All bacterial strains recovered from blood were identified from available microbiology department records of all 13 hospitals in San Francisco County from January 1, 1996, to March 31, 1999. For three hospitals (4,5,9), data from 1996 had been purged and were no longer available. For each isolate, data were collected on organism type and susceptibility pattern. Information was also obtained on the ward, age, and gender of the patient. Only the first positive blood culture of a given species was included for a single patient throughout the study period, regardless of susceptibility pattern. Cultures positive for S. epidermidis were considered representative of clinical bacteremia if at least two isolates with identical susceptibility patterns were obtained from a minimum of two separate sets of blood cultures. Each such set was considered a single bacteremic event. All other cultures positive for S. epidermidis were excluded, as were other common skin contaminants (e.g., Propionibacterium acnes, Peptostreptococcus, Corynebacterium).

All but one hospital used automated systems (VITEK [bioMerieux Vitek, Hazelwood, MO] or Microscan [Baxter Laboratories, West Sacramento, CA]) for the susceptibility testing of gram-negative bacteria. All hospitals used Kirby-Bauer disk-diffusion techniques for the evaluation of susceptibility profiles for Streptococcus pneumoniae and other streptococcal species according to National Committee for Clinical Laboratory Standards (NCCLS) guidelines. Kirby-Bauer results were supplemented by e-testing. One hospital (12) performed all susceptibility testing using Kirby-Bauer disk-diffusion techniques. All microbiology laboratories used MIC breakpoints established by the NCCLS. The annual number of blood culture sets processed by each microbiology laboratory was also obtained.

Data from five hospitals were obtained as text files or Microsoft Excel (Redmond, WA) files and subsequently imported into a Microsoft Access (Redmond, WA) database. Data from four hospitals were obtained in printed form, scanned as image files, converted into text files using TextBridge Pro 98 (ScanSoft Inc., Peabody, MA), and imported into the database. All scanned entries were verified for accuracy. Data from the remaining four hospitals were obtained from stored index cards, entered manually into the database, and verified for accurate entry.

We also obtained bed size and census data for all 13 hospitals in San Francisco County. Total hospital admissions were tabulated from quarterly administrative records from 1996 through 1998. For each hospital, average length of hospital stay was obtained from publicly available resources (25).

A total of 11,573 bacterial strains were recovered from blood cultures by the 13 hospitals. After excluding duplicate cultures, we had 8,072 remaining clinical isolates.

Information on hospital size, census, and blood culture volume is provided in Table 1. Despite being distinct and nonadjoining, hospitals 4 and 5 are reported together because of unified microbiology and administrative centers. Hospital 13 is a skilled nursing facility. On average, 74,600 sets of blood cultures were processed each year in San Francisco County; 9.9% of these were positive for bacterial species.

Figure 1

Figure 1. All bacterial species that were isolated from blood in >100 persons, January 1996 through March 1999, all hospitals in San Francisco County, California. Each bar is divided by yearly totals. Total...

Of the 8,072 isolates, Staphylococcus aureus (1,858), Escherichia coli (1,634), and S. pneumoniae (725) were the most common organisms. The numbers of S. aureus, Enterococcus faecalis, Bacteroides fragilis, E. coli, and Serratia marcescens increased annually during the 4-year period. Fourteen species had >100 isolates and were considered for further analysis (Figure 1). These 14 organisms accounted for 85% to 86% of all yearly totals.

Gram-Positive Organisms

Figure 2

Figure 2. Increase in percentage of Staphylococcus aureus isolates resistant to methicillin and increase in percent of Enterococcus faecium resistant to vancomycin on a yearly basis from 1996 through the first quarter of...

The proportion of MRSA and E. faecium resistant to vancomycin (VRE faecium) increased annually (Figure 2). Countywide, the proportion of MRSA isolates rose from 18.1% (1996) to 26.1% (1999) (p<0.001). In total, MRSA constituted 22.4% of all S. aureus isolates, including 19.6% of emergency department isolates and 15.9% of isolates in the outpatient setting (Table 2). Methicillin resistance in S. aureus isolates was <15% in patients <20 years of age and >20% in all other age groups, with the exception of 30- to 39-year-olds (17.6%) (Table 3). Proportional resistance did not vary by patient gender.

During the study period, 124 VRE isolates and 157 vancomycin-sensitive Enterococcus isolates were unspeciated. Among the speciated E. faecium isolates, the percentage resistant to vancomycin rose from 0% to 66.7% in the 4-year period (p<0.001). VRE faecium was most frequently isolated from inpatient adult wards, but six isolates were cultured from emergency department and outpatient settings. Over 66% of E. faecium isolates from skilled nursing facilities were resistant to vancomycin (Table 2). VRE isolates exceeded 40% in all age groups with the exception of ≤10-year-olds (22.2%) (Table 3). Of note, VRE faecium isolates showed increasing annual resistance to doxycycline (from 30% to 68%; p = 0.02).

Figure 3

Figure 3. Plot of the number of methicillin-resistant Staphylococcus aureus (MRSA) and vancomycin-resistant Enterococcus (VRE) isolates by hospital day of admission. An early peak is noted, corresponding to patients entering the hospital with...

Even after the skilled nursing facility (hospital 13) was excluded, the percentage of MRSA and VRE faecium isolates varied substantially among individual hospitals (MRSA 12.5% to 37.5%, VRE faecium 12.5% to 80.0%). There was no correlation between proportional resistance and number of ICU or total hospital beds, number of annual admissions, or average length of ICU or total hospital stay. However, for both organisms, there was increasing proportional resistance among adult wards in the following order: outpatient wards, emergency department, medical and surgical floors, medical and surgical ICUs, and skilled nursing facility wards. Notably, a substantial number of VRE and MRSA isolates were cultured within the first 24 to 48 hours of hospital admission (Figure 3).

Of the 678 isolates of S. pneumoniae, 13.6% were resistant to penicillin. Proportional resistance increased from 10.8% (21 isolates) in 1996 to 14.6% (27 isolates) in 1998, but this increase was not statistically significant. Proportional resistance was highest at the extremes of age (16.4% in patients >70 years of age and 19.3% in patients <10 years of age (Table 3).

Gram-Negative Organisms

Countywide, E. coli (1,634 isolates), Klebsiella pneumoniae (428 isolates), and Pseudomonas aeruginosa (260 isolates) were the most frequently isolated gram-negative bacilli. This was true in all inpatient and outpatient wards with the exception of skilled nursing facilities, where Proteus mirabilis was the most common gram-negative isolate after E. coli. Proportional resistance by ward for selective gram-negative organisms is shown in Table 4.

Among E. coli isolates, resistance to trimethoprim-sulfamethoxazole averaged 28% and resistance to ciprofloxacin averaged 3%. There was no resistance to fluoroquinolones among E. coli isolates from pediatric wards. Increasing annual resistance to ticarcillin-clavulanate was seen in both E. coli (6% to 16%, p = 0.03) and K. pneumoniae (0% to 18%, p = 0.007) isolates.

Figure 4

Figure 4. Yearly percent resistance to ciprofloxacin, ceftazidime, imipenem, and piperacillin in Pseudomonas aeruginosa isolates from blood. Increasing proportional resistance occurred in three of the four antibiotics commonly used to treat this organism....

P. aeruginosa isolates showed increasing annual countywide resistance to ciprofloxacin (7% to 21%, p = 0.005), ceftazidime (6% to 16%, p = 0.02), and imipenem (2% to 18%, p = 0.004) (Figure 4). Resistance to each of these three antibiotics exceeded 10% in adult ICU and adult medical and surgical wards. In fact, in these settings, ciprofloxacin resistance approached 20% countywide. No isolates resistant to ciprofloxacin were cultured from pediatric wards (Table 4). Resistance to gentamicin (15%) and piperacillin-tazobactam (12%) also increased but was not statistically significant.

There were 182 E. cloacae and 116 S. marcescens isolates from January 1996 through March 1999. Ciprofloxacin resistance averaged 4% among E. cloacae isolates and 6% among S. marcescens isolates. S. marcescens isolates also showed increasing annual proportional resistance to gentamicin (0% to 14%, p = 0.02) and piperacillin (4% to 29%, p = 0.01).

Resistance to ceftazidime, which can be predictive of inducible and extended-spectrum-beta-lactamases, was found in the following overall mean proportions in the study period: E. coli (1%), P. mirabilis (1%), K. pneumoniae (1%), S. marcescens (8%), P. aeruginosa (13%), and E. cloacae (39%). Only P. aeruginosa isolates demonstrated an increasing linear annual trend (p = 0.02).

San Francisco County has a population of approximately 735,000 and covers 46.7 square miles (26). It comprises multiple racial and ethnic groups (black 10.9%, Hispanic 13.9%, Asian 29.1%, and Native American 0.5%) and is served by 13 hospitals. We have shown that county surveillance of bacterial resistance is a useful addition to local hospital surveillance, particularly as antibiotic resistant bacteria increasingly spread from hospital to hospital and into the community at large.

Across the county, annual proportions of MRSA and VRE isolates significantly increased over the 4-year period. Pseudomonal strains resistant to fluoroquinolones, ceftazidime, or imipenem also increased annually. These data allow us to distinguish countywide outbreaks and trends from single-hospital changes in resistance patterns, and enable infection control efforts to expand or narrow to the appropriate scale. With awareness programs, county surveillance can broaden physicians’ knowledge of their hospital’s effects on the community, as well as the effects of neighboring hospitals on resistance patterns in their particular hospital.

Additionally, county surveillance that includes subcategorization of isolates by ward is invaluable in identifying patients at high risk and locations for transmission of resistant bacteria. Not surprisingly, we report our highest proportion of MRSA and VRE isolates from ICU and nursing home units. Nevertheless, ward variability across hospitals was substantial. Large interhospital differences can lead to further study of ward practices that foster or abate transmission. Awareness can prompt hospital infection control personnel to ensure well-described preventive measures such as swabbing and isolation precautions for VRE and MRSA in ICU settings (27–30) and nasopharyngeal swabbing and eradication of MRSA in hemodialysis wards (1,31). We also identify several MRSA and VRE isolates from outpatient and emergency department settings. Whether or not these represent true community-acquired strains or strains from patients recently released from hospital settings, they suggest that highly resistant bacterial outpatient infections and infectivity are increasing, a result consistent with recent studies (19,21,22).

We also evaluated whether the wide variability in the proportions of resistant bacteria among San Francisco hospitals was linked to hospital indices. In contrast to previous nationwide sampling studies, none of this variability was correlated with the number of hospital ICU beds (3), total beds (1,2,31), annual admissions (32), or annual mean length of stay. This may be due to our small number of hospitals, leading to limited power to detect such correlations. Alternatively, local community and hospital factors (e.g., increasing care of moderately ill patients at home [8], increasing home intravenous antibiotics [15], active transfer of patients between hospitals [33,34], and community-acquired resistant organisms) may now be diminishing the effect of hospital size, census and length of stay on proportional resistance.

There are many additional advantages to a countywide surveillance system for antimicrobial resistance. First, it can help reassess therapy. The 28% trimethoprim-sulfamethoxazole resistance in E. coli raises questions about the optimal empiric treatment of urinary tract infections in patients at high risk for bacteremia or urosepsis. Likewise, distinguishing resistance in the outpatient versus inpatient setting can guide empiric therapy in the appropriate setting.

Second, in studying a larger populace, we can obtain sufficient numbers to study uncommon organisms. Similarly, countywide surveillance provides a means to identify and confirm novel resistant pathogens. In our study, one MRSA isolate with intermediate resistance to vancomycin and three vancomycin-resistant S. epidermidis isolates were noted in microbiology laboratory reports. As with the organisms recently reported in the United States (35–37), these were reported from nontertiary hospitals. Nevertheless, our reports are unconfirmed and likely represent laboratory error. However, if surveillance could be expedited to real-time use, such reports could be investigated and confirmed rather than dismissed. At its worst, this suggests that highly resistant organisms may be escaping deserved attention and reaction.

Third, countywide surveillance engenders further hypotheses and research regarding interhospital and community transmission of resistant organisms. For example, our finding that 20% of emergency department S. aureus isolates are methicillin-resistant provides a flag to further study which county areas have the highest percentages of resistant S. aureus and which risk factors are involved (e.g., recent hospital admission [21,22], associated hemodialysis centers [8,38] or nursing homes [21,22], or intravenous drug use [39]). The finding that P. mirabilis bacteremia is more common in nursing home wards raises questions about preventing urinary tract infections in that setting. The relative lack of fluoroquinolone resistance in pediatric gram-negative isolates is likely due to the avoidance of fluoroquinolones in children because of potential detrimental cartilage effects. The study of fluoroquinolone-resistant organisms during the transition years from pediatric to adult medicine may provide insight into the quantity and duration of antibiotic needed to produce selection pressure, as well as the speed and durability of emerging resistance.

Our surveillance method had several limitations. Chart review would have been an invaluable addition in distinguishing between community-acquired and recent hospital or nursing home acquisition of resistant organisms. Second, we did not collect or confirm isolates; thus, although our results reflect microbiologic data actually presented to ordering physicians, they are subject to laboratory differences in speciation and susceptibility determination. Notably, not all organisms were fully speciated or tested against the antibiotics of interest. Fortunately, this occurred in a small proportion of bacteria. Third, we do not provide information on antibiotic use, which is known to be a major determinant of bacterial antibiotic resistance. Fourth, countywide trends can be driven by trends seen in the largest hospitals, particularly since smaller hospitals often lack sufficient numbers of isolates to make statistical analyses meaningful. This was notable for our data on proportional increases in VRE, which was largely driven by three hospitals. On the contrary, MRSA trends were not limited to a few hospitals, nor were they limited to the largest hospitals in the county. No hospital showed a significant decreasing trend in the proportion of MRSA isolates.

Areas for improvement include methods to computerize microbiologic data storage in a universal format. This would expedite surveillance and allow real-time collection and identification of unusually resistant organisms, as well as provide sentinel data regarding countywide outbreaks. In addition, linking of patient information to microbiologic data would have expedited acquisition of sex, gender, and ward information. Furthermore, despite NCCLS guidelines, a fair amount of variability exists in laboratory practices and susceptibility panels. Further standardization of these practices would help ensure the reliability of merging data among hospitals.

Without a doubt, the greatest utility of countywide surveillance lies in its ability to ask screening questions that prompt a more thorough investigation of specific hospitals, wards, or age groups at particular risk for acquiring or transmitting highly resistant organisms.

We have shown how several such questions were raised by our surveillance of bacteremias in San Francisco County and described its many advantages. We have further defined our limitations and difficulties in performing such surveillance, in the hope that this will be helpful to further similar surveillance efforts.