Using Hospital Antibiogram Data To Assess Regional Pneumococcal Resistance to Antibiotics

Cheryl R. Stein, David J. Weber, Meera Kelley


Emerging Infectious Diseases. 2003;9(2) 

In This Article


We examined the practicability of collecting hospital antibiogram data in North Carolina, a state with a population of 8,049,313 people and a land area of 48,711 square miles divided into 100 counties.[17] We also assessed pneumococcal susceptibility to multiple antimicrobial agents using the aggregated antibiogram data.

This study was conducted in North Carolina from April to September 2001. A study packet was mailed in April 2001 to the directors of clinical microbiology laboratories at all 114 North Carolina hospitals identified by the Centers for Medicare and Medicaid Services (CMS). Hospitals subsequently identifying themselves as specialty hospitals (e.g., psychiatric, drug treatment) were excluded from all analyses. The packet included a letter describing the project, a questionnaire on hospital characteristics and laboratory testing methods, a request for submission of S. pneumoniae antibiogram data for each year from 1996 to 2000, and a prepaid, preaddressed, return express mail envelope. The nature of the information provided on hospital antibiograms does not necessarily allow for identification of duplicate specimens from the same patient, differentiation of susceptibility results by source of specimen, or determination of conformance to National Committee for Clinical Laboratory Standards (NCCLS) guidelines for susceptibility cut points.

From the antibiograms, the numbers of pneumococcal isolates tested and numbers of isolates testing susceptible were added across all hospitals for each antimicrobial agent for each year of data to create statewide summary totals. Data for drugs that predictably elicit the same susceptibility result were combined: penicillin/oxacillin, cefotaxime/ceftriaxone, and levofloxacin/ofloxacin. Antibiogram data from each hospital for each year of data were assessed for inclusion. Data from antibiograms reporting 1) testing results cumulative over >1 year, 2) percentages of susceptible isolates without providing the total number of isolates tested, 3) more than one susceptibility value for the same drug for the same period, or 4) results by nursing unit or named patient were all excluded for the year and drug in question. We also excluded data for drugs other than penicillin if more isolates were tested for penicillin susceptibility than for the other drugs, and the antibiogram did not clearly indicate that the subgroup selected for additional testing was based on the source of the specimen (i.e., bloodstream). Testing only penicillin-nonsusceptible isolates for susceptibility to other drugs could yield misleading results because penicillin-resistant isolates are more likely to be resistant to other drugs as well.[8]

The aggregated statewide summary totals were used to calculate yearly susceptibility proportions for nine different antibiotics for the entire state. Nonsusceptible isolates encompassed those identified as either intermediate- or high-level resistant. Susceptibility proportions for penicillin were also stratified by geographic region of the state. Hospitals were categorized into three regions (west, central, east) by the county the respondent listed on the questionnaire. The Committee on the Protection of the Rights of Human Subjects, University of North Carolina School of Medicine, granted Institutional Review Board approval.

North Carolina's pneumococcal susceptibility pattern from 1997 through 2000 was compared to patterns shown by national surveillance systems tracking S. pneumoniae susceptibility. Data from published reports were included if they covered a period of no more than 12 months and identified the source of isolates as respiratory, invasive, or both. If the surveillance period overlapped two calendar years while covering one respiratory season, the data were classified by the latter year. For instance, isolates collected from October 1999 through April 2000 were labeled as year 2000 data.

We used the Cochran-Armitage trend test, which tests for trends in binomial proportions across levels of an ordinal covariate, to evaluate temporal patterns in the data. A two-sided p-value ≤0.05 was considered statistically significant. Trend tests were performed by using SAS version 8.1 (SAS Institute, Inc., Cary, NC). Exact binomial 95% confidence intervals were calculated for proportions by using Stata version 7.0 (Stata Corporation, College Station, TX).