To evaluate the effect of using specific antimicrobial agents on the prevalence of VRE among a wide variety of ICUs in U.S. hospitals, the Hospital Infections Program at the Centers for Disease Control and Prevention, in cooperation with the Rollins School of Public Health of Emory University (Atlanta, Georgia), began the second phase of Project ICARE (Intensive Care Antimicrobial Resistance Epidemiology) in 1996. By using data from hospitals participating in Project ICARE, we studied the relationship between institutional characteristics (such as rates of nosocomial infection or antimicrobial use) and the prevalence of VRE in adult ICUs.

Methods

Top Methods Results Discussion Author & Article Info References

Setting

Hospitals that participated in the ICU surveillance component of the National Nosocomial Infections Surveillance (NNIS) system were invited to participate in the second (January 1996 through December 1997) and third (April 1998 through July 1999) phases of Project ICARE. One hundred seventy-one adult ICUs from 60 hospitals reported the required data to the ICARE component of the NNIS system during the study period of January 1996 through July 1999. The surveillance method and definitions of the NNIS system and Project ICARE have been previously described (18-20).

Data Elements

Participating hospitals reported monthly nosocomial infection data from at least one ICU. These data included information on device-associated infections (for example, central line–associated bloodstream infection, ventilator-associated pneumonia, and catheter-associated urinary tract infection), number of ICU patient admissions, days of device use by patients (device-days), and total inpatient days for the ICU. These data permitted calculation of device-associated infection rates (infections per 1000 device-days), device use rates (percentage of patient-days in which specific devices were used), and average length of stay.

Hospitals also reported monthly on the amount (in grams) of select antimicrobial agents administered to patients and the susceptibility results of isolates recovered from clinical specimens from hospitalized patients. Microbiological data were aggregated for each ICU separately, all non-ICU inpatient wards combined, and all outpatient areas combined. Pharmacy data were reported for the same hospital strata, but pharmacy data from outpatient areas were not routinely available and were often inaccurate because of the nature of U.S. retail pharmacies for outpatient prescriptions. To minimize information bias in aggregating outpatient antimicrobial use, we declined to analyze these data. The reported amounts of antimicrobial agents were standardized by conversion to defined daily doses; for parenteral vancomycin, one defined daily dose is equivalent to 2 g. Antimicrobial agents of a similar class or activity were grouped (Table 1).

The microbiology laboratories at participating hospitals reported antimicrobial susceptibility results for all enterococcal isolates recovered from all clinical specimens associated with hospital- or community-acquired infection or colonization. We excluded duplicate isolates, which were defined as isolates of the same organism with the same antimicrobial resistance pattern recovered from the same patient, regardless of the site of isolation (for example, blood, sputum, urine, or wound) during the same calendar month. In addition, we excluded susceptibility reports from isolates obtained for infection control surveillance. By excluding these surveillance isolates, the VRE prevalence more closely reflects data routinely aggregated as part of the laboratories' cumulative susceptibility report (cumulative antibiogram). Finally, the validity of the susceptibility data was assessed previously, and participating laboratories were determined to reliably identify enterococcus and correctly test vancomycin susceptibility. This was done through a proficiency testing program at these laboratories (21) and confirmatory testing of up to 20 vancomycin-resistant enterococci isolates from phase 2 Project ICARE hospitals (20). Enterococci were considered resistant if the minimum inhibitory concentration of vancomycin was greater than or equal to 32 µg/mL or if the zone diameter by disk diffusion was less than 14 mm.

Monthly data were pooled for the study period for each ICU, and pooled rates were calculated for device-associated infection, device use, average length of stay, prevalence (percentage) of VRE, and antimicrobial use (defined daily dose/1000 patient-days). For example, the pooled mean rate of vancomycin use was calculated for each ICU by dividing the total number of defined daily doses by the total number of patient-days reported over the study period by that ICU and multiplying that value by 1000; the rate is expressed as defined daily doses per 1000 patient-days. For this analysis, evaluation of antimicrobial use was limited to agents that have been commonly implicated as either a risk factor or a protective factor for VRE acquisition (1, 6-8, 17), such as the ampicillin drug group; first-generation, second-generation, or extended-spectrum (third-generation) cephalosporins; vancomycin; and the piperacillin and ticarcillin groups (Table 1). Antimicrobial agents were evaluated as single agents and as groups of similar agents, such as the ß-lactamase inhibitor group (for example, piperacillin–tazobactam, ticarcillin–clavulanic acid, ampicillin–sulbactam, and amoxicillin–clavulanic acid) or the antipseudomonal penicillin group (for example, the piperacillin and ticarcillin groups). Rates for VRE were calculated for each ICU and for all non-ICU inpatient areas combined. For specific ICUs testing fewer than 10 enterococcal isolates for susceptibility during the study period, the prevalence of VRE was considered to be of low accuracy, and such ICUs were excluded from further analysis.

As participants in the NNIS system, hospital personnel had categorized each ICU at their hospital by the types of patients served—cardiac care, medical, general surgical, cardiothoracic, combined medical–surgical (units in which <80% of patients could be classified into a single unit type), neurosurgical, respiratory, trauma, burn, or "other." For the current analysis, respiratory, trauma, and burn units were grouped with "other" ICUs.

Statistical Analysis

Data were analyzed by using SAS software, release 6.12 (SAS Institute, Inc., Cary, North Carolina). We evaluated selected factors (such as ICU or hospital characteristics, device use rates, nosocomial infection rates, or rates of antimicrobial use) that were potentially associated with increased or decreased rates of VRE. The relationship between VRE rates and categorical variables was assessed by comparing the median values of VRE rates by using the Kruskal–Wallis test. Continuous variables were evaluated in two ways. First, the relationship was assessed by using the Spearman rank-correlation coefficient. Second, the relationship was assessed in univariate weighted linear regression models. The weight for each VRE prevalence was calculated as the reciprocal of the variance. The resulting model thus considered the weighted effect of the number of enterococci reported by each ICU. This model helped us identify antimicrobial exposures associated with VRE after accounting for the varied amount of data submitted by each ICU. All reported P values are two tailed.

To assess the joint influence of these antimicrobial use factors together with hospital or ICU characteristics on VRE prevalence, we used stepwise weighted linear regression techniques to identify the most important main effects. Each eligible antimicrobial agent or group of related antimicrobial agents (Table 1) was entered into the unadjusted models as continuous and ordinal (that is, according to quartile) data. This allowed us to identify exposures for which there was a better model fit when the data were evaluated as ordinal rather than continuous data. Continuous variables that were eligible for inclusion in the model-building process as potential independent predictors (significant at P < 0.05 by weighted univariate linear regression or strong biological plausibility, as with cephalosporin use) were then eligible to enter the model. We had no specific exposure variables in mind and tested all variables as potential confounders by looking at the change in the coefficient and P values for main effect variables (for example, antimicrobial use) after the inclusion of potential confounders (such as prevalence of VRE in the non-ICU setting). By evaluating both the degree of statistical significance between a potential effect variable and the impact on the coefficients of the other variables in the model, we determined whether the variable should remain in the model. Detection of potential outlier data points and their influence on main effect factors were also assessed in the modeling process.

Role of the Funding Source

The funding source had no role in the collection, analysis, and interpretation of the data or in the decision to submit the paper for publication.

Results

Top Methods Results Discussion Author & Article Info References

Description of Sites

During the study period, 60 hospitals followed the surveillance protocol and reported a median of 12 months (interquartile range, 11 to 24 months) of data from a total of 171 adult ICUs. Hospitals were from 26 states (4 New England states, 15 Mid-Atlantic, 19 South Atlantic, 12 East Central, 6 West Central, and 4 Pacific) and had a median hospital bed size of 385 (range, 147 to 1206); 30 (50%) reported a major affiliation with a teaching institution (major teaching centers), and 6 (10%) were Veterans Affairs medical centers. Forty-five (26%) ICUs were excluded from further analysis because susceptibility results on fewer than 10 enterococcal isolates were reported. The remaining 126 ICUs included 37 medical–surgical ICUs, 20 cardiac care ICUs, 26 medical ICUs, 23 general surgical ICUs, 9 cardiothoracic ICUs, 6 neurosurgical ICUs, and 5 "other" ICUs (2 burn, 2 trauma, and 1 respiratory). These ICUs reported similar median values for the number of ICU beds and the number of days that patients received care during the study period (patient-days) (Table 2). However, the average length of stay, rates of device use, and device-associated infections (excluding rates of central line–associated bloodstream infection) varied significantly among the ICU types (Table 2).

Rates of VRE

In the 126 ICUs, the median prevalence of VRE was 10.0% (range, 0% to 59%). Median rates did not vary significantly by ICU type (Table 2). However, rates of VRE were significantly higher among ICUs at major teaching centers (median rate, 12.6% vs. 5.6% overall; median absolute difference, 7.3 percentage points [95% CI, 4.0 to 11.2 percentage points]) or larger hospitals (those with >500 beds) (12.5% vs. 6.5%; median absolute difference, 5.0 percentage points [CI, 1.1 to 9.5 percentage points]). Correlations of the rate of VRE in an ICU with the rate among all non-ICU areas combined (r = 0.63 [CI, 0.52 to 0.73]) or for all outpatient areas combined (r = 0.53 [CI, 0.39 to 0.66]) were statistically significant and strongly positive. In addition, correlation between the rate of VRE in an ICU with the ICU-specific rate of ventilator use (r = 0.20 [CI, 0.01 to 0.38]) was weakly positive and of marginal significance.

Hospital classification as a Veterans Affairs medical center was not associated with VRE rate (9.9% in Veterans Affairs hospitals vs. 12.1% in other hospitals; median absolute difference, 5.3 percentage points [CI, –2.5 to 10.7 percentage points]). In addition, rate of VRE was not associated with ICU-specific rate of central line–associated bloodstream infection (r = 0.03 [CI, –0.20 to 0.26]), ventilator-associated pneumonia (r = 0.12 [CI, – 0.12 to 0.35]), catheter-associated urinary tract infection (r = –0.01 [CI, –0.24 to 0.22]), or certain device use rates (for central venous catheter, r = 0.14 [CI, –0.10 to 0.36]; for urinary catheter, r = –0.10 [CI, –0.33 to 0.14]; for number of ICU beds, r = –0.03 [CI, –0.26 to 0.20]).

Antimicrobial Use

Rates of use for each of the antimicrobial agents and for groups of agents differed dramatically and significantly among ICU types. For most of the drug groups evaluated, the rate of use was highest in the medical, medical–surgical, and general surgical ICUs and was lowest in the coronary care, neurosurgical, and cardiothoracic ICUs (except for vancomycin use, which was high in the neurosurgical and cardiothoracic ICUs) (Table 2).

In the unweighted univariate analysis, the rate of VRE in an ICU was significantly correlated with ICU-specific use of vancomycin (r = 0.44 [CI, 0.29 to 0.57]) (Figure). However, apart from marginal significance in correlation with piperacillin–tazobactam (r = 0.20 [CI, 0.03 to 0.36]), rates of use of other antimicrobial agents were not significantly correlated with VRE rates according to the Spearman correlation. The association between antimicrobial use and prevalence of VRE was further assessed in weighted linear regression models, by which several additional agents showed statistically significant associations: ß-lactamase inhibitor combination group (P = 0.04), first-generation cephalosporin group (P = 0.03), and ampicillin group (P = 0.06). Significant association of the antipseudomonal penicillin group with VRE prevalence (P = 0.008) was due to rates of association between VRE prevalence with ticarcillin use (with or without clavulanic acid; P < 0.001) rather than with piperacillin use (with or without tazobactam; P > 0.2). Among the other individual antimicrobial agents evaluated, none were significantly associated with VRE prevalence (data not shown).

View larger version (13K): [in this window] [in a new window]   Figure. Correlation of the rate of parenteral vancomycin (defined daily doses per 1000 patient-days) and the prevalence of vancomycin-resistant enterococci (percentage of enterococci that are vancomycin resistant) in 126 intensive care units.P values were determined for the Spearman correlation coefficient (r = 0.44 [95% CI, 0.29 to 0.57]) and weighted linear regression (parameter estimate = 0.08; P = 0.001).

Multivariable Analysis

To assess the independent importance of factors identified in univariate analysis and the relative impact of antimicrobial use in these ICUs, we performed a stepwise weighted linear regression analysis. The final model included 123 ICUs that reported sufficient data from all factors under investigation (3 ICUs were from one hospital that did not report on all factors in the final model). The most significant factor associated with VRE prevalence in these ICUs was the rate of VRE in the non-ICU inpatient areas (Table 3). The rate of vancomycin use in the ICU was the most significant "modifiable" ICU-specific factor associated with VRE prevalence. Analysis after adjustment for ICU type showed cardiothoracic and neurosurgical ICUs to be associated with significantly lower VRE prevalence. According to this adjusted model, the only other antimicrobial use showing a significant association with VRE prevalence was especially high doses (>75th percentile of overall use) of third-generation cephalosporins (223 defined daily doses/1000 patient-days). No other antimicrobial agent or group of antimicrobial agents was significantly associated with a higher or lower VRE rate. Finally, no other ICU-specific factor among those considered modifiable (that is, nosocomial infection factors) was associated with VRE prevalence.

Discussion

Top Methods Results Discussion Author & Article Info References

Prevention and control of hospital-acquired infections associated with VRE have focused on measures to prevent cross-transmission between patients and practices to control antimicrobial use. Specifically, diminishing unnecessary use of vancomycin is proposed as a control measure by the Hospital Infection Control Practices Advisory Committee (9). Our data support these recommendations by demonstrating that increased non-ICU VRE rates and increased rates of vancomycin use are associated with higher rates of VRE isolated from nonsurveillance cultures from adults in ICUs. In addition, higher rates of third-generation cephalosporin use were associated with higher rates of VRE. These associations between specific antimicrobial use and VRE prevalence were significant even after adjustment for the amount of VRE reported outside of the ICU setting and for case mix (using ICU type as a proxy measure for case mix).

This type of ecologic study has been performed at single institutions evaluating hospital-wide rates of VRE infection or colonization, and results have been conflicting (6, 17). In part, this may have arisen from small sample sizes. By contrast, our final analysis was based on a median of 12 months of microbiological and pharmacy data from each of 123 adult ICUs. Performing multivariate analysis in such a multicenter study makes our findings more applicable to the adult ICUs typically found in U.S. hospitals. The results support the assertion that reducing the rates of use of these specific antimicrobial agents could decrease the incidence of VRE in these ICUs. In contrast to a previous study (17), our investigation showed no association between use of piperacillin (with or without tazobactam) or the antimicrobial agents in the ß-lactamase inhibitor combination group and lower rates of VRE. In our univariate analysis, higher rates of use of these agents had a modest positive correlation with VRE rate. However, when we controlled for ICU type and VRE prevalence in non-ICU areas, use of agents in the ß-lactamase inhibitor combination group was not an independent predictor of VRE (as either a protective or risk factor).

The effect of third-generation cephalosporin use on VRE rate was found to be significant in the multivariate model only after adjustment for VRE prevalence in non-ICU areas by using weighted multivariable analysis; this finding suggests that the ICU-specific use of these agents has the greatest impact on VRE prevalence in ICUs in which VRE prevalence in non-ICU areas of the same hospital has poor association with ICU-specific prevalence of VRE. Furthermore, the effect of these agents on VRE prevalence proved significant only above a certain threshold (the 75th percentile for all ICUs); use above this threshold was associated with significantly increased risk. In contrast, incremental increases in vancomycin use were associated with equal corresponding increases in risk for higher VRE prevalence. These findings may indicate that incremental changes in vancomycin use can affect ICU-specific rates of VRE while major reductions in rates of third-generation cephalosporin can affect rates of VRE. The reasons for these apparently different types of associations deserve further study.

Our ICU-level analysis has several limitations. First, estimating cross-transmission frequency by molecular studies to determine the genetic relatedness of all VRE isolated from participating ICUs was beyond the scope of this study. Therefore, our model lacks a factor to account for cross-transmission of epidemic strains in any particular ICU; such transmission may explain some of the variation observed in the data, and the role of such transmission relative to antimicrobial use cannot be directly assessed from our data. However, to better address this, we did contact all ICUs that reported higher (top 90th percentile) rates of VRE, and none reported an outbreak of VRE in their ICU. Second, no standardized severity-of-illness measure was used in our comparisons (22, 23). However, we could include the type of ICU and status of major teaching center as proxy measures for case mix. Furthermore, we suspect that the average length of stay and rates of device use in a given ICU are proxy measures for severity of illness, although these factors were not significant in the multivariate model. Third, our analysis of pooled data over a study period did not evaluate a possible temporal association between specific antimicrobial use and VRE; this evaluation will be possible in a future analysis after receipt of additional data. Finally, our study was of an ecologic nature and did not include collection of data on patient-specific exposures or risk factors or on specific clinical outcomes regarding infection, treatment, or mortality associated with VRE. While ecologic studies have several practical advantages, their lack of individual-level data limits the extent to which causal inferences can be drawn and makes control of confounding more challenging (24). On a related issue, we did not ascertain the clinical relevance of the isolates. However, we have analyzed these and related data from the ICU component of the National Nosocomial Infections Surveillance System and have determined that the aggregated susceptibility data from isolates reported in Project ICARE (including the data in this study) accurately reflect the resistance prevalence related only to isolates associated with hospital-acquired infections (25). Despite such limitations, our data demonstrate an association between the rates of use of some antimicrobial agents and rates of antimicrobial resistance at an institutional level for 60 hospitals. These findings are supported by those of single-institution studies that used patient-level analysis (1, 5, 7).

Unit-level analysis of this type has several advantages compared with the traditional patient-level analysis. Collection of unit-level data on rates of use and resistant organisms can be incorporated into routine infection control surveillance activities. The data collection mechanisms can be automated in pharmacies and microbiology laboratories. Then, such data from ongoing surveillance can be used to monitor the effect of control interventions (such as implementing guidelines on use of antimicrobial agents or enhanced isolation efforts) on rates of antimicrobial use and resistance; they can also be included in existing quality-improvement processes at hospitals (26, 27).

Several previous studies have demonstrated that vancomycin is frequently used inappropriately (10-16). Our data can be used to support efforts at improving the judicious use of specific antimicrobial agents in adult ICUs. Methods to accomplish this are not well established, but several approaches have been recommended (26, 27); the most efficient methods will probably differ among institutions. Systems such as those used in Project ICARE for collection of pharmacy and microbiology data can provide insight into factors associated with increased rates of antimicrobial-resistant bacteria and antimicrobial use itself. A previous analysis of Project ICARE data showed that after adjustment for type of ICU, the major determinants of vancomycin use in adult ICUs were the endemic rate of methicillin-resistant Staphylococcus aureus bacteremia in the hospital and the rate of central line–associated bloodstream infections in the ICU (28, 29). These findings suggested that reducing rates of methicillin-resistant S. aureus or central line–associated nosocomial infections—traditional priorities of infection control activities—should help to reduce use of vancomycin in adult ICUs (29). Similarly, our analysis identifies use of vancomycin and use of third-generation cephalosporins at very high amounts as modifiable factors within an ICU; reducing use of these agents should reduce rates of VRE. More studies are needed to identify modifiable factors associated with use of other antimicrobial agents. Moreover, these data emphasize that traditional efforts to reduce cross-transmission may be needed hospital-wide because rates of VRE outside of the ICUs greatly affected ICU-specific VRE rates.