Abstract

To evaluate the potential bias of analyzing aggregated data, we separately examined antibiotic exposure and resistance data for 35,423 patients admitted to a university hospital in Utah, from both an individual-patient perspective and group-level perspective. From 1994 through 1998, use of defined daily doses (per 1000 patient-days) of fluoroquinolones, third-generation cephalosporins, ampicillin-sulbactam, and imipenem increased by 82%, 38%, and 99%, and decreased by 38%, respectively, whereas group-level resistance rates of Enterobacteriaceae or Pseudomonas species changed only minimally. However, in individual-patient—level analyses performed by multivariable proportional hazards regression, exposure to a fluoroquinolone, third-generation cephalosporin, ampicillin-sulbactam, or imipenem was a strong risk factor for resistance to fluoroquinolones (adjusted hazard ratio [AHR], 4.0; P < .001), third-generation cephalosporins (AHR, 3.5; P < .001), ampicillin-sulbactam (AHR, 2.3; P = .008), or imipenem (AHR, 5.7; P < .001), respectively. Thus, group-level and individual-patient—level analyses of antibiotic-use-versus-susceptibility relations yielded divergent results. Multicenter studies should include individual-patient—level data to elucidate more fully the relation between antibiotic exposure and resistance.

The importance of accumulating and comparing data regarding antimicrobial use and resistance from different institutions is increasingly recognized [1, 2]. However, these data sets often lack individual-patient—level information on antibiotic exposure, which has forced investigators to rely on group-level data to assess the relationship between antibiotic use and resistance. As a potential methodological shortcoming, the analysis of aggregated data may be limited by “ecologic bias,” which is the failure of group-level—effect estimates to reflect the biological effect at the individual-patient level. This bias is a result of the fact that, unlike individual-level studies, ecologic studies do not link individual outcome events to individual exposure histories [3, 4].

To evaluate the hazard of conducting group-level analyses and to determine the magnitude of this potential ecologic bias, we separately analyzed, from both an individual-patient—level and group-level perspective, the antibiotic-use and microbiological-resistance data from a large cohort of hospitalized patients at a single institution. Specifically, by means of 2 different epidemiological approaches, we assessed whether the nosocomial isolation of Enterobacteriaceae or Pseudomonas species that are resistant to fluoroquinolones, third-generation cephalosporins, ampicillin-sulbactam, or imipenem was associated with in-hospital exposure to these classes of antimicrobial agents.

Methods

The study cohort consisted of patients who had been admitted to the medical and surgical services at a 400-bed tertiary care hospital in Salt Lake City from January 1994 through December 1998. Other inclusion criteria were an age of >15 years and length of hospital stay of >2 days.

Individual-patient data on age, sex, International Classification of Disease—9 codes, procedures, medications, and microbiological findings were available. Data were collected retrospectively from administrative, pharmacy, and laboratory computerized databases with use of a relational database-management system [5].

Use of 4 types of antimicrobial agents (third-generation cephalosporins, fluoroquinolones, ampicillin-sulbactam, and imipenem) formed the primary exposure of interest. For the group-level analyses, antibiotic use was expressed as defined daily dose (DDD) per 1000 patient-days (1 DDD being the standard adult daily dose of an antibiotic agent for 1 day's treatment) [6, 7]. For the individual-patient—level analyses, the main exposures of interest were measured in days of antibiotic exposure [8]. To adjust for confounding by other antibiotics and to facilitate time-dependent analysis, treatment with other antibiotics was measured in average number of antibiotic doses per hospital day.

For both approaches, the main outcome of interest was the isolation of a clinical specimen of Enterobacteriaceae that was resistant to one of the studied antimicrobial classes >2 days after admission or the isolation of Pseudomonas species resistant to ceftazidime, imipenem, or ciprofloxacin >2 days after admission (i.e., nosocomial isolation). In correspondence with clinical practice, intermediate susceptibility to the selected antibiotics was considered to be resistance. Standard media and techniques for isolation, identification, and antibiotic susceptibility testing were used.

Initially, group-level data were analyzed graphically. Temporal trends were assessed by plotting DDDs per 1000 patient-days for the specific antibiotic class of interest against susceptibility percentages of unique nosocomial isolates, according to time and space (e.g., year and ward). Thereafter, changes in resistance patterns over time were analyzed using linear regression with the help of the general linear model procedure.

For individual-patient—level analyses, multivariable Cox proportional hazards regression was used in separate models to determine the relative risk of antibiotic resistance while controlling for length of stay. For all case patients, the observation continued from time of admission to the recovery of fluoroquinolone-resistant Enterobacteriaceae or Pseudomonas species (first model), recovery of third-generation cephalosporin-resistant Enterobacteriaceae or Pseudomonas species (second model), recovery of ampicillin-sulbactam—resistant Enterobacteriaceae (third model), or recovery of imipenem-resistant Enterobacteriaceae or Pseudomonas species (fourth model). Patients who did not reach an outcome status were censored at time of discharge or death. Patients with resistant microorganisms detected before the present hospitalization or within 48 h after admission were excluded. Antibiotic exposures were incorporated as time-dependent variables. Potential confounding variables were examined, including diagnosis-related groups, principal diagnosis and procedure, ward location, age, sex, and other antibiotic exposure. Assumptions of the Cox proportional hazards model were tested [9]. Results are reported as adjusted hazard ratios (AHRs) with 95% CIs. Statistical analysis was performed using STATA software, version 6.0 (STATA).

Results

During the 5-year study period, 35,423 medical and surgical patients were admitted. Important characteristics of the study population are shown in table 1. Table 2 lists a classification of nosocomially isolated Enterobacteriaceae and Pseudomonas species that were resistant to third-generation cephalosporins, fluoroquinolones, imipenem, and ampicillin-sulbactam. These microorganisms were isolated from clinical specimens without the detection of any important nosocomial outbreak of gram-negative infections.

Table 1

Important characteristics of the study population.

Table 1

Important characteristics of the study population.

Table 2

Nosocomially isolated gram-negative microorganisms associated with antibiotic resistance.

Table 2

Nosocomially isolated gram-negative microorganisms associated with antibiotic resistance.

From 1994 through 1998, the use of fluoroquinolones, third-generation cephalosporins, and ampicillin-sulbactam increased from 67.5 to 123.0 DDDs per 1000 patient-days (relative increase, 82%), from 42.1 to 58.0 DDDs per 1000 patient-days (relative increase, 38%), and from 36.0 to 71.5 DDDs per 1000 patient-days (relative increase, 99%), respectively (figure 1A–figure 1C). In contrast, the use of imipenem decreased from 55.6 to 34.6 DDDs per 1000 patient-days (relative decrease, 38%; figure 1D). Despite the increasing use of third-generation cephalosporins, fluoroquinolones, and ampicillin-sulbactam, the proportion of Enterobacteriaceae and Pseudomonas species that were susceptible to these agents was relatively stable during the 5-year period. For instance, from 1994 through 1998, the proportion of fluoroquinolone-susceptible Enterobacteriaceae decreased from 99.7% to 95.2% (P = .57), whereas the proportion of third-generation cephalosporin—susceptible Enterobacteriaceae increased from 91.3% to 92.3% (P = .81; figure 1A, 1B). Likewise, despite important changes in the prescription of ceftazidime and imipenem, the proportion of ceftazidime-resistant P. aeruginosa (relative decrease, 1.9%; P = .96) and imipenem-resistant P. aeruginosa (relative increase, 4.1%; P = .74) changed only minimally.

Figure 1

A, Third-generation cephalosporin susceptibility among Enterobacteriaceae (solid line) and use in defined daily doses (DDDs; dotted line). B, Fluoroquinolone susceptibility among Enterobacteriaceae (solid line) and use in DDDs (dotted line). C, Ampicillin-sulbactam susceptibility among Enterobacteriaceae (solid line) and use in DDDs (dotted line). Susceptibility data for 1994 were not available. D, Imipenem susceptibility among Pseudomonas aeruginosa (solid line) and use in DDDs (dotted line). Data are from medical and surgical services, University of Utah Medical Center, 1994–1998.

Figure 1

A, Third-generation cephalosporin susceptibility among Enterobacteriaceae (solid line) and use in defined daily doses (DDDs; dotted line). B, Fluoroquinolone susceptibility among Enterobacteriaceae (solid line) and use in DDDs (dotted line). C, Ampicillin-sulbactam susceptibility among Enterobacteriaceae (solid line) and use in DDDs (dotted line). Susceptibility data for 1994 were not available. D, Imipenem susceptibility among Pseudomonas aeruginosa (solid line) and use in DDDs (dotted line). Data are from medical and surgical services, University of Utah Medical Center, 1994–1998.

Data regarding antibiotic use and susceptibility correlated poorly for most of the stratified antibiotic-use-versus-susceptibility relations. For instance, as shown in figure 2A the correlation between yearly fluoroquinolone use and fluoroquinolone-resistant Pseudomonas species was weak (prevalence ratio per year, 1.002; P = .8). The proportion of Pseudomonas species that were resistant to ceftazidime, imipenem, or ciprofloxacin across different wards also demonstrated an inconsistent relation to the use of these antibiotics. Only the stratified ward-level analysis of ampicillin-sulbactam—resistant Enterobacteriaceae showed a significant drug-use-versus-susceptibility relation (prevalence ratio, 2.15; P < .001; figure 2C). In contrast, the ward-level analysis of third-generation cephalosporin—resistant Enterobacteriaceae yielded a smaller and statistically insignificant relation (prevalence ratio, 1.76; P = .2; figure 2B).

Figure 2

A, Group-level analysis of fluoroquinolone resistance in Pseudomonas species, stratified by year. Linear regression coefficient (R2) = .1 (Y = 16 + 0.035 × DDD). B, Group-level analysis of third-generation cephalosporin resistance in Enterobacteriaceae, stratified by ward location. R2 = .66 (Y = 0.2 + 0.17 × DDD). C, Group-level analysis of ampicillin-sulbactam resistance in Enterobacteriaceae, stratified by ward location. R2 = 1.0 (Y = 26 + 0.16 × DDD). BMT, bone marrow transplant unit; BURN, burn unit; ICU, intensive care unit.

Figure 2

A, Group-level analysis of fluoroquinolone resistance in Pseudomonas species, stratified by year. Linear regression coefficient (R2) = .1 (Y = 16 + 0.035 × DDD). B, Group-level analysis of third-generation cephalosporin resistance in Enterobacteriaceae, stratified by ward location. R2 = .66 (Y = 0.2 + 0.17 × DDD). C, Group-level analysis of ampicillin-sulbactam resistance in Enterobacteriaceae, stratified by ward location. R2 = 1.0 (Y = 26 + 0.16 × DDD). BMT, bone marrow transplant unit; BURN, burn unit; ICU, intensive care unit.

As shown in figure 3A the probability of isolation of fluoroquinolone-resistant Enterobacteriaceae or Pseudomonas species 60 days after admission was 0.01 (95% CI, 0–0.02) for patients without fluoroquinolone exposure and 0.16 (95% CI, 0.10–0.25) for patients treated with fluoroquinolones (P < .001, by means of the log-rank test). figure 3B–figure 3D displays similar cumulative hazard curves for resistance to third-generation cephalosporins (P < .001), ampicillin-sulbactam (P = .006), and imipenem (P < .001).

Figure 3

Cumulative hazard estimates for the 75-day resistance rate, according to (A) fluoroquinolone resistance by fluoroquinolone exposure, (B) third-generation cephalosporin resistance by third-generation cephalosporin exposure, (C) ampicillin-sulbactam resistance by ampicillin-sulbactam exposure, and (D) imipenem resistance by imipenem exposure. Data are from the University of Utah Medical Center, 1994–1998.

Figure 3

Cumulative hazard estimates for the 75-day resistance rate, according to (A) fluoroquinolone resistance by fluoroquinolone exposure, (B) third-generation cephalosporin resistance by third-generation cephalosporin exposure, (C) ampicillin-sulbactam resistance by ampicillin-sulbactam exposure, and (D) imipenem resistance by imipenem exposure. Data are from the University of Utah Medical Center, 1994–1998.

After performance of multivariable analyses at the individual-patient level, fluoroquinolone exposure was determined to be a strong risk factor for fluoroquinolone resistance (AHR, 4.0; P = .001). Likewise, exposure to a third-generation cephalosporin was a strong risk factor for resistance to that drug (AHR, 3.5; P < .001), as were use of imipenem for imipenem resistance (AHR, 5.7; P < .001) and use of ampicillin-sulbactam for ampicillin-sulbactam resistance (AHR, 2.3; P = .008). As shown in table 3, all antibiotic effects were statistically significant, were not explained by confounding due to ward location or exposure to other antibiotics, and were specific for each drug-use-versus-susceptibility relation. An additional patient-level analysis revealed that the strongest association between antibiotic use and resistance was found for subjects from whom susceptible strains had been previously recovered.

Table 3

Individual patient-level risk factor analysis of the effect of antibiotic exposure in multivariable Cox regression models.

Table 3

Individual patient-level risk factor analysis of the effect of antibiotic exposure in multivariable Cox regression models.

We observed that fluoroquinolone use had a “protective” effect against recovery of third-generation cephalosporin—resistant Enterobacteriaceae (AHR, 0.4; P = .003). Likewise, use of a third-generation cephalosporin was “protective” against recovery of ampicillin-sulbactam—resistant Enterobacteriaceae (AHR, 0.6; P = .02).

Discussion

Antibiotic exposure remains one of the most important risk factors for the acquisition of antibiotic-resistant, gram-negative microorganisms by hospitalized patients [10–12]. We applied ecologic methods similar to those reported in previous studies [1, 2, 13] to analyze drug-use-versus-susceptibility relations for gram-negative pathogens over time and across location within a single institution. Group-level trends between antibiotic use and resistance were observed, but they were small in magnitude and not statistically significant (except for the ward-level analysis of ampicillin-sulbactam). When the same antimicrobial agents were analyzed at the individual-patient level, significant associations between antibiotic exposure and resistance were seen for all studied agents. Thus, use of group-level data to make causal inferences about effects of antimicrobial agents on resistance at the individual-patient level may be subject to ecologic bias.

Different types of epidemiological studies have been used to quantify the association between antibiotic exposure and gram-negative resistance in hospitalized patients. These studies included outbreak reports, laboratory-based surveys, randomized trials, and prospective or retrospective cohort studies based on analyses of individual-patient—level data or aggregated data [10, 12, 14–17]. The different methodological approaches are not mutually transposable, and the lack of uniformity makes the comparison of different studies difficult. Thus, we identified an urgent need to perform a parallel analysis of individual-patient—level and group-level data in order to facilitate further comparative studies. On the basis of our findings, we suggest that group-level analytic methods alone do not reliably elucidate causal relationships between antibiotic exposure and resistance in gram-negative microorganisms. Moreover, multicenter studies that do not include individual-patient—level data on antibiotic exposures may yield distorted results; therefore, such studies should incorporate individual-patient—level information to validate the results.

Group-level analyses, as stated forcefully in numerous critical reviews about ecologic studies [3, 4], may distort cross-level inferences. Four types of biases may contribute to inconsistencies between individual-patient– and group-level analyses in studies examining drug-use-versus-susceptibility relationships. First, disparate methods may be used for quantifying antibiotic exposure [8]. Thus, antibiotic exposure on an individual level may be similar, unrelated, or even opposite to trends in antibiotic use on an institutional level. Second, divergent outcome definitions may be applied (prevalence vs. incidence measures). Moreover, the temporal relationship between antibiotic exposure and resistance may vary according to the examined time windows. Group-level analyses including only short periods of time may be inaccurate, because selection of resistant strains may take time [1].

Third, group-level analyses of antibiotic resistance are sensitive to statistical distortion, because critical information is lost in the process of data aggregation [4]. For instance, in ecologic studies, no distinction is made between antibiotic exposure for the time periods before or after isolation of a resistant microorganism. Ecologic estimates may even suffer from biases that have no analogue in individual-level studies and from unusual biases due to misclassification [3, 18]. Fourth, cross-transmission of nosocomial pathogens may create nonlinear dynamics [19]. For instance, exposure to ceftazidime may increase the risk of acquisition of Klebsiella pneumoniae resistant to ceftazidime, not just for the exposed individual but also for close contacts. Conversely, as shown in our analysis, fluoroquinolones may decrease the transmission and recovery of ceftazidime-resistant K. pneumoniae by eliminating K. pneumoniae carriage. Thus, the effect of an individual-level exposure can be decreased or amplified as a result of an interaction between the individual and the group effect. This group-level or contextual effect may be particularly important for gram-positive pathogens, such as vancomycin-resistant enterococci [20].

Notwithstanding their many pitfalls, ecologic studies provide a potentially useful function in studies of infectious agents, because they allow measurement of the total effect of an exposure. This is important, because the total effects of antibiotics encompass not just the direct effects on the individual who receives the antibiotic but also the indirect effects mediated by effects on transmissibility or on the likelihood of transmission of susceptible organisms.

Several authors have advocated that individual hospitals monitor the association between antimicrobial use and resistance within specific patient-care areas, because hospital-wide surveillance studies can mask the occurrence of important resistance problems within specific units [21, 22]. We agree that antibiotic restriction policies should be preferentially based on data from specific patient-care areas, rather than on hospital-wide surveillance data. However, it is also worth noting that even ward-level analyses of aggregate data may show biased effect estimates, as suggested by our analysis. Thus, care should be applied before installing antibiotic restriction policies that may be distorted by ecologic bias, even on the ward level.

In conclusion, group-level analytic methods alone do not reliably elucidate causal relationships between antibiotic use and the resistance of gram-negative organisms. Therefore, multicenter studies that do not include individual-patient—level data on antibiotic exposure may yield biased results. Our work provides a strong methodological foundation and important information that will be critical for future studies that will compare antibiotic use and associated antibiotic resistance patterns in different settings.

Acknowledgment

We thank Mark Lipsitch, Harvard School of Public Health, Boston, for his assistance in this project.

References

1
Monnet
DL
Archibald
LK
Phillips
L
Tenover
FC
McGowan
JE
Gaynes
RP
Antimicrobial use and resistance in eight US hospitals: complexities of analysis and modeling
Infect Control Hosp Epidemiol
 , 
1998
, vol. 
19
 (pg. 
388
-
94
)
2
Fridkin
SK
Steward
CD
Edwards
JR
, et al.  . 
Surveillance of antimicrobial use and antimicrobial resistance in United States hospitals: project ICARE phase 2
Clin Infect Dis
 , 
1999
, vol. 
29
 (pg. 
245
-
52
)
3
Greenland
S
Morgenstern
H
Ecological bias, confounding, and effect modification
Int J Epidemiol
 , 
1989
, vol. 
18
 (pg. 
269
-
74
)
4
Greenland
S
Divergent biases in ecologic and individual-level studies
Stat Med
 , 
1992
, vol. 
11
 (pg. 
1209
-
23
)
5
Evans
RS
Pestotnik
SL
Classen
DC
, et al.  . 
A computer-assisted management program for antibiotics and other antiinfective agents
N Engl J Med
 , 
1998
, vol. 
338
 (pg. 
232
-
8
)
6
Natsch
S
Hekster
YA
de Jong
R
Heerdink
ER
Herings
RM
van der Meer
JW
Application of the ATC/DDD methodology to monitor antibiotic drug use
Eur J Clin Microbiol Infect Dis
 , 
1998
, vol. 
17
 (pg. 
20
-
4
)
7
Carling
PC
Fung
T
Coldiron
JS
Parenteral antibiotic use in acute-care hospitals: a standardized analysis of 14 institutions
Clin Infect Dis
 , 
1999
, vol. 
29
 (pg. 
1189
-
96
)
8
Harbarth
S
Viot
M
Beeler
I
Klastersky
J
Szucs
T
Variation in antimicrobial utilization for febrile neutropenia in cancer patients
Infection
 , 
2000
, vol. 
28
 (pg. 
375
-
8
)
9
Christensen
E
Multivariate survival analysis using Cox's regression model
Hepatology
 , 
1987
, vol. 
7
 (pg. 
1346
-
58
)
10
Lucet
JF
Chevret
S
Decre
D
, et al.  . 
Outbreak of multiply resistant Enterobacteriaceae in an intensive care unit: epidemiology and risk factors for acquisition
Clin Infect Dis
 , 
1996
, vol. 
22
 (pg. 
430
-
6
)
11
D'Agata
E
Venkataraman
L
DeGirolami
P
, et al.  . 
Colonization with broad-spectrum cephalosporin-resistant gram-negative bacilli in intensive care units during a non-outbreak period: prevalence, risk factors, and rate of infection
Crit Care Med
 , 
1999
, vol. 
27
 (pg. 
1090
-
5
)
12
Richard
P
Delangle
MH
Raffi
F
Espaze
E
Richet
H
Impact of fluoroquinolone administration on the emergence of fluoroquinolone-resistant gram-negative bacilli from gastrointestinal flora
Clin Infect Dis
 , 
2001
, vol. 
32
 (pg. 
162
-
6
)
13
Rahal
JJ
Urban
C
Horn
D
, et al.  . 
Class restriction of cephalosporin use to control total cephalosporin resistance in nosocomial Klebsiella
JAMA
 , 
1998
, vol. 
280
 (pg. 
1233
-
7
)
14
Ballow
CH
Schentag
JJ
Trends in antibiotic utilization and bacterial resistance
Diagn Microbiol Infect Dis
 , 
1992
, vol. 
15
 (pg. 
37
-
42
)
15
Jaccard
C
Troillet
N
Harbarth
S
, et al.  . 
Prospective randomized comparison of imipenem-cilastatin and piperacillin-tazobactam in nosocomial pneumonia or peritonitis
Antimicrob Agents Chemother
 , 
1998
, vol. 
42
 (pg. 
2966
-
72
)
16
Harbarth
S
Rohner
P
Safran
E
Garbino
J
Auckenthaler
R
Pittet
D
Secular trends in amikacin and gentamicin resistance among gram-negative bloodstream isolates in a university hospital between 1989 and 1994
Clin Microbiol Infect
 , 
1998
, vol. 
4
 (pg. 
199
-
204
)
17
Carmeli
Y
Troillet
N
Eliopoulos
GM
Samore
MH
Emergence of antibiotic-resistant Pseudomonas aeruginosa: comparison of risks associated with different antipseudomonal agents
Antimicrob Agents Chemother
 , 
1999
, vol. 
43
 (pg. 
1379
-
82
)
18
Brenner
H
Savitz
DA
Jockel
KH
Greenland
S
Effects of nondifferential exposure misclassification in ecologic studies
Am J Epidemiol
 , 
1992
, vol. 
135
 (pg. 
85
-
95
)
19
Koopman
JS
Longini
IM
Jr
The ecological effects of individual exposures and nonlinear disease dynamics in populations
Am J Public Health
 , 
1994
, vol. 
84
 (pg. 
836
-
42
)
20
Austin
DJ
Anderson
RM
Transmission dynamics of epidemic methicillin-resistant Staphylococcus aureus and vancomycin-resistant enterococci in England and Wales
J Infect Dis
 , 
1999
, vol. 
179
 (pg. 
883
-
91
)
21
Bryce
EA
Smith
JA
Focused microbiological surveillance and gram-negative beta-lactamase-mediated resistance in an intensive care unit
Infect Control Hosp Epidemiol
 , 
1995
, vol. 
16
 (pg. 
331
-
4
)
22
White
RL
Friedrich
LV
Mihm
LB
Bosso
JA
Assessment of the relationship between antimicrobial usage and susceptibility: differences between the hospital and specific patient-care areas
Clin Infect Dis
 , 
2000
, vol. 
31
 (pg. 
16
-
23
)
Financial support: Max-Kade Research Fellowship Award (to S.H.).
Presented in part: 39th Interscience Conference on Antimicrobial Agents and Chemotherapy, San Francisco, September 1999 (abstract no. 730).

Comments

0 Comments