The Centers for Disease Control (CDC) estimates that ∼2 million people develop antibiotic-resistant infections per year in the United States.Reference Baur, Gladstone and Burkert1 A major contributor to antibiotic resistance includes excessive use of antibiotics in both inpatient and outpatient settings.Reference Baur, Gladstone and Burkert1, Reference Dantes, Mu and Hicks2 The proportion of healthcare-associated infections (HAIs) caused by antibiotic-resistant pathogens can be viewed geospatially on a relatively new database hosted by CDC.3 We explored possible causes of state-to-state variability in prevalence of multidrug-resistant (MDR) Pseudomonas aeruginosa, extended-spectrum cephalosporin resistant (ESC-R) Escherichia coli, and methicillin-resistant Staphylococcus aureus (MRSA) HAIs. We included population characteristics considering historical associations with resistance, such as race and age.Reference Dantes, Mu and Belflower4
Methods
The primary source of data used in this study was the CDC Antibiotic Resistance Patient Safety Atlas,3 which contains state-level datasets on outpatient antibiotic prescriptions per 1,000 population stratified by age group and agent classification, prevalence of antibiotic stewardship programs in acute-care hospitals, and summary data on pathogens associated with HAIs tested and testing resistant to select antibiotics reported to the National Healthcare Safety Network (NHSN). State-level summary data on aspects of state HAI programs are also available in the database. The Centers for Medical and Medicare Services (CMS) Hospital Compare, Long-Term Hospital Compare, and Nursing Home Compare datasets were utilized for state-level characteristics on acute-care facilities (ACFs), long-term acute-care facilities (LTACFs), and skilled nursing facilities (SNFs).5–7 State population estimates were obtained from the US Census Bureau. Only data from 2011–2014 were included in the analysis.
Primary outcomes included state-level prevalence, defined as the total number of E. coli, S. aureus, or P. aeruginosa reported as resistant aggregated across all hospitals in the state divided by the total number of each pathogen reported. MDR P. aeruginosa was defined as P. aeruginosa testing intermediate or resistant to at least 1 drug in 3 of the following classes: extended-spectrum cephalosporin, fluoroquinolone, aminoglycoside, carbapenem, or piperacillins/tazobactam. ESC-R E. coli was defined as E. coli testing resistant to ceftriaxone, ceftazidime, cefepime, or cefotaxime. MRSA was defined as S. aureus testing resistant to methicillin, oxacillin, or cefoxitin.Reference Sievert, Ricks and Edwards8 States that had <10 isolates tested for each pathogen were excluded.
Predictors of resistance included inpatient antibiotic stewardship (measured as the percentage of hospitals in the state incorporating all 7 CDC Core Elements into their program) and state-level outpatient antibiotic prescription rates by antibiotic class (referred to as “outpatient prescribing”). Potential confounders included number and types of healthcare facilities, population characteristics (eg, population >65 years or African-American population greater than the national percentage), having a state-specific requirement to report HAI data to the NHSN or a state-program to validate NHSN data. Validation programs indicate that the state health department had access to NHSN data, performed an assessment of missing or implausible values on at least 6 months of the year’s data, and contacted identified facilities.
A Wilcoxon ranked-sum test was used to examine differences in distribution of HAI antibiotic resistance prevalence between states with and without NHSN mandate and validation standards. Multivariable logistic regression with events/trials and forward selection method was utilized to fit models for each outcome and to identify confounders to retain for further evaluation. Events were the number of isolates resistant and trials were the number of isolates tested; this method accounts for the sizeable differences in the number of isolates tested between US states. A Wald test was used to determine inclusion in the model. Significant confounders were then included to evaluate the Pearson partial correlation coefficients and P values of the main predictors (ie, stewardship and outpatient prescribing), this method was chosen to better reflect the state-level summary metrics as ecologic data. All statistical analyses were performed using SAS version 9.4 software (SAS Institute, Cary, NC).
Results
Across all states, 5,322 MDR isolates were MRSA (45% of S. aureus tested), 2,333 were ESC-R E. coli (12.7% of tested), and 1,066 were MDR P. aeruginosa (12.8% of tested) (Table 1). There was no statistical difference in the median resistance rates between states based on presence of NHSN mandates or validation programs (data not shown). Potential confounders of isolates found to be resistant were identified (for each resistant phenotype) as follows: proportion of the state’s population being >65 years (MDR P. aeruginosa, ESC-R E. coli), large African-American population (ESC-R E. coli and MRSA), having a state-validation program (MDR P. aeruginosa, MRSA), number of LTACFs (MDR P. aeruginosa) or number of SNF bed-days (MRSA), and the state’s population (ESC-R E.coli). Due to statistical significance, an interaction term between outpatient fluoroquinolone prescribing and largeAfrican-American population was included in the model for MRSA prevalence.
Adjusting for the observed confounding, the partial Pearson correlation coefficients of the associations between resistance and either inpatient stewardship or outpatient prescribing were calculated (Table 2). Outpatient fluoroquinolone and cephalosporin prescribing were positively and significantly correlated with ESC-R E. coli prevalence. For MRSA, only outpatient fluoroquinolone prescribing was positively and significantly correlated prevalence; however, the magnitude of the correlation was slightly elevated in states with a large population of African Americans. Correlation between outpatient prescribing and inpatient MDR P. aeruginosa prevalence was only of appreciable magnitude but not significant (P = 0.09).
Table 1. Demographic Characteristics of Key State-Level Antibiotic Resistance Indicators, 2011–2014
Note. MRSA, methicillin-resistant S. aureus; CDC, Centers for Disease Control and Prevention; MDR, multidrug resistant; IQR, interquartile range.
a All 50 states plus the District of Columbia had data on all the predictors; however, some states had <10 isolates tested and thus were dropped from specific analysis.
Table 2. Pearson Partial Correlation Coefficients (R2) for Intensity of Antibiotic Stewardship and Outpatient Prescribing Rates on MRSA, ESC-R E.coli, and MDR P. aeruginosa
Note. MRSA, methicillin-resistant Staphylococcus aureus; ESC-R E. coli, extended-spectrum cephalosporin resistant Escherichia coli; MDR, multidrug resistant.
a Adjusted for number of SNF bed days, % of the population that was African American, validation, and mandate.
b Adjusted for % of the population that was African American, total population, and percent of the population ≥65 years of age.
c Adjusted for % of the population ≥65 years of age, number of long-term acute-care facilities, and validation.
Discussion
Intensity of antibiotic stewardship did not explain geographic variability in inpatient MDR P. aeruginosa, ESC-R E. coli, or MRSA prevalence. Conversely, outpatient fluoroquinolone and cephalosporin prescribing explained some geographic variability in ESC-R E. coli prevalence. Outpatient fluoroquinolone prescribing also explained some geographic variability in MRSA prevalence that was slightly elevated in states with a larger population of African Americans. These observations suggest that outpatient prescribing frequency may have a direct impact on the resistance phenotypes of hospital-onset infection pathogensReference Sievert, Ricks and Edwards8 or have indirect effects such as promotion or alteration of the gastrointestinal tract flora among outpatients. Other explanations may include outpatient prescription data reflecting recently discharged patients or serving as a proxy measure for how antibiotics are prescribed among inpatients.
We did not observe any association between stewardship intensity among inpatient setting and inpatient HAI resistance metrics. One prominent reason is that the NHSN survey data available only reported frequency of hospitals that incorporated all 7 of the CDC Core Elements, whereas hospitals that incorporate fewer CDC Core Elements may contribute to reducing antibiotic-resistant phenotype prevalence but are not included in the percentage of antibiotic stewardship programs in place. Additionally, even those with all 7 components may vary in effectiveness, and no validation of inpatient stewardship data is available. Additional reasons for this observation may include lack of any time-series component preventing any temporal relationship between timing of stewardship implementation and prevalence of the target antibiotic-resistant phenotypes or lack of infection control assessment. Notably, the resistance data were limited to only 3 types of NHSN-defined HAIs; therefore, they do not reflect the majority of HAIs nor community or nursing-home–onset infections. Regardless, these data do suggest that stewardship in the outpatient setting could have an impact on inpatient resistance phenotypes and should be studied in more detail. In conclusion, we observed interesting associations with some biologic plausibility using publicly available data summarized at the state-level from both CMS5–7 and CDC.3 More granular data would allow for more conclusive analysis in the future should it become publicly available.
Author ORCIDs
Julianne N. Kubes, 0000-0002-6717-3104
Acknowledgments
We acknowledge Lindsey Weiner and Michael Garber for consultation and advisement on this article. We also acknowledge the Rollins School of Public Health for providing the statistical software used in this article.
Financial support
No financial support was received.
Conflicts of interest
S.K.F. reports that his institution has received research grants from Pfizer. All other authors report no conflicts of interest relevant to this article.
