Antimicrobial susceptibility data on E. coli collected by the MYSTIC and SENTRY systems were forwarded to GAARD coordinators at APUA for descriptive and inferential analysis of temporal, demographic, and geographic trends. MYSTIC was launched by AstraZeneca in 1997 to study bacterial resistance in specialist and general hospital units in hospitals using meropenem (15). At present, 52 sites from 19 countries are contributing results. Each center isolates up to 100 gram-positive and 100 gram-negative aerobic bacteria per year from routine diagnostic samples from hospitalized patients, excluding repeat patient isolates. Antimicrobial susceptibility tests are performed by broth microdilution by using NCCLS reference methods (16) either on-site (for non-US laboratories) or by a reference laboratory (for US participants). More than 9,000 isolates are processed annually, with at least 9 antimicrobial drugs tested per strain.

Bristol-Myers Squibb established the SENTRY program in 1997 as a global program for the surveillance of resistance in bacterial and fungal populations (17). SENTRY has expanded from 75 sites in 1997 to 94 laboratories in 35 countries in 2003. Bacterial isolates are obtained from diagnostic specimens taken in the course of routine clinical management of both hospitalized and community patients. Each site collects a defined number of consecutively identified strains within a number of distinct protocols, e.g., blood isolates, urine isolates, and respiratory isolates, excluding repeat patient isolates. Strains, including basic patient demographic data, are shipped to a coordinating laboratory for centralized identification and susceptibility testing by broth microdilution panels according to NCCLS reference guidelines (16). Forty-five to 50 antimicrobial drugs are monitored each year, with ≈30 tested per strain; >200,000 strains are processed annually.

Through integrated analysis of data from multiple sources, the GAARD project seeks to realize a number of benefits: 1) increased statistical power in detecting evolutionary events of public health importance and elucidating risk factors for resistance emergence and spread; 2) greater geographic, demographic, and temporal coverage of bacterial populations than is possible under any single system with limited resources; and 3) cross-validation of findings from complementary data sources with distinct strategies for site recruitment, patient identification, specimen collection, and laboratory testing, which should prompt deeper investigation of seemingly discordant findings (21).

For countries in which a direct comparison of results from the 2 systems was possible, resistance frequencies from MYSTIC were typically higher than from SENTRY. In only 2 instances were higher SENTRY estimates significant (ciprofloxacin in Belgium and Canada). Observation of such incongruent findings should prompt a focused review for possible rationales, such as laboratory testing errors, differences among patient populations sampled, criteria for specimen selection, antimicrobial use patterns, or local outbreaks of resistant organisms. Because SENTRY estimates for Belgium reflect the experience of a single institution while MYSTIC data include results from 9 sites, the MYSTIC results may better reflect the situation in that country.

One of the most substantial findings of the multivariate analysis is that the surveillance system was not associated with nonsusceptibility in any of these compounds, in contrast to the findings of the univariate analyses. Thus, the finding that MYSTIC estimates of nonsusceptibility were consistently higher than SENTRY isolates in paired comparisons may be completely attributable to differences in the demographics of the patient subpopulations sampled. In this study, the principal contributor identified was the proportion of ICU patients represented in the 2 systems. Such findings should increase confidence in the reliability and validity of findings reported separately from the 2 programs. The observation of consistent differences in uncontrolled comparisons of results between systems also highlights the importance of including relevant demographic information in reports on antimicrobial susceptibility rates.

An arbitrary categorization of countries into relatively low, medium, and high resistance is shown in Table 3 for a few key first- and second-line antimicrobial drugs used to treat E. coli infections. The intervals indicated were selected to provide some degree of separation between groups of countries using the observed estimates and should not be interpreted as having a direct clinical implication for therapy decisions. The high rates of resistance to both ampicillin and trimethoprim/sulfamethoxazole in many countries observed in this study should prompt close review of treatment success rates in settings in which they are commonly used in empiric therapy (22).

The use of surveillance data to guide antimicrobial therapy guidelines is a complicated issue that must address the constraints of available resources and therapeutic alternatives, local resistance and antimicrobial use patterns, and potential epidemiologic biases in available data. A number of studies have addressed empiric and quantitative approaches for using surveillance data in treatment guidelines for urinary tract infections and pyelonephritis, including cost-effectiveness studies and establishing resistance thresholds to guide therapy decisions (23–27).

Several significant results were noted in the univariate analyses of temporal trends. Such changes over time could be due to real shifts in the bacterial populations, changes in the number or type of participating institutions, changes in specimen collection practices, or spurious correlations, as no statistical corrections were made for multiple comparisons. The significant decrease to 4 or more agents in Brazil, Chile, and Italy in particular is worth highlighting for further exploration; Chile has successfully implemented and enforced new national legislation banning the sale of antimicrobial drugs without a prescription since 1999, and this legislation has produced substantial reductions in total antimicrobial drug use in the country (28).

Significant findings from the multivariate analysis of core antimicrobial drugs were mentioned above: higher rates of nonsusceptibility in isolates from ICU patients, older patients, and male patients and in isolates from Latin America, the western Pacific, and southern Europe. When all other variables were controlled for, nonsusceptibility to ciprofloxacin showed a statistical increase in over time, while nonsusceptibility to piperacillin/tazobactam decreased. This decrease in nonsusceptibility to piperacillin/tazobactam was significant in 11 countries in univariate analyses and merits further investigation into contributory factors. While temporal trends in the multivariate analysis may reflect, to some degree, the high proportion of US isolates in the SENTRY database, demographic characteristics of SENTRY isolates within and outside the United States were comparable, with only a small but significant difference seen for sex (44.2% [n = 1,058] male in the United States vs. 48.1% [n = 2,331] male outside the United States for 2001 data, p = 0.034).

The higher rate of nonsusceptibility among isolates from male patients has been previously noted for ciprofloxacin resistance (10,12,29) and ascribed to epidemiologic differences between men and women with E. coli infections. Urinary tract infections in male patients are more frequently complicated or healthcare-associated than those in the typical female patient, and infection in men may be associated with higher rates of previous antimicrobial drug usage and time in the hospital setting (29).

The finding of higher resistance in isolates from ICU patients to most agents is not unexpected, given the high selection pressure exerted by intensive antimicrobial use in this setting and the ease of transmission of resistant pathogens on the hands of healthcare workers. The observation that ICU isolates did not have higher rates of resistance to ciprofloxacin, most frequently used in the outpatient setting, suggests that risk factors for ciprofloxacin resistance are distinct from those of the other, principally second-line, agents studied. This dichotomy was also observed with respect to age. For ciprofloxacin, in contrast to the other core antimicrobial drugs, older age was associated with a significant protective effect, i.e., lower nonsusceptibility (OR 0.39, 95% CI 0.29–0.52, p<0.001), than seen in younger patients. One hypothesis holds that resistance in certain antimicrobial drugs, such as intravenous or second-line agents, is more closely associated with patterns of prescribing in hospitals and in older patients, while resistance in others, such as ciprofloxacin, is more correlated with patterns of antimicrobial drug use in the community. This hypothesis merits further investigation in a variety of geographic and clinical settings (30,31). Given the ubiquity of E. coli as a commensal pathogen in the human gut and in animal populations, resistance in E. coli may be a sensitive indicator of distinct therapeutic and nontherapeutic, appropriate and inappropriate uses of antimicrobial drugs (32). Another APUA-coordinated project, Reservoirs on Antibiotic Resistance, is a 5-year scientific collaboration that addresses this issue by exploring the movement of resistance determinants within commensal bacterial populations and between commensals and human pathogens (33).

Both the MYSTIC and SENTRY surveillance networks rely on routinely generated test results, a strategy with advantages over purely research-oriented, resource-intensive special surveys. These advantages include sustainability, more complete organism and geographic coverage, monitoring of baseline trends, infection control alerts, and outbreak detection. However, potential biases may be introduced that must be considered, such as selectively testing patients whose infections did not respond to treatment or who had more severe disease. Such biases may be amplified in the outpatient setting and in low-resource countries where treatment is frequently empiric with limited diagnostic testing. Results from routinely generated sample collections could usefully be compared to findings from periodic validation surveys in which greater resources are expended in identifying and testing representative patient populations (34–36).

With antimicrobial resistance continuing to evolve and present a global public health challenge, appropriately designed and implemented surveillance systems are a priority. Collaboration among existing surveillance systems can improve the quality, breadth, and impact of data for guiding and evaluating clinical and public health policy.