%) were high-level penicillin nonsusceptible. Of 121 excluded patients for whom penicillin susceptibility tests were performed, 80 (66%) were infected with a penicillin-susceptible isolate, and 41 (34%) were infected with a penicillin-nonsusceptible isolate ( P 1.2). Table 1 compares baseline characteristics between patients with PNSP and PSSP. No significant differences were found regarding age, sex, race, HIV infection/aids status, primary source of infection, and type of acquisition ( P 1.05). with PNSP had a higher Charlson comorbidity score than did patients with PSSP [13]. Previous antibiotic intake as risk factor for PNSP. Previous antibiotic use within the last 1 and 6 months before zero time is shown in table 2. Other antimicrobials prescribed within 6 months before zero time were as follows: vancomycin ( n p 6), aminoglycosides ( n p 5), tetracyclines ( n p 5), rifampin ( n p 3), and clindamycin ( n p 3). Separate analyses were not performed for these antibiotics because of the small sample size. A total of 155 patients (51%) received antibiotics during the 6 months before zero time: 118 (39%) received antibiotics for therapeutic indications as 1 treatment course, 17 (6%) received them as long-term prophylaxis only, and 20 (7%) received them for both treatment and prophylactic indications. Use of the following antibiotics was significantly associated with PNSP ( P!.05): penicillin or penicillin derivatives, any b-lactam, sulfonamides, and macrolides within the 1 and 6 months before zero time, and cephalosporins within S. pneumoniae Bacteremia CID 2003:36 (1 May) 1133

3 Table 1. Associations of baseline variables with penicillin-nonsusceptible Streptococcus pneumoniae (PNSP). Characteristic with PNSP with PSSP (n p 205) OR (95% CI) P Age, median years Age of!6 years 31 (32) 45 (22) 1.65 ( ).069 Male sex 62 (63) 129 (63) 1.01 ( ).954 African American race 73 (74) 148 (72) 1.12 ( ).674 Tobacco use 40 (41) 98 (48) 0.75 ( ).253 Injection drug abuse 11 (11) 20 (10) 1.17 ( ).693 HIV infection/aids 26 (27) 40 (20) 1.49 ( ).166 Charlson comorbidity score of (48) 73 (36) 1.67 ( ).040 Lungs as source of infection a 74 (76) 168 (82) 0.68 ( ).191 Nosocomially acquired infection 6 (6) 5 (2) 2.61 ( ).109 NOTE. Data are no. (%) of patients, unless otherwise indicated. PSSP, penicillin-susceptible S. pneumoniae. a Other sources: otitis ( n p 26), unknown ( n p 26), meningitis ( n p 4), skin ( n p 3), and septic arthritis ( n p 2). the 6 months before zero time. Fluoroquinolone exposure within the 1 and 6 months before zero time was not associated with PNSP ( P p.34 and P p.18, respectively). We conducted bivariate analyses to identify associations between the uses of different antibiotic classes within the previous 6 months (table 3). Use of any antibiotic of the b-lactam, sulfonamide, macrolide, or fluoroquinolone class was related to use of any drug of the other 3 classes ( P!.05). Subgroup analysis of all patients aged 6 years ( n p 227) revealed that risk factors were identical to those for the total cohort that is, use of b-lactams (OR, 6.45; 95% CI, ; P!.001), sulfonamides (OR, 2.60; 95% CI, ; P p.013), and macrolides (OR, 7.08; 95% CI, ; P!.001) during the 6 months before zero time. A Charlson comorbidity score of 11 (OR, 2.70; 95% CI, ; P p.001) and HIV infection/ AIDS (OR, 1.91; 95% CI, ; P p.037) were also associated with the outcome. Fluoroquinolone use was not related to the outcome ( P 1.05). Duration of previous antibiotic use. Table 4 shows the median duration of total treatment days within the 6 months before zero time with b-lactams, sulfonamides, and macrolides in patients with penicillin-susceptible S. pneumoniae and PNSP. with PNSP had significantly longer total exposure to b-lactams (15 vs. 10 days; P p.001). In contrast, total duration of therapy with sulfonamides or macrolides did not differ significantly between both groups ( P p.199 and P p.290, respectively). The risk for PNSP among patients with either 1 or 2 treatment courses compared with patients without previous antibiotic use is shown in table 5. Receipt of 1 or 2 courses of b-lactams or macrolides and receipt of 2 courses of sulfonamides were associated with PNSP (P!.001). The odds for 2 courses of b-lactams (OR, 34.36) were significantly higher than the odds for only 1 course (OR, 3.95; P!.001). Differences between the odds for 2 versus only 1 treatment course were of borderline significance or were nonsignificant for the sulfonamide and macrolide groups (P p.041 and P p.172, respectively). Ten patients (3%) received b-lactams (median duration of therapy, 2 days; range, 1 5 days) or macrolides (median duration of therapy, 2 days; range, 1 4 days) as short-term applications. These patients did not receive any additional antibiotics within the study period. Six patients received first- or second-generation cephalosporins as short-term applications. The indication in most patients was perioperative surgical prophylaxis. When compared with 148 patients (49%) who did not receive previous antibiotics, this small subgroup had a significantly higher likelihood of developing due to a penicillin-nonsusceptible isolate (OR, 5.73; 95% CI, ; P p.014). Regression analysis. We then performed logistic regression analysis to determine antibiotic classes that were independently related to penicillin nonsusceptibility. Factors associated with the outcome with P!.20 in unadjusted models were considered as candidates for the final model. Interaction terms, including b-lactams, sulfonamides, and macrolides, were also included in the model. Independent predictors were b-lactam use in the 6 months before zero time (adjusted OR, 5.61; 95% CI, ; P!.001) and macrolide use in the 6 months before zero time (adjusted OR, 2.83; 95% CI, ; P p.004). Logistic regression analysis of patients aged 6 years revealed identical independent risk factors for this subgroup that is, use of b CID 2003:36 (1 May) Ruhe and Hasbun

4 Table 2. Association of previous antibiotic use with penicillin-nonsusceptible Streptococcus pneumoniae (PNSP). No. (%) of patients Antibiotic use With PNSP With PSSP (n p 205) OR (95% CI) Any in the previous 6 months 76 (78) 79 (39) 5.51 ( )!.001 b-lactams Previous 1 month 26 (27) 20 (10) 3.34 ( )!.001 Previous 6 months 65 (66) 48 (23) 6.44 ( )!.001 Penicillin or penicillin derivatives Previous 1 month 22 (22) 12 (6) 4.66 ( )!.001 Previous 6 months 52 (53) 32 (16) 6.11 ( )!.001 Cephalosporins Previous 1 month 8 (8) 9 (4) 1.94 ( ).182 Previous 6 months 39 (40) 22 (11) 5.50 ( )!.001 Sulfonamides Previous 1 month 19 (19) 17 (8) 2.66 ( ).005 Previous 6 months 22 (22) 26 (13) 1.99 ( ).029 Macrolides Previous 1 month 14 (14) 10 (5) 3.25 ( ).005 Previous 6 months 28 (29) 18 (9) 4.16 ( )!.001 Fluoroquinolones Previous 1 month 5 (5) 6 (3) 1.78 ( ).343 Previous 6 months 13 (13) 17 (8) 1.69 ( ).175 NOTE. PSSP, penicillin-susceptible S. pneumoniae. lactams (adjusted OR, 5.16; 95% CI, ; P!.001) and macrolides (adjusted OR, 4.97; 95% CI, ; P!.001) during the 6 months before zero time. DISCUSSION Various factors, such as young age, previous hospitalization, HIV infection, and previous antibiotic use, have been reported to be associated with increased risk of PNSP colonization or infection [3]. Previous studies on the importance of previous antibiotic use were limited to b-lactams or to the pediatric population, or they were of cross-sectional design. To our knowledge, we have performed the first study involving a combined pediatric and adult cohort to have investigated the association between the duration of exposure to all major antibiotic classes and the subsequent occurrence of invasive PNSP infections. On bivariate analyses, use of penicillin and penicillin derivatives, cephalosporins, all b-lactams combined, macrolides, and sulfonamides in the 6 months before zero time were all highly significantly associated with PNSP (P.05). Separate analyses of antibiotic consumption within the last 1 month before zero time also showed a significant correlation between the outcome and the use of penicillin, b- lactams, macrolides, and sulfonamides ( P!.01). The associa- tion of previous antibiotic exposure within the last 1- and 6-month periods with PNSP infection could be explained by prolonged carriage of nonsusceptible pneumococci selected by previous antibiotic use [14]. In contrast, fluoroquinolone use in the last 1 and 6 months before zero time was not a significant predictor ( P p.34 and P p.18, respectively). These data suggest a potential benefit of responsible fluoroquinolone use in the further prevention of penicillin resistance in S. pneumoniae. However, recent reports describing increasing resistance to fluoroquinolones and subsequent associated treatment failures warrant their cautious use [15, 16]. On multivariate analyses, previous b-lactam use and previous macrolide use within the 6 months before zero time were both independently associated with the outcome of PNSP. These results were confirmed by separate analysis of patients aged 6 years. Unlike the findings of 2 large crosssectional studies [6, 7], in our study, b-lactam use (adjusted OR, 5.61) was more strongly correlated with penicillin resistance than was macrolide use (adjusted OR, 2.83). Previous use of sulfonamide, which was recently reported to be the only antibiotic associated with PNSP carriage in children in one Swedish study, was not an independent factor in our model [17]. P S. pneumoniae Bacteremia CID 2003:36 (1 May) 1135

5 Table 3. Associations of use of different antibiotic classes within the 6 months before diagnosis of pneumococcal. Antibiotic b-lactams (n p 113) Sulfonamides (n p 48) No. (%) of patients Macrolides (n p 46) Fluoroquinolones (n p 30) b-lactams (n p 113) 27 (56) a 30 (65) a 17 (57) b Sulfonamides (n p 48) 27 (24) a 24 (52) a 9 (30) b Macrolides (n p 46) 30 (27) a 24 (50) a 10 (33) a Fluoroquinolones (n p 30) 17 (15) b 9 (19) b 10 (22) b a P!.01. b.01 P!.05. We were interested in determining whether repeated exposure to an antibiotic correlated with a higher likelihood of PNSP. For this purpose, we divided patients into those who received one antibiotic course and those who received 2 courses over the 6-month period. who received 2 b- lactam treatment courses in the previous 6 months had a higher risk of subsequent PNSP compared with patients who received only 1 course ( P!.001). Differences in the number of courses for the sulfonamide and macrolide groups (1 vs. 2 courses) were of borderline significance or were nonsignificant ( P p.04 and P p.17, respectively). Similarly, the total median duration of previous treatment with b-lactams was significantly higher (15 days) in the penicillin-nonsusceptible group compared with the penicillin-susceptible group (10 days). No significant differences were found for the sulfonamide or macrolide group. These findings correlate with the findings of the study by Guillernot et al. [18], who examined the risk for carriage of PNSP in children. Longer duration of oral b- lactam treatment (15 days) in the preceding 30 days contributed to the likelihood of being a PNSP carrier (OR, 3.5; P p.02). Carsenti-Etesse et al. [19] studied the in vitro development of pneumococcal resistance to b-lactams after serial passage in media containing various subinhibitory concentrations of different b-lactams. The MIC breakpoint of mg/ml was reached after a mean of 18 daily amoxicillin passages, and the breakpoint of 2 mg/ml was reached in 5 isolates after a mean of 7.8 exposures to cefixime. These observations are consistent with stepwise acquisition of resistance through multiple penicillin-binding protein alterations and might provide an explanation for our clinical findings [20]. We also examined patients at the other end of the treatment spectrum: 10 patients with very short-term applications of b- lactams and/or macrolides (median, 2 days for both) also had a significantly higher risk of PNSP compared with patients without recent antibiotic intake (OR, 5.73; P p.014 ). To our knowledge, this is the first report to describe a correlation between very brief antibiotic exposure and penicillinnonsusceptible pneumococcal infections. Larger studies are certainly needed to validate these data. Our study has several advantages over previous work. To our knowledge, this is the first large study of both adult and pediatric patients to have evaluated the impact of use of all major antibiotics and duration of antibiotic exposure on the acquisition of PNSP. Furthermore, our heterogeneous study population makes our results more generalizable across age and socioeconomic status. Finally, we had a large number of outcomes in our study, which enabled us to perform valid multivariable analyses without overfitting our logistic regression model [21]. A disadvantage of our study is the retrospective design. It is possible that patients received antibiotic prescriptions by an outside provider without documentation in our hospitals medical records. However, meticulous abstraction of all existing medical and pharmacy records at all 4 hospitals was undertaken to minimize this problem. In addition, patients enrolled from 3 of the 4 medical centers (Charity and University Hospital of the Medical Center of Louisiana, Veterans Affairs Medical Cen- Table 4. Median duration of treatment with different antibiotics in the 6 months before diagnosis of pneumococcal. Treatment with PNSP with PSSP (n p 205) P b-lactams.001 No. (%) of patients 65 (66) 48 (23) Length of treatment, median days Sulfonamides.199 No. (%) of patients 22 (22) 26 (13) Length of treatment, median days Macrolides.290 No. (%) of patients 28 (29) 18 (9) Length of treatment, median days 10 7 NOTE. PNSP, penicillin-nonsusceptible Streptococcus pneumoniae; PSSP, penicillin-susceptible S. pneumoniae CID 2003:36 (1 May) Ruhe and Hasbun

6 Table 5. Number of treatment courses with different antibiotics in the 6 months before diagnosis of infection and association with penicillinnonsusceptible Streptococcus pneumoniae (PNSP). No. (%) of patients Antibiotic, no. of courses With PNSP With PSSP (n p 205) OR (95% CI) No antibiotic 22 (22) 126 (61) 1.00 b-lactams 1 Course 29 (30) 42 (20) 3.95 ( )! Courses 36 (37) 6 (3) ( )!.001 Sulfonamides 1 Course 4 (4) 12 (6) 1.91 ( ) Courses 18 (18) 14 (7) 7.36 ( )!.001 Macrolides 1 Course 20 (20) 15 (7) 7.64 ( )! Courses 9 (9) 2 (1) ( )!.001 P ter) receive most of their medical care from these institutions. These patients constituted 81% of the total study population. Potential use of free drug samples in these hospitals is recorded as a regular treatment course in the medical records and should not have significantly affected our data. However, differences regarding dose or duration of individual antibiotic treatment (e.g., as a result of management errors or different treatment indications) or noncompliance with the therapy were not accounted for in our analysis. Second, data on resistance to non b-lactams, especially macrolides and sulfonamides, were not available for the majority of the isolates. Thus, we cannot completely exclude the possibility that the demonstrated association of previous intake of these antibiotics with the outcome of penicillin resistance is at least partially the result of confounding. Pneumococcal isolates were not available for serotyping or molecular subtyping. Consequently, other epidemiological factors, such as horizontal spread in the community, could not be considered. Last, we selected patients with penicillin-susceptible pneumococcal as control group. This might have introduced selection bias into our study and may have led to an overestimation of the ORs for previous antibiotic use as risk factors for PNSP infection [22]. In summary, our study provides detailed information about previous antibiotic exposure in patients with PNSP. On multivariate analysis, previous b-lactam and macrolide use in the last 6 months before zero time were associated with PNSP, confirming recent findings from crosssectional and pediatric studies. Use of fluoroquinolones is not related to the development of PNSP. Multiple treatment courses and short-term applications of b-lactams and macrolides place patients at increased risk for developing PNSP. Judicious use of b-lactams and macrolides might reduce the incidence of PNSP infection. Acknowledgments We thank Megan O Brien for statistical advice and Anupama Menon for technical support. References 1. Whitney CG, Farley MM, Hadler J, et al. Increasing prevalence of multidrug-resistant Streptococcus pneumoniae in the United States. N Engl J Med 2000; 343: Reinert RR, Al-Lahham A, Lemperle M, et al. Emergence of macrolide and penicillin resistance among invasive pneumococcal isolates in Germany. J Antimicrob Chemother 2002; 49: Kristinsson KG. Effect of antimicrobial use and other risk factors on antimicrobial resistance in pneumococci. Microb Drug Resist 1997;3: Melander E, Ekdahl K, Joensson G, et al. Frequency of penicillinresistant pneumococci in children is correlated to community utilization of antibiotics. Pediatr Infect Dis J 2000; 19: Arason VA, Kristinsson KG, Sigurdsson JA, et al. Do antimicrobials increase the carriage of penicillin resistant pneumococci in children? Cross sectional prevalence study. BMJ 1996; 313: Garcia-Rey C, Aguilar L, Baquero F, et al. Importance of local variations in antibiotic consumption and geographical differences of erythromycin and penicillin resistance in Streptococcus pneumoniae. J Clin Microbiol 2002; 40: Pihlajamäki M, Kotilainen P, Kaurila T, et al. Macrolide-resistant Streptococcus pneumoniae and use of antimicrobial agents. Clin Infect Dis 2001; 33: Goldstein FW. Penicillin-resistant Streptococcus pneumoniae: selection by both b-lactam and non b-lactam antibiotics. J Antimicrob Chemother 1999; 44: Winston LG, Perlman JL, Rose DA, et al. Penicillin-nonsusceptible Streptococcus pneumoniae at San Francisco General Hospital. Clin Infect Dis 1999; 29: Meynard JL, Barbut F, Blum F, et al. Risk factors for isolation of S. pneumoniae Bacteremia CID 2003:36 (1 May) 1137

7 Streptococcus pneumoniae with decreased susceptibility to penicillin G from patients infected with human immunodeficiency virus. Clin Infect Dis 1996; 22: American College of Chest Physicians/Society of Critical Care Medicine Consensus Conference Committee. Definitions for sepsis and organ failure and guidelines for the use of innovative therapies in sepsis. Crit Care Med 1992; 20: National Committee for Clinical Laboratory Standards (NCCLS). Methods for dilution antimicrobial susceptibility tests for bacteria that grow aerobically: approved standard. 5th ed. Document M7-A5. Wayne, PA: NCCLS, Charlson ME, Pompei P, Ales KL, et al. A new method of classifying prognostic comorbidity in longitudinal studies: development and validation. J Chronic Dis 1987; 40: Austrian R. Some aspects of the pneumococcal carrier state. J Antimicrob Chemother 1986; 18: Chen DK, McGeer A, Azavedo JC, et al. Decreased susceptibility of Streptococcus pneumoniae to fluoroquinolones in Canada. N Engl J Med 1999; 341: Davidson R, Cavalcanti R, Brunton JL, et al. Resistance to levofloxacin and failure of treatment of pneumococcal pneumonia. N Engl J Med 2002; 346: Melander E, Mölstad S, Persson K, et al. Previous antibiotic consumption and other risk factors for carriage of penicillin-resistantstreptococcus pneumoniae in children. Eur J Clin Microbiol Infect Dis 1998; 17: Guillernot D, Carbon C, Balkau B, et al. Low dosage and long treatment duration of b-lactam. JAMA 1998; 279: Carsenti-Etesse H, Durant J, De Salvador F, et al. In-vitro development of resistance to b-lactam antibiotics in Streptococcus pneumoniae. J Antimicrob Chemother 1995; 36: Negri MC, Morosini MI, Loza E, et al. In vitro selective antibiotic concentrations of b-lactams for penicillin-resistant Streptococcus pneumoniae populations. Antimicrob Agents Chemother 1994; 38: Concato J, Feinstein AR, Holford TR. The risk of determining risk with multivariable models. Ann Intern Med 1993; 118: Harris AD, Karchmer TB, Carmeli Y, et al. Methodological principles of case-control studies that analyzed risk factors for antibiotic resistance: a systematic review. Clin Infect Dis 2001; 32: CID 2003:36 (1 May) Ruhe and Hasbun