Antibiotic Use in Small Community Hospitals

Abstract

Background. Antibiotic use and misuse is driving drug resistance. Much of US healthcare takes place in small community hospitals (SCHs); 70% of all US hospitals have <200 beds. Antibiotic use in SCHs is poorly described. We evaluated antibiotic use using data from the National Healthcare and Safety Network antimicrobial use option from the Centers for Disease Control and Prevention.

Methods. We used Intermountain Healthcare's monthly antibiotic use reports for 19 hospitals from 2011 to 2013. Hospital care units were categorized as intensive care, medical/surgical, pediatric, or miscellaneous. Antibiotics were categorized based on spectrum of coverage. Antibiotic use rates, expressed as days of therapy per 1000 patient-days (DOT/1000PD), were calculated for each SCH and compared with rates in large community hospitals (LCHs). Negative-binomial regression was used to relate antibiotic use to predictor variables.

Results. Total antibiotic use rates varied widely across the 15 SCHs (median, 436 DOT/1000PD; range, 134–671 DOT/1000PD) and were similar to rates in 4 LCHs (509 DOT/1000PD; 406–597 DOT/1000PD). The proportion of patient-days spent in the respective unit types varied substantially within SCHs and had a large impact on facility-level rates. Broad-spectrum antibiotics accounted for 26% of use in SCHs (range, 8%–36%), similar to the proportion in LCHs (32%; range, 26%–37%). Case mix index, proportion of patient-days in specific unit types, and season were significant predictors of antibiotic use.

Conclusions. There is substantial variation in patterns of antibiotic use among SCHs. Overall usage in SCHs is similar to usage in LCHs. Small hospitals need to become a focus of stewardship efforts.

antibiotic stewardship, benchmarking, small community hospital, antibiotic use

Antibiotic use and misuse is driving the increase in drug resistance and has been declared a priority at all levels of the United States government. To improve antibiotic use and develop benchmarks, accurate and standardized measurement of use are needed. A significant portion of US healthcare takes place in small community hospitals (SCHs). In 2012, 72.4% of all US hospitals had <200 beds and the mean bed size in all hospitals was 160 [1]. However, our understanding of antibiotic use in these facilities is extremely limited. Most data on antibiotic use and antibiotic stewardship (AS) are from large academic medical centers [2, 3].

Monitoring antibiotic use is important for AS initiatives, and is listed on the Centers for Disease Control and Prevention (CDC) Checklist for Core Elements of Hospital Antibiotic Stewardship Programs. The 2014 National Healthcare and Safety Network (NHSN) annual survey reported that increasing hospital size is a major predictor of a hospital meeting all of the CDC's core elements of AS [4]. Unfortunately, AS programs (ASPs) and infectious diseases consultation are limited in many small hospitals [5–10]. Describing antibiotic use in SCHs is the first step toward a better understanding of how to implement effective ASPs in these settings. Evaluating antibiotic usage patterns in these facilities is a high priority, given that SCHs constitute the majority of acute care hospitals and national AS requirements are forthcoming [11–13].

To measure inpatient antibiotic use, the CDC launched the NHSN antimicrobial use (AU) option in 2011. The NHSN AU is a secure, internet-based surveillance system used to collect aggregated, monthly inpatient antibiotic use data [14]. Although submission to the NHSN AU system is voluntary, participation in the program will undoubtedly grow with the National Action Plan for Combating Antibiotic-Resistant Bacteria, which outlined a goal of “routine reporting of antibiotic use and resistance data to the NHSN” by 2020 [11]. The objective of this study was to describe antibiotic use and variability within SCHs, using large community hospitals (LCHs) for comparison.

METHODS

Intermountain Healthcare

Intermountain Healthcare (hereafter Intermountain) is a nonprofit, integrated, healthcare network in Utah and Idaho and has a long history in antibiotic use measurement [15, 16] and electronic decision _support [17]. Intermountain operates 15 SCHs (licensed beds, 14–146); 4 LCHs (licensed beds, 245–472); 1 free-standing children's hospital; and 2 specialty hospitals (an orthopedic hospital and an advanced community hospital focused on bone marrow transplantation). Three of 15 SCHs have designated pediatric units, and 7 SCHs have active intensive care units (ICUs). Data from the children's hospital and 2 specialty hospitals were not included in these analyses.

NHSN AU Data

Intermountain facilities use a locally developed electronic medical record system. Antibiotic use data has been submitted to the NHSN AU module since January 2011. As described and validated elsewhere [16, 18], antibiotic usage data files are generated monthly and manually uploaded to the NHSN AU module. The data include patient care location and facility-wide antibiotic use expressed in days of therapy per 1000 patient-days present (abbreviated DOT/1000PD). Use of individual antibiotics and classes of antibiotics can be expressed per unit and per facility. The data are aggregated by month for each patient care location and facility wide. Patient-level data are not submitted. Days of therapy data (numerator) are obtained from the electronic medication administration records and patient-day data (denominator) are obtained from patient room-trace data [16].

Antibiotic Spectrum Categorization

We categorized antibiotic agents into 5 groups based on antibiotic spectrum and ability to treat multidrug-resistant organisms (MDROs) (Figure 1). Category 1 antibiotics are narrower-spectrum agents, and category 5 antibiotics are the broadest-spectrum antibiotics or associated with MDROs. Categories 4 and 5 were classified as broad-spectrum antibiotics.

Figure 1.

Antibiotic classification schema. From left to right, category 1 through 5 antibiotics.

To measure case mix, all inpatient encounters (not limited to Medicare beneficiaries) within Intermountain hospitals are assigned to an all-patient refined-diagnosis related group (APR-DRG). The APR-DRGs are then assigned weights based on charge data collected. An average encounter is assigned a value of 1. For example, a normal vaginal delivery is assigned a weight of 0.34, and coronary artery bypass graft surgery, a weight of 2.64. The case mix index (CMI) is calculated by summing the APR-DRG weights of all the encounters and dividing the sum by the number of encounters. Monthly CMIs as well as a 3-year mean (2011–2013) CMI were calculated for each SCH and LCH within Intermountain.

Statistical Analysis

An antibiotic day of therapy is defined as any amount of a specific antibiotic agent administered in a calendar day. The total number of patient-days during a calendar month is defined as the aggregate number of patients in a patient care location or facility anytime throughout a day, summed over all patients and all days during a calendar month. The NHSN-defined patient care locations of the 15 SCHs and 4 LCHs were consolidated into 4 nonoverlapping unit-type categories: medical/surgical unit, ICU, pediatric medical/surgical unit, and miscellaneous unit. The miscellaneous category was created to include patient care locations with historically low antibiotic usage (eg, well-baby nursery, maternity, labor and delivery, and psychiatry). Using NHSN AU data from January 2011 through December 2013, we calculated monthly and 3-year antibiotic use rates for each facility, unit type, and antibiotic category. Pearson correlations were used to describe the association across facilities between overall 3-year antibiotic use rates and the proportion of broad-spectrum antibiotic use.

Generalized linear fixed and mixed-effect regression models were fit to relate rates of antibiotic use to predictor variables of interest. Each model assumed a negative binomial distribution for the number of antibiotic days over the designated period of the analysis to account for overdispersion resulting from correlated antibiotic use within the same patients on different days and for other sources of clustering. We used the total number of patient-days as an offset and applied a logarithmic link function to relate the antibiotic use rate to the predictor variables. We used model-based standard errors to determine confidence intervals and P values in our primary analyses but considered robust empirical standard errors in our sensitivity analyses. We first used fixed-effect models, with antibiotic use over 1-month intervals within each facility as the unit of analysis, to describe mean rates of antibiotic use and overall trends during the 3-year period within the specific large and small hospitals included in the study. These models were fit separately for the small and large hospitals, with equal weight applied to each hospital, and included the 36 study months and the interaction between study month and facility as predictor variables.

We also considered a mixed-effect model of antibiotic use over 1-month intervals within each floor type within each of the 15 SCHs as the unit of analysis. This model related antibiotic use to a more comprehensive set of predictor variables, including unit types, CMI, and year. The model also included sine and cosine terms to account for seasonal trend. Hospital and time in months were treated as random effects so that statistical inferences could be extended beyond the specific SCHs in the study. We estimated the remaining coefficient of variation for the 3-year antibiotic use rates across the 15 SCHS after accounting for the predictor variables by appropriately transforming the estimated variance of the random effect for center assuming a log-normal distribution for the antibiotic use rates among the 15 SCHs.

RESULTS

Characteristics of SCHs and LCHs within the Intermountain network are presented in Table 1. The proportion of patient-days spent in patient Care areas designated as miscellaneous ranged from 0% to 64%. Three-year mean antibiotic use rates by facility for the 15 SCHs and 4 LCHs are shown in Figure 2A. Among the SCHs, antibiotic use varied greatly (median, 436 DOT/1000PD; range, 134–671 DOT/1000PD) and was similar to that in LCHs (509 DOT/1000PD; 406–597 DOT/1000PD). Mean CMIs varied between hospitals, ranging from 0.29 to 1.3 (Figure 2A). SCHs had a lower CMI (median, 1.05; interquartile range [IQR], 1.0–1.13) than LCHs (median, 1.59; IQR, 1.55–1.62). Antibiotic use rates in the 15 SCHs overlap with those in the LCHs despite significantly lower CMI. For comparison, hospital L2 (in Table 1 and Figure 2) is a LCH with an academic affiliation and level 1 trauma services, 6 ICUs, and an active solid organ transplant program.

Table 1.

Description of Intermountain's Small and Large Community Hospitals in 2013^a

Hospital	Licensed Beds, No	Total Patient-Days, No.	Patient-Days, No. (%)
			Medical/Surgical	ICU	Pediatric	Miscellaneousb
Small community hospitals
S11	146	44 174	13 976 (32)	3351 (8)	4619 (10)	22 228 (50)
S14	97	33 319	8082 (24)	802 (2)	3000 (10)	21 435 (64)
S6	89	27 491	11 363 (41)	1895 (7)	4677 (17)	9556 (35)
S10	71	23 448	8913 (38)	933 (4)	…	13 602 (58)
S1	48	11 642	7765 (67)	1531 (13)	…	2346 (20)
S15	24	10 118	6091 (60)	…	…	4027 (40)
S5	25	8764	5338 (61)	1472 (17)	…	1954 (22)
S4	30	7312	6045 (83)	408 (6)	…	859 (11)
S8	42	4800	3892 (81)	…	…	908 (19)
S12	19	2976	2561 (86)	…	…	415 (14)
S9	18	2162	1824 (84)	…	…	338 (16)
S13	19	1959	1959 (100)	…	…
S3	16	1899	1576 (83)	…	…	323 (17)
S2	14	1505	1505 (100)	…	…	…
S7	18	1275	1275 (100)	…	…	…
Large community hospitals
L2	472	182 298	99 177 (49)	44 226 (22)	…	58 040 (29)
L3	395	118 587	60 330 (51)	27 320 (23)	5130 (4)	25 807 (22)
L4	309	112 303	50 259 (45)	14 973 (13)	2703 (2)	44 368 (40)
L1	245	84 819	54 053 (64)	10 127 (12)	1834 (2)	18 805 (22)

Abbreviation: ICU, intensive care unit.

^a Small community hospitals had <200 beds; large community hospitals, 200–500 beds.

^b Miscellaneous bed days include labor and delivery, nursery, maternity, and psychiatry.

Figure 2.

A, Three-year mean antibiotic use rates (bars) and mean case mix index (pyramids) for Intermountain's 15 small community hospitals (SCHs) and 4 large community hospitals (LCHs); S and L in hospital numbers denote SCH and LCH status, respectively. DOT/1000PD, days of therapy per 1000 patient-days. B, Three-year mean antibiotic use rates and 95% confidence intervals for Intermountain's 15 SCHs and 4 LCHs, excluding miscellaneous units (labor and delivery, nursery, maternity, and psychiatry).

Given the large contribution of patient-days in miscellaneous units in some hospitals (Tables 1 and 2), antibiotic use rates were recalculated after removing these data from the rate calculations (Figure 2B). Exclusion of miscellaneous units from the rate calculation results in significant changes in the ranking of hospitals by antibiotic use. The CMI is calculated using hospital-level data; therefore, CMI cannot be calculated for the data in Figure 2B.

Table 2.

Three-Year (2011–2013) Median Antibiotic Use Rates for Intermountain's 15 Small Community Hospitals, by Unit Type and Antibiotic Category

Antibiotic Category	Antibiotic Use Rate by Unit Type, Median (IQR), DOT/1000PD
	ICU	Adult Medical-Surgical	Pediatric Medical-Surgical	Miscellaneousa	Overall
Category 1	87 (42–125)	137 (83–200)	221 (172–293)	36 (12–64)	91 (38–167)
Category 2	133 (85–183)	99 (65–134)	26 (13–49)	1 (0–12)	59 (8–119)
Category 3	124 (94–180)	113 (69–156)	70 (25–118)	0 (0–4)	74 (3–130)
Category 4	252 (175–333)	137 (84–181)	145 (63–194)	5 (0–18)	103 (13–182)
Category 5	191 (145–273)	54 (24–88)	6 (0–19)	0 (0–0)	24 (0–91)
Overall	881 (755–1041)	607 (452–715)	491 (426–582)	54 (24–108)	500 (111–715)

Abbreviations: DOT/1000PD, days of therapy per 1000 patient-days; ICU, intensive care unit; IQR, interquartile range.

^a Miscellaneous units included labor and delivery, nursery, maternity, and psychiatry units.

The antibiotic use rates by antibiotic category varied by patient care area for the 15 SCHs (Table 2). ICUs had the highest total antibiotic use, with a median of 881 DOT/1000PD (IQR, 755–1041) while also having the highest rates of broad-spectrum usage. Miscellaneous units had lowest antibiotic use rates, with a median of 54 DOT/1000PD (IQR, 24–108) and also the lowest rates of broad-spectrum usage. Within medical-surgical units, categories 1 and 4 antibiotics predominated, and use of broad-spectrum antibiotics was less than that observed in the ICUs. With miscellaneous units excluded, the proportion of total antibiotic use represented by broad-spectrum antibiotics varied significantly in SCHs, ranging from 8% to 36% of total use (Figure 3). Broad-speCtrum use correlated moderately with total antibiotic use (Pearson correlation, 0.30).

Figure 3.

Stacked bar graph of antibiotic spectrum prescribed by hospital in small and large community hospitals, dichotomized on narrow-spectrum (categories 1–3; dark gray) and broad-spectrum/multidrug-resistant organism (MDRO)–active (categories 4–5; light gray) antibiotics, with miscellaneous unit data excluded. S and L in hospital numbers denote small and large community hospital status, respectively.

Temporal trends in antibiotic use rates for the 15 SCHs and 4 LCHs over the 3-year study period were examined (Supplementary Figure 1). The aggregated antibiotic use rate was stable for the 15 SCHs this period, with a relative rate difference per year of 1.48% (95% confidence interval, −.39% to 3.39%). In contrast, antibiotic use decreased for LCHs during the same time period with a relative rate difference per year of −3.54% (95% confidence interval: −4.44% to −2.63%). Of note, none of the 15 SCHs had active ASPs during this period, whereas 3 of the 4 LCHs had active, infectious diseases–led ASPs.

Table 3 presents the results of the multivariable mixed-effeCt analysis relating antibiotic usage to predictor variables, based on NHSN AU data and CMI in the 15 SCHs. Greater antibiotic usage was associated with ICU units, higher CMI, and winter season. Similar results were obtained in a sensitivity analysis using robust empirical standard errors (data not shown). The standard deviation (SD) of the rates of antibiotic use across the 15 SCHs was 43.1% as large as the mean rate of antibiotic use. Adjustment for floor type reduced the SD from 43.1% to 26.6%, corresponding to a 38.2% relative reduction. Adjustment for both floor type and CMI reduced the SD to 23.6%, corresponding to a 45.2% relative reduction. Adjustment for all of the predictors in the model reduced the SD between facilities to 22.5%, corresponding to a 47.9% relative reduction.

Table 3.

Multivariable Analysis Relating Antibiotic Use to Proportion of Patient-Days Spent in Specific Care Units, Year, and Facility Mean CMI Among Intermountain's Small Community Hospitals

Effect	Antibiotic Use Rate Ratio (95% CI)	P Value
Adult medical/surgical unit vs ICU	0.775 (.725– .828)	<.001
Pediatric medical/surgical unit vs ICU	0.666 (.608–.729)	<.001
Miscellaneous unita vs ICU	0.095 (.089–.102)	<.001
Year 2012 vs 2011	1.029 (.957–1.108)	.44
Year 2013 vs 2011	1.063 (.948–1.191)	.30
Facility average CMI (per 1-unit increase)	2.801 (1.916–4.095)	<.001
Amplitude of seasonal effect	1.054 (1.021–1.087)	.001

Abbreviations: CI, confidence interval; CMI, case mix index; ICU, intensive care unit.

^a Miscellaneous units labor and delivery, nursery, maternity, and psychiatry units.

DISCUSSION

Antibiotic use in SCHs has not been well described. To our knowledge, this is the first in-depth analysis of antibiotic use rates and selection patterns among SCHs in the United States. Using data contained within the NHSN AU module, we demonstrated that SCHs have similar antibiotic prescribing rates and similar antibiotic prescribing patterns when compared with LCHs, despite a less complex patient population. In addition, there is considerable variability in total antibiotic usage and the spectrum of antibiotics prescribed within SCHs. With the majority of US hospitals having <200 beds [1] and many without ASPs or infectious diseases _support, helping SCHs improve antibiotic prescribing must become a priority for the infectious diseases community.

Comparing the antibiotic use rates from Intermountain's SCHs with those from other SCHs is difficult owing to the paucity of data from SCHs and lack of consensus on benchmarking techniques. Few studies have evaluated antibiotic use within healthcare networks [2, 19–22], and only 2 have published on the topic in the last 15 years [2, 19]. In 1 of these studies, antibiotic usage data were collected via administrative billing and claims data for 130 hospitals (mean number of beds, 288; range, 20–1020) [19]. The total mean (SD) antibiotic drug use was 776 (120) DOT/1000PD for the 50 most commonly used antibiotics [19]. A subsequent study evaluated facility-level antibiotic use in 70 US academic medical centers in the University HealthSystem Consortium, using inpatient billing files. The mean bed size for the 70 academic medical centers was 544 (range, 185–1156), with a mean CMI for adult patients of 1.62 (.06–1.99). The mean (SD) total antibiotic drug use was 839 (106) DOT/1000PD [2]. Both of the studies above describe facility antibiotic usage rates and did not account for the proportion of patient-days in units with traditionally low antibiotic use. The majority of facilities included in these studies were academic medical centers, few had <200 beds, and they used a different method for capturing antibiotic use. Given these limitations, comparing the facility-level antibiotic use in these studies to our network of SCHs is difficult and could result in invalid conclusions.

Hospital length of stay, geography, teaching status, infectious diseases–related International Classification of Diseases, Ninth Revision, codes, CMI, number of operations performed, and number of bacteremia episodes have all been evaluated as potential factors predictive of antibiotic use in risk-adjusted models [23–30]. Identifying factors that are predictive of antibiotic use and accounting for them in risk-adjusted models allows for more accurate comparison of antibiotic prescribing practices across facilities and/or care units (ie, benchmarking). Several studies have assessed the impact of facility size on antibiotic consumption [23–29, 31]. The largest study was a 1-day antibiotic use point prevalence survey conducted by the CDC that evaluated 11 282 patients from 183 hospitals and 10 states. Among the 4073 patients evaluated in small hospitals (<150 beds), 52.8% received ≥1 antibiotic drug, compared with 48.6% of patients in hospitals with 150–399 beds and 47.7% in hospitals with ≥400 beds (P < .001; χ²) [31]. Only 1 study has found that antibiotic use increased with increasing hospital size [28].

In our model, we limited the variables in our risk-adjusted antibiotic usage model to those obtained from the NHSN data and facility-level CMI. CMI is readily available from hospital-level data and has been shown to correlate with antibiotic use [32]. Hospital size, unit type, and proportion of patient-days spent in each unit type are all easily accessible within the AU module. Importantly, in our network of hospitals, the proportion of patient-days in miscellaneous unit types was highly variable. Consistent with prior reports [26], the proportion of patient-days in specific unit types was highly predictive of facility-level antibiotic prescribing rates in our model. Accounting for the proportion of patient-days in specific unit types will be critical for valid comparisons between facilities and developing benchmarking tools. In our model, accounting for the proportion of patient-days spent in specific care units and facility mean CMI significantly reduced the interhospital antibiotic use variability, highlighting the importance of including these predictors in benchmarking tools. However, substantial variation remained after adjusting for these variables. This variation suggests that there is an unmet need to improve the quality of antibiotic use through ASPs.

Facility-level antibiotic usage rates were not predictive of broad-spectrum antibiotic use, an association that has not been described previously. This highlights the need for usage metrics complementary to DOT/1000PD that incorporate antibiotic spectrum into antibiotic usage metrics. Metrics could strictly weight antibiotics based on spectrum covered [33] or weight antibiotics based on spectrum covered and ability to treated MDROs, as we did in our spectrum categorization. A weighted usage metric would account for both antibiotic deescalation and antibiotic discontinuation strategies.

Our study has inherent limitations. The NHSN AU data are limited to aggregated monthly data stratified by CDC defined floor type. The lack of patient-level clinical data has the potential to complicate benchmarking strategies. In addition, Utah may differ demographically from other states, and the characteristics of the 15 SCHs in a western state with low population density may differ from those of other small hospitals, limiting generalizability. Finally, the DOT/1000PD metric is being used as a surrogate measure of antibiotic prescribing quality; however, elevated DOT/1000PD rates have yet to be consistently tied to worse clinical outcomes or increased antibiotic resistance. Further studies will need to critically evaluate the clinical significance of the DOT/1000PD metric, its utility as a means to benchmark facilities, and its reflection on antibiotic prescribing quality and clinical outcomes.

The NHSN AU module is the first national database for antibiotic use data that has the potential to be used for antibiotic use benchmarking. Recently, the CDC has developed the Standardized Antimicrobial Administration Ratio (SAAR) to summarize AU data and allow interhospital comparison in the form of an observed-to-expected ratio while controlling for facility-level factors. The SAAR metric does not include CMI as a facility-level predictor and includes only adult and pediatric medical, surgical, and medical/surgical ICUs, which obviates the need to account for miscellaneous unit types. The SAAR metric will allow ASPs to quickly identify hospital units that are using more antibiotics than hospitals of similar composition. This would act as a signal for an AS team to more critically evaluate antibiotic usage in these units and assess prescribing appropriateness and the need for AS interventions.

SCHs in the United States will face significant challenges meeting the forthcoming national AS requirements [11–13] and improving antibiotic use due to limited access to infectious diseases physician and/or pharmacist leadership, limited information technology _support, and lack of AS guidance directed specifically to SCHs. Regardless of the challenges, given the high rates of antibiotic use and similar prescribing patterns compared with LCHs that we have described, implementing ASPs in SCHs is critical. Further research into the effectiveness of standard and novel (eg, telehealth AS and infectious diseases consultation) AS strategies in these unique settings is needed. The infectious diseases community must respond and assist SCHs in establishing effective ASPs to meet the needs of our patients and address antibiotic resistance.

Notes

Financial support. Pfizer Independent Grants for Learning & Change provided all project funding and The Joint Commission provided administrative oversight.

Potential conflicts of interest. E. S. has consulted for Durata Therapeutics and has received a grant from Allergan. A. L. H. has also received grants from the Agency for Healthcare Research and Quality and Merck. A. T. P. has consulted for Co-crystal, has received grants from BioFire, royalties from Antimicrobial Therapy, and payment for educational presentations from Medscape, and is on the board of Alios Pharmaceutical. All other authors report no potential conflicts. All authors have submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest. Conflicts that the editors consider relevant to the content of the manuscript have been disclosed.