Susceptibility testing results were identified for 14,595 S. aureus clinical isolates obtained from 10,892 patients. A total of 3,725 isolates were from 2,749 patients with an associated MRN; 10,870 were from 8,143 patients identified with an APID. For all patients, the isolate-to-patient ratio was 1.3.

Figure 1

Figure 1. Effect of duplicate isolate removal strategies on the number of Staphylococcus aureus isolates and percentage susceptible to oxacillin for all patients in Hawaii, 2002. The 95% confidence interval for the proportion...

Figure 1 depicts the effect of duplicate isolate removal on S. aureus susceptibility to oxacillin for all isolates. NCCLS and Cerner methods produced similar estimates of susceptibility for any given analysis period, i.e., the difference between the 2 methods was insignificant. Furthermore, the difference in susceptibility percentage between a 90-day and 365-day period was <1% by either NCCLS or Cerner criteria, which was insignificant. With both Cerner and NCCLS methods, the general trend for estimates of susceptibility increased as the period of analysis lengthened.

No removal resulted in the lowest susceptibility estimate (67%) observed, even lower than that for most resistant (Figure 1). Overall, an inverse relationship was observed between number of isolates included in the analysis and proportion of susceptible isolates (Figure 1). For both Cerner and NCCLS methods, point estimates of susceptibility rose slightly as the period of analysis increased and the number of isolates included in the susceptibility calculations decreased.

The patterns observed for all isolates combined remained unchanged when stratified by different clinical settings, i.e., ICU, non-ICU, and outpatient (Table 3). Within a given clinical setting, the difference in susceptibility frequencies with the 90- and 365-day intervals by either Cerner or NCCLS was <1%.

Differences in the magnitude of the effect of duplicate isolate removal were observed across different clinical settings (Table 3). For example, the effect of removal in the non-ICU and ICU environments was an increase of 8% and 6%, respectively, in the susceptible proportion when comparing 365-day NCCLS results to no removal; this comparison resulted in an increase of 2% in the outpatient setting.

In a subanalysis restricted to the 2,749 patients (3,725 isolates) with an associated MRN, results were highly analogous to those observed for the larger cohort as a whole. Specifically, for each clinical setting, the difference in the susceptibility estimate between NCCLS and Cerner methods was insignificant for any given period of analysis, and the difference in susceptibility percentage between a 90- and 365-day time period was <1% with either NCCLS or Cerner. In addition, the outpatient setting continued to show the least effect of duplicate isolate removal when compared to inpatient settings.

Figure 2

Figure 2. Effect of duplicate isolate removal strategies on the number of Staphylococcus aureus isolates and percentage susceptible to oxacillin for all patients in Hawaii, 2002. The 95% confidence interval for the proportion...

Finally, we examined the effect of each deduplication strategy for inpatients (ICU and non-ICU) at major hospitals. Although the hospitals were of different sizes, and their respective rates of MRSA differed by what was seen with no removal, the effects of deduplication observed for each hospital individually were similar to those observed for the population-based surveillance dataset as a whole. Figure 2 illustrates the results of isolate deduplication from 2 of the hospitals with the largest number of S. aureus isolates in 2002.

To our knowledge, this report is the first to compare the effect of different deduplication strategies on susceptibility patterns derived from a statewide, population-based antimicrobial resistance surveillance system. The relatively large sample of S. aureus isolates in this study was obtained from multiple healthcare settings by a variety of clinical laboratories. Because of the diversity of practice patterns represented in the study, we believe our findings are likely to be applicable to other facilities and agencies conducting antimicrobial resistance surveillance for MRSA.

The major findings from this analysis are the following: 1) NCCLS and modified Cerner methods yield similar results for a given analysis period; 2) with both NCCLS and modified Cerner, the number of total isolates included and the percentage that are MRSA decrease slightly as the period of analysis increases; 3) point estimates of the proportion of MRSA produced for a 90- or 365-day analysis period were statistically similar by either NCCLS or Cerner; and 4) the effect of deduplication was greater for inpatient settings compared to outpatient settings.

We also found that no removal produced the highest estimates of MRSA resistance, even higher than most resistant for the same analysis period. While this finding may seem at first paradoxical, it demonstrates the influence that practice patterns may have on reported rates of MRSA and why deduplication is critical. In our setting, cultures of samples from patients with MRSA were obtained more frequently than cultures from those with methicillin-susceptible strains (MSSA) so that, on average, individual patients with MRSA contributed more isolates, which could be included in the estimation of MRSA rates. This fact also explains why we observed, for both NCCLS and Cerner, an inverse relationship between the length of the analysis period and the rate of MRSA, i.e., the greater number of isolates included with shorter periods of analysis meant more MRSA isolates relative to MSSA strains. If reculturing patient samples is influenced by prior susceptibility testing results, an institution's MRSA percentages might be determined by the practice patterns of physicians working within the institution. Furthermore, trends in antimicrobial resistance could be obscured if practice patterns changed and reculturing samples from MSRA patients became more or less common.

Duplicate isolate removal facilitates comparing data among institutions and monitoring trends over time. However, at present, no clear consensus has been reached on the definition of duplicate isolates, and duplicate isolates cannot be easily identified with certainty in clinical practice (2,4–6). Our study found that NCCLS and modified Cerner methods yield similar results, and for either a 90-day or a 365-day analysis period, the produced estimates fall between results produced by the most resistant and most susceptible methods. Thus, NCCLS and Cerner might both be considered reasonable approaches. However, the NCCLS method has 1 major advantage: NCCLS is the only method that does not require the infection control practitioner to simultaneously compare susceptibility results for multiple isolates obtained from a given patient during the analysis period. With NCCLS, one simply includes the susceptibility results for the first isolate obtained during the analysis period. This straightforward approach would minimize opportunity for error and result in more consistent implementation of the deduplication process (5).

Regardless of which deduplication strategy is selected, the question remains which analysis period to adopt (5,6,11). A longer surveillance period increases the probability that an isolate representing a truly new resistance event (rather than a duplicate) will be removed (11,12). The purpose of antimicrobial resistance surveillance is to assess temporal trends, evaluate intervention efforts, and ultimately improve clinical outcomes on a population-based level. While the treating clinician will need to consider the susceptibility results for each isolate obtained from the patient, and perhaps promptly change therapy in response to new developments (13,14), population-based recommendations for antimicrobial treatment are not likely to be altered on the basis of 3, 10, or even 30 days of surveillance data. Therefore, adopting either a 90- or 365-day analysis period for MRSA surveillance appears reasonable. At the statewide level in Hawaii, the results seen with the 90- and 365-day NCCLS methods were nearly identical, so either option would be appropriate.

Antibiograms should be individualized for specific clinical areas within an institution (e.g., ICUs) (15,16). In Hawaii, we observed differences in both MRSA rates as well as the magnitude of the effect of deduplication among the ICU, non-ICU, and outpatient settings; outpatient settings had the least effect. The greater effect of deduplication among inpatients may result from both the higher rate of MRSA among hospitalized patients and a greater likelihood of inpatients, especially those with MRSA, to have samples recultured compared to outpatients.

A major limitation of this analysis is that, because of medical confidentiality issues, we did not have patients' names. Since unique identities were determined by using an MRN generated by the treating facility or medical plan, a patient whose sample was cultured in >1 clinical facility during the analysis period might be miscounted as 2 persons. A related concern is that some laboratories did not provide the patient's MRN; for these patients, we had to use other information to generate an APID. While the APID process was not perfect and a small proportion of persons may have been misclassified as nonunique, a subanalysis that used only records with MRNs produced the same pattern of results as the analysis that used the larger dataset that incorporated the APID. This finding suggests that any misclassifications that resulted from using the APID did not substantially alter the relative effect of the different deduplication strategies we studied.

A second limitation is that we only assessed oxacillin resistance among S. aureus, so that conclusions regarding the effect of deduplication on other microorganisms must be made with caution. In addition, our analysis was not stratified by specific anatomic culture site (e.g., blood vs. skin); therefore, the effect of various deduplication strategies on isolates from specific culture sites could not be addressed.

A third limitation is that we did not include deduplication methods that take into account patterns of phenotypic resistance to multiple antimicrobial agents simultaneously (i.e., antibiotypes), as is practiced in some European countries (12). Tracking resistance by antibiotypes may show the actual number of infectious events or the selection of resistance occurring within the surveillance period. While these tasks are important for surveillance in some settings, the main purpose of our study was to evaluate a variety of uncomplicated strategies for generating communitywide susceptibility reports to specifically monitor MRSA trends and guide selection of empiric therapy. Nevertheless, further work is needed to examine the role of antibiotype surveillance in population-based antimicrobial surveillance systems.

We conclude that the NCCLS recommendation of including the first isolate of a given species per patient per analysis period, irrespective of body site, antimicrobial susceptibility profile, or other phenotypic characteristics, yielded results similar to other duplicate isolate removal methods and is straightforward in its implementation. Application of the techniques we examined had the same effect regardless of the institution. To aid our understanding of MRSA in both infection control practice and public health, we urge the widespread adoption of an industry standard. We suggest that adopting the 90- or 365-day NCCLS method would be appropriate, taking into account the goals of surveillance and the resources required.