For the survey of 1998 susceptibility results, we received responses from 47 of 52 hospitals in the DVCCN. However, in two instances, the results from multiple hospitals were combined at a single laboratory and could not be separated for reporting purposes. Of the 45 laboratories reporting results, 33 laboratories reported >10 pneumococcal blood isolates during 1998. Among the 33 sites, the median number of pneumococcal blood isolates during 1998 was 19 (range 10-92).

Figure 1 displays the frequency distribution of the proportion of isolates with nonsusceptibility to penicillin as reported by each hospital in 1998. The proportion of nonsusceptible isolates ranged from 0% to 67%, with a mean proportion of 21%. Sixty-one percent of hospitals were >5% above or below the mean. If we only included hospitals with >20 pneumococcal isolates in 1998, a total of 16 hospitals would be included in the analysis; the proportion of nonsusceptible isolates ranged from 0% to 36%, with a mean proportion of 20%. Fifty-six percent of hospitals were >5% above or below the mean proportion of nonsusceptible isolates.

For the 1999 and 2000 survey results, we received responses from 31 hospital laboratories. Of these, 23 hospital laboratories reported >10 pneumococcal isolates in 1 of the 2 study years. The mean proportion of isolates with reduced susceptibility to penicillin was 19% in 1999 and 24% in 2000. Fifty-three percent of sites were >5% above or below the mean in 1999, and 65% of sites were >5% above or below the mean in 2000.

The Table summarizes the relationship between each hospital’s annual proportion of nonsusceptible pneumococcal blood isolates during the 3 years of the study. For hospitals reporting >10 isolates in 1998, 1999, and 2000 (N = 15), the proportion of nonsusceptible isolates at each hospital was poorly correlated between any 2 years. For 1998 to 1999, the Spearman correlation coefficient was –0.07 (p = 0.82), for 1999 to 2000, the Spearman correlation coefficient was 0.28 (p = 0.32), and for 1998 to 2000, the Spearman correlation coefficient was –0.03 (p = 0.91). Across all 3 years, the overall intraclass correlation coefficient was 0.18.

Location of hospital by county did not correlate with the proportion of pneumococcal blood isolates nonsusceptible to penicillin (p = 0.45, 0.37, 0.08 in 1998, 1999, and 2000, respectively). Figure 2 is a proportional symbol map with each hospital location represented by a circle whose radius corresponds to the proportion of nonsusceptible pneumococcal blood isolates at each hospital in 1998. Overall, Figure 2 shows geographic clustering of the 33 hospitals reporting >10 pneumococcal blood isolates in 1998, 14 of which are gathered in Philadelphia, the most heavily populated county in our five-county study area. No geographic clustering in the proportion of nonsusceptible pneumococci isolated at each hospital was evident from our data. This finding is supported by a Moran correlation coefficient of –0.01, demonstrating a lack of statistically significant spatial autocorrelation in the proportion of nonsusceptible hospital pneumococci at hospitals across the five-county region (p = 0.80).

Pneumococcal drug resistance has increased at a fast rate. Most resistant isolates now demonstrate reduced susceptibility to multiple antimicrobial drug classes, including β-lactams, macrolides, and tetracyclines. However, substantial geographic variability in the proportion of pneumococci resistant to different antimicrobial drugs at the international and national levels suggests that the impact of resistance on empirical treatment decisions may vary across regions. Indeed, at least one guideline recommends that the selection of empirical therapy for outpatients with community-acquired pneumonia should be influenced by regional antimicrobial drug susceptibility patterns for S. pneumoniae (2).

For most physicians, the obvious source of information on community-acquired pneumonia is the microbiologic susceptibility results from their local hospital laboratory. Hospitals frequently provide this information in the form of antibiograms, specifically designed to aid in antimicrobial drug selection. The validity of this information is largely unknown. This study demonstrates that substantial variability exists in hospital-reported rates of pneumococcal drug resistance over a small geographic region and that the rates at each site are poorly correlated over time. In addition, using a variety of approaches, we were unable to demonstrate substantial geographic clustering of the data at the individual hospital level, suggesting that the individual hospital rates poorly reflect an underlying rate of resistance in the community.

Other studies have demonstrated similar heterogeneity in the proportion of drug-resistant pneumococci reported by hospitals over small geographic areas. The Centers for Disease Control and Prevention reported levels of pneumococcal resistance to penicillin among invasive isolates from all hospitals within the seven regions in the United States that comprise their Active Bacterial Core Surveillance program. The proportion of penicillin-nonsusceptible S. pneumoniae ranged from 15.3% to 38.3% across the seven regions. The proportion of hospitals within each region that were >5% below or above the average for the region ranged from 35% to 76%. No hospital characteristics, including proportion of isolates from children or from black patients, predicted deviation from the regional average (11). More recently, 16 hospitals in Brooklyn, New York, participated in boroughwide surveillance for S. pneumoniae in 1997 and 1999. Aggregating the hospitals into the western and eastern ends of the borough demonstrated a significant difference in the proportion of penicillin-susceptible isolates, 57% versus 75% (p = 0.046) (15).

Our hypothesis that individual hospital rates of drug-resistant S. pneumoniae should demonstrate geographic clustering is based on the assumption that rates of resistance should reflect the underlying distribution of drug resistance in the source community. Prior research has demonstrated that patients with community-acquired pneumonia are hospitalized on average <5 miles from their place of residence (16). Thus, hospitals within the same geographic region should admit patients with similar underlying rates of drug resistance. Moreover, if the pneumococcal resistance rates reported by individual hospitals should be used to guide physician antibiotic prescribing decisions for patients with community-acquired pneumococcal infections, hospitals serving similar communities should report similar rates of resistance.

Beyond reflecting true variability in the underlying community levels of drug resistance, hospital-level variation in the proportion of penicillin-susceptible pneumococcal isolates reflects at least four phenomena. First, chance error can create significant variability in hospital-level results particularly since many hospitals have only a small number of invasive pneumococcal isolates per year. We excluded hospitals with <10 isolates in each study year from our analyses to reduce the role of chance in our findings, but the relatively small number of isolates at each site undoubtedly contributed to the significant year-to-year variability and lack of geographic clustering. However, a minimum of 10 isolates per year is the NCCLS-recommended minimum number of isolates for reporting cumulative antimicrobial susceptibility results for any species in any reporting period (8).

Second, variation in testing strategies at individual hospitals may create bias in the proportion of isolates identified as nonsusceptible if the thresholds for sending microbiologic tests vary according to characteristics that are likely to influence the probability of a drug-resistant infection. While this may be an important determinant of variability in drug-susceptibility rates among clinical isolates from respiratory and sinus samples, we believe that it is less likely to explain variability in susceptibility rates among blood isolates since thresholds for sending blood cultures are more standardized across clinical practices, particularly in the management of patients hospitalized with community-acquired pneumonia.

Third, hospitals may differ in their methodologic approach and quality assurance for pneumococcal susceptibility testing. However, a study based on a nationwide College of American Pathologists proficiency survey of hospital laboratories demonstrated that, while the use of specific susceptibility tests varies substantially, few major interpretive errors occurred in assigning the susceptibility to standardized pneumococcal isolates (17). In addition, prior studies have found that the results of susceptibility data generated at local hospital facilities provide a reasonable surrogate for susceptibility data generated at centralized facilities (18).

Finally, geographic proximity of hospitals does not ensure that the patient populations served by those hospitals come from the same communities. Indeed, substantial literature has established important variation in hospital referral patterns that creates diversity in the source of patient populations among geographically proximate hospitals. Future research will need to consider the role of referral patterns in explaining some of the hospital-level variability observed in this and other studies.

One limitation of this study is our focus on a single geographic region in eastern Pennsylvania. Patterns and determinants of hospital-level variation in pneumococcal drug susceptibility may produce different results in other regions. In addition, because we focused on hospitals providing adult care, we cannot generalize our results to pediatric hospitals.

Regardless of whether individual hospitals provide valid information on local levels of pneumococcal susceptibility, variability in these levels is only clinically meaningful if the range of susceptibility crosses some threshold for the empirical choice of specific antimicrobial drugs. Currently in the United States, empirical use of penicillins for the treatment of community-acquired pneumonia is uncommon relative to the use of fluoroquinolones and macrolides (19). However, as resistance to these classes of antimicrobial drugs continues to grow, physicians will need to determine appropriate thresholds for switching to yet newer antimicrobial agents. In these settings, valid information on local rates of pneumococcal drug susceptibility will become increasingly important. Ideally, data should be derived from large samples reflective of the region for which the empirical recommendations apply. Continued research is needed to determine whether susceptibility data provided by local hospital microbiology laboratories can ever serve this purpose.