Skip to main content
NCBI home page
eLife logoLink to eLife
. 2020 Feb 5;9:e52307. doi: 10.7554/eLife.52307

Potential impact of outpatient stewardship interventions on antibiotic exposures of common bacterial pathogens

Christine Tedijanto 1,✉, Yonatan H Grad 2,3, Marc Lipsitch 1,2
Editors: Neil M Ferguson4, Miles P Davenport5
PMCID: PMC7025820  PMID: 32022685

Abstract

The relationship between antibiotic stewardship and population levels of antibiotic resistance remains unclear. In order to better understand shifts in selective pressure due to stewardship, we use publicly available data to estimate the effect of changes in prescribing on exposures to frequently used antibiotics experienced by potentially pathogenic bacteria that are asymptomatically colonizing the microbiome. We quantify this impact under four hypothetical stewardship strategies. In one scenario, we estimate that elimination of all unnecessary outpatient antibiotic use could avert 6% to 48% (IQR: 17% to 31%) of exposures across pairwise combinations of sixteen common antibiotics and nine bacterial pathogens. All scenarios demonstrate that stewardship interventions, facilitated by changes in clinician behavior and improved diagnostics, have the opportunity to broadly reduce antibiotic exposures across a range of potential pathogens. Concurrent approaches, such as vaccines aiming to reduce infection incidence, are needed to further decrease exposures occurring in ‘necessary’ contexts.

Research organism: None

Introduction

Antibiotic consumption is a known driver of antibiotic resistance. In developed nations, over 80% of antibiotic consumption for human health occurs in the outpatient setting (European Centre for Disease Prevention and Control, 2018; Public Health Agency of Canada, 2018; Public Health England, 2017; Swedres-Svarm, 2017), and US-based studies conducted across different subpopulations have estimated that 23–40% of outpatient prescriptions may be inappropriate (Chua et al., 2019; Fleming-Dutra et al., 2016; Olesen et al., 2018b). Inappropriate antibiotic use leads to increased risk of adverse events (Linder, 2008; Shehab et al., 2008), disruption of colonization resistance and other benefits of the microbial flora (Buffie and Pamer, 2013; Khosravi and Mazmanian, 2013), and bystander selection for antibiotic resistance, with little to no health gains for the patient. Recent work on bystander selection estimates that, for 8 of 9 potential pathogens of interest, over 80% of their exposures to commonly used antibiotic classes in the outpatient setting occur when the organisms are asymptomatically colonizing the microbiome, not causing disease (Tedijanto et al., 2018). A corollary of extensive bystander selection is that reductions in use will prevent antibiotic exposures for species throughout the microbiome.

We use publicly available data to quantify potentially avertable exposures of bacterial species to commonly used antibiotics under hypothetical changes in prescribing practice. The set of scenarios included here are intended as thought experiments to explore the upper bounds of avertable antibiotic exposures. Reductions in antibiotic consumption have historically had variable impacts on resistance levels, likely dependent on setting, baseline consumption and resistance patterns, and fitness costs (Andersson and Hughes, 2010; Lipsitch, 2001). In addition, treatment strategies are primarily guided by randomized controlled trials assessing immediate clinical outcomes, with less careful consideration given to the microbiome-wide effect of such decisions. We apply analytical methods to parse out species-level effects of changes in prescribing practice in order to better understand the potential impact of antibiotic stewardship.

Materials and methods

Scenarios of interest

This analysis includes sixteen antibiotics that are frequently prescribed in the outpatient setting and nine potentially pathogenic bacterial species that are commonly carried in the normal human microbiome. For each antibiotic-species pair, we estimate the proportion of antibiotic exposures experienced by that species that could be averted under four hypothetical scenarios. The scenarios range from broad elimination of unnecessary prescribing to focused modifications of antibiotic use for specific indications as follows:

  1. Eliminate unnecessary antibiotic use across all outpatient conditions.

  2. Eliminate all antibiotic use for outpatient respiratory conditions for which antibiotics are not indicated.

  3. Eliminate all antibiotic use for acute sinusitis.

  4. Prescribe nitrofurantoin for all cases of cystitis in women.

We use the results reported by Fleming-Dutra and colleagues as estimates of unnecessary antibiotic use in ambulatory care settings (Fleming-Dutra et al., 2016). In this study, the authors convened a group of experts to estimate the proportion of appropriate antibiotic use by condition and age group based on clinical guidelines. When guidelines could not be used for this task, the rate of appropriate antibiotic prescribing was estimated by benchmarking against the lowest-prescribing US region. Based on the same paper, we consider viral upper respiratory tract infection, influenza, non-suppurative otitis media, viral pneumonia, bronchitis, and allergy and asthma to be outpatient respiratory conditions for which antibiotics are not indicated.

Antibiotic treatment of acute bacterial sinusitis is currently guideline-recommended (Rosenfeld et al., 2015). However, bacteria are an infrequent cause of acute sinusitis, and due to the self-limiting nature of the syndrome, evidence to support antibiotic treatment is weak (Burgstaller et al., 2016; Fokkens et al., 2007). The cause of sinusitis (whether bacterial, viral, or noninfectious) can be very difficult to distinguish in practice, and as a result of this challenge and others, antibiotics continue to be prescribed at over 80% of US outpatient visits with a primary diagnosis of acute sinusitis (Smith et al., 2013). In contrast, antibiotics are always recommended for urinary tract infections (UTIs). Despite being recommended as a second-line therapy for cystitis due to concerns about resistance, fluoroquinolones are the most common treatment, prescribed at 40% of outpatient visits for uncomplicated UTI (Kabbani et al., 2018). We explore the hypothetical scenario of treating all cases of cystitis in women with nitrofurantoin, a recommended first-line therapy with good potency against common uropathogens, low levels of resistance, and decreased risk of collateral damage to the intestinal microbiome due to its propensity to concentrate in the bladder (Gupta et al., 2011; Gupta et al., 1999; Stewardson et al., 2015). Nitrofurantoin is considered clinically appropriate unless the patient has chronic kidney disease, is showing signs of early pyelonephritis, or has had a prior urinary isolate resistant to nitrofurantoin (Hooton and Gupta, 2019). In our analysis, we exclude patients with a simultaneous diagnosis of pyelonephritis. We assume that other contraindications are fairly rare in the general population.

Data sources and methodology

Extending our recent work on bystander selection (Tedijanto et al., 2018), we used the 2015 National Ambulatory Medical Care Survey and National Hospital Ambulatory Medical Care Survey (NAMCS/NHAMCS) (National Center for Health Statistics, 2015), the Human Microbiome Project (HMP) (Huttenhower et al., 2012; Human Microbiome Project Consortium, 2012), and assorted carriage and etiology studies (Bäckhed et al., 2015; Bluestone et al., 1992; Bogaert et al., 2011; Brook et al., 1994; Celin et al., 1991; Chira and Miller, 2010; Edlin et al., 2013; Ginsburg et al., 1985; Gunnarsson et al., 1997; Gupta et al., 1999; Gwaltney et al., 1992; Hammitt et al., 2006; Lif Holgerson et al., 2015; Huang et al., 2009; Jain et al., 2015; Mainous et al., 2006; Pettigrew et al., 2012; Regev-Yochay et al., 2004; Verhaegh et al., 2010; Wubbel et al., 1999; Yassour et al., 2016) (details in Figure 1—source data 1) to estimate national outpatient antibiotic exposures by drug, species, and condition. NAMCS/NHAMCS are annual cross-sectional surveys designed to sample outpatient visits in the United States, and up to five diagnoses and thirty medications may be associated with each visit. We used methodology developed by Fleming-Dutra and colleagues (Fleming-Dutra et al., 2016), and applied in other studies (Olesen et al., 2018a), to group diagnosis codes into conditions and link antibiotic prescriptions with the most likely indication. For this analysis, visits with a diagnosis of acute cystitis (ICD-9-CM: 595.0), unspecified cystitis (595.9) or unspecified UTI (599.0), without a concurrent diagnosis of pyelonephritis (590.1, 590.8), renal abscess (590.2), or kidney infection (590.9), were considered to be associated with cystitis. In this analysis, we maintain the assumption that one antibiotic prescription is equivalent to one exposure; antibiotic exposures experienced by a given species and associated with a given condition are roughly estimated as the product of antibiotic prescriptions for that condition and species carriage prevalence, which is dependent on disease etiology (target exposures) and asymptomatic carriage prevalence (bystander exposures). For diagnoses where etiology was not readily available, we assumed that none of our species of interest were causative agents. These assumptions have been enumerated in detail in prior work (Tedijanto et al., 2018). We applied proportions of unnecessary antibiotic prescribing by condition and age group estimated by Fleming-Dutra and colleagues based on expert opinion, clinical guidelines, and regional variability in use (Fleming-Dutra et al., 2016). For relevant scenarios (1 and 2), we applied the proportions of unnecessary use evenly across all antibiotic prescriptions. Antibiotics and antibiotic classes are identified by the Lexicon Plus, 2008 classification scheme (https://www.cerner.com/solutions/drug-database).

The proportion of avertable antibiotic exposures for each species is defined by Equation 1. The equation adopts previously described notation (Tedijanto et al., 2018) with modifications. A listing of all variables and descriptions can be found in Table 1. Let a represent antibiotic, s represent species, i represent ICD-9-CM diagnosis code, and g represent age group. Throughout the analysis we have weighted outpatient visits to be nationally representative using the sampling and nonresponse weights provided in NAMCS/NHAMCS. Let Xas be the number of avertable exposures by antibiotic and species, Tas be the total number of exposures by antibiotic and species, daig be the number of prescriptions of antibiotic a associated with diagnosis code i in age group g, psig be the carriage prevalence of species s among those with diagnosis code i in age group g, and qaig be the proportion of avertable exposures by diagnosis code and age group in the given scenario. For example, in the scenario assessing elimination of all unnecessary antibiotic use, qaig is the proportion of avertable antibiotic use by diagnosis and age group (Fleming-Dutra et al., 2016). Alternatively, in the scenario assessing elimination of non-nitrofurantoin treatment for cystitis, qaig is 1 when a is not nitrofurantoin, i is a diagnosis code associated with cystitis, and the patient is female, and 0 elsewhere. Carriage prevalences (psig) are assumed to be constant within three age groups (under 1 year, 1–5 years, over 5 years old) (Tedijanto et al., 2018), while proportions of avertable antibiotic use (qaig) were reported for three different age groups (0–19 years, 20–64 years, 65 years old and over) (Fleming-Dutra et al., 2016). G is the smallest set of age groups that accounts for this granularity (under 1 year, 1–5 years, 6–19 years, 20–64 years, 65 years old and over). For antibiotic prescriptions that occurred at visits without any ICD-9-CM diagnosis codes (i = 0), we applied the carriage prevalence among healthy individuals.

Table 1. Notation, descriptions, and sources for variables in Equation 1.

Variable Definition Source
daig Number of prescriptions (using nationally representative weights) of antibiotic a associated with ICD9-CM diagnosis code i in age group g NAMCS/NHAMCS 2015
esig Proportion of cases of condition defined by ICD9-CM diagnosis code i in age group g caused by species s Published etiology studies (see Figure 1—source data 1)
ps0g Asymptomatic carriage prevalence of species s in age group g Human Microbiome Project (HMP) and published carriage studies
psig Carriage prevalence of species s among individuals diagnosed with ICD9-CM code i in age group g esig+(1−esig)ps0g
qaig Proportion of prescriptions of antibiotic a associated with ICD9-CM diagnosis code i in age group g that are avertable under the given scenario Based on article by Fleming-Dutra et al. (Fleming-Dutra et al., 2016) with adjustments as described in Figure 1—source data 2
Xas Number of exposures of antibiotic a experienced by species s that are avertable under the given scenario ∑g=1G∑i=0Idaig×psig×qaig
Tas Total number of exposures of antibiotic a experienced by species s ∑g=1G∑i=0Idaig×psig
Open in a new tab

Equation 1. Proportion of avertable exposures by species and antibiotic.

XasTas=∑g=1G∑i=0Idaig×psig×qaig∑g=1G∑i=0Idaig×psig

In scenarios where unnecessary use is eliminated (1 and 2), we assume that only bystander exposures are affected. This presumes that perfect discrimination between bacterial and non-bacterial etiologies is possible. This results in a slight modification to the numerator of Equation 1 -- psig is changed to ps0g as all eliminated exposures would have occurred during asymptomatic carriage. For cases where esig is 1, we assume zero avertable exposures. In addition, we make slight modifications to qaig when we estimate the proportion of bacterial cases to be larger than the proportion of necessary prescriptions (Figure 1—source data 2; Bluestone et al., 1992; Brook et al., 1994; Celin et al., 1991; Brook, 2016). As a sensitivity analysis, we include the proportion of avertable exposures for each antibiotic-species pair under Scenario 1 if antibiotic use was eliminated equally across both target and bystander exposures (Figure 1—figure supplement 1). All analysis was conducted in R version 3.6.1.

Results

Results for all four scenarios are depicted in Figure 1

Figure 1. Heatmaps showing the estimated percentage of species exposures to each antibiotic or antibiotic class that could be averted by scenario.

Scenarios are defined as elimination of (A) unnecessary antibiotic prescriptions across all outpatient conditions, (B) all antibiotic use for outpatient respiratory conditions for which antibiotics are not indicated, (C) all antibiotic use for acute sinusitis, and (D) non-nitrofurantoin treatment of cystitis in women. Drug class results include prescriptions of all antibiotics in that class, as identified by the Lexicon Plus classification system. Sensitivity and other additional analyses are shown in Figure 1—figure supplements 1–3. Abbreviations: Antibiotics (y-axis): AMX-CLAV = amoxicillin-clavulanate, MACR/LINC = macrolides/ lincosamides, TMP-SMX = sulfamethoxazole-trimethoprim; Organisms (x-axis): EC = E. coli, HI = H. influenzae, KP = K. pneumoniae, MC = M. catarrhalis, PA = P. aeruginosa, SA = S. aureus, SAg = S. agalactiae, SP = S. pneumoniae, SPy = S. pyogenes; PY = person years.

Figure 1—source data 1. Summary of etiology studies.
Figure 1—source data 2. Modifications to proportions of unnecessary antibiotic prescriptions based on published etiology studies and antibiotic use in NAMCS/NHAMCS.
Modifications were made if the reported proportion of bacterial cases for a given condition exceeded the estimated proportion of appropriate prescriptions reported in Fleming-Dutra et al. (2016).
Figure 1—source data 3. Estimated antibiotic class exposures by exposed and causative species.

Figure 1.

Figure 1—figure supplement 1. Sensitivity analysis of proportions of avertable exposures across all outpatient conditions when proportion of unnecessary use is applied equally across target and bystander exposures (Equation 1 without modification).

Figure 1—figure supplement 1.

Figure 1—figure supplement 2. Sensitivity analyses for Scenario 3 (elimination of all antibiotic use for acute sinusitis).

Figure 1—figure supplement 2.

(A) Unnecessary prescribing for acute sinusitis estimated from Fleming-Dutra et al. Based on Fleming-Dutra et al. (2016), we estimate that the proportion of unnecessary antibiotic prescriptions for acute sinusitis is 18%, 100%, and 34% for individuals who are 0-19, 20-64, and over 65 years old, respectively (see Figure 1—source data 2). (B) Estimates of bacterial etiology used to approximate necessary antibiotic use for acute sinusitis. The following upper bound estimates for the proportion of acute sinusitis cases with bacterial etiology were used: 2% for adults and 13% for children (applied to 0-19 year old age group) (Sande and Gwaltney, 2004; Snow et al., 2001; Wald et al., 1991). (C) Estimates of bacterial etiology among those presenting to primary care clinics used to approximate necessary antibiotic use. Patients visiting the clinic are likely to have more persistent, severe, or worsening symptoms, which may be indications of bacterial etiology. Studies have estimated that approximately 38% of adults and 17% of children (upper bounds) visiting primary care for symptoms of acute upper respiratory tract infection have bacterial rhinosinusitis (Aitken and Taylor, 1998; Benninger et al., 2000; Williams et al., 1992).
Figure 1—figure supplement 3. Avertable exposures under Scenario 1 (elimination of all unnecessary antibiotic use across outpatient conditions) using 2010-2011 NAMCS/NHAMCS data with (left) and without (right) NHAMCS outpatient department data.

Figure 1—figure supplement 3.

Open in a new tab

Scenario 1

We estimate that elimination of unnecessary antibiotic prescriptions across all outpatient conditions would prevent 6% to 48% (IQR: 17% to 31%) of antibiotic-species exposures (Figure 1A). The smallest reduction is associated with S. pyogenes exposures to cefdinir and the largest with H. influenzae, E. coli, and P. aeruginosa exposures to azithromycin. If all unnecessary antibiotic use could be prevented, over 30% of exposures to amoxicillin-clavulanate, penicillin, azithromycin, clarithromycin, levofloxacin, and doxycycline across most potential pathogens of interest could be averted. Of particular interest, elimination of all unnecessary prescribing in the outpatient setting could reduce exposures of S. pneumoniae to penicillins and macrolides by 27% and 37%, respectively, and of S. aureus to penicillins and quinolones by 27% and 21%. For E. coli and K. pneumoniae, approximately one-quarter of exposures to cephalosporins and one-fifth of exposures to quinolones could be averted.

Scenario 2

Scenario 2, elimination of all antibiotic use for outpatient respiratory conditions where antibiotics are not indicated, is a subset of Scenario 1 and accounts for a substantial portion of avertable exposures included in the first scenario. As in Scenario 1, the results are primarily driven by drug and depend on the amount of use of that drug for the affected conditions (in this case, conditions for which antibiotics are not indicated). Pathogen characteristics affecting the proportion of avertable exposures can be better understood by looking across all species for a single drug. For example, we focus on the row for azithromycin. Based on NAMCS/NHAMCS 2015, 32% of all azithromycin use is associated with the affected conditions. Since these conditions are never caused by our organisms of interest, the maximum proportion of avertable exposures for any species is 32%. However, the proportion of avertable exposures is modulated by the number of azithromycin exposures that occur for treatment of other conditions – for example, a lower proportion of avertable exposures for S. pyogenes, M. catarrhalis, and S. agalactiae indicates that they have relatively frequent exposure to azithromycin during treatment of other conditions that are unaffected in this scenario, both as a bystander and causative pathogen.

Scenario 3

In Scenario 3, we explore elimination of all antibiotic use for acute sinusitis. Overall, acute sinusitis accounts for just 3% of prescriptions of our included antibiotic classes, resulting in the low proportions of avertable exposures across most bacteria-antibiotic combinations. However, proportions of avertable exposures are relatively high for amoxicillin-clavulanate and clarithromycin, antibiotics for which a large proportion of their use is due to acute sinusitis (15% and 9%, respectively). In this scenario, we make the assumption that all use for acute sinusitis is unnecessary. In reality, a small number of cases are truly bacterial and antibiotics are indicated for patients with persistent, severe or worsening symptoms (Rosenfeld et al., 2015). We explore varying levels of unnecessary use based on etiological data in Figure 1—figure supplement 2 (Aitken and Taylor, 1998; Benninger et al., 2000; Fleming-Dutra et al., 2016; Sande and Gwaltney, 2004; Snow et al., 2001; Wald et al., 1991; Williams et al., 1992).

Scenario 4

In this scenario, we consider a program to prescribe only nitrofurantoin for all cases of acute cystitis in women, replacing all other antibiotics that are currently in use. As a result of this change, we estimate that exposures of E. coli and K. pneumoniae to cephalosporins could be reduced by 16% and 17%, respectively, and exposures to quinolones could decrease by 26% and 28%. Uniquely in this scenario among those we considered, several of our species of interest cause a large proportion of UTI cases (E. coli, K. pneumoniae, and P. aeruginosa). To explore the effects of being a causative pathogen on the proportion of avertable exposures, we focus on the results for ciprofloxacin. Overall, 27% of ciprofloxacin use is associated with acute cystitis in women. Organisms that are not causative pathogens of cystitis have proportions of avertable exposures close to or less than 27%, but a substantially higher proportion of exposures could be averted among causative pathogens - 33%, 36%, and 44% of ciprofloxacin exposures to E. coli, K. pneumoniae, and P. aeruginosa, respectively. Although E. coli is the most common cause of UTIs, its estimated proportion of avertable exposures in this scenario is lower than that of P. aeruginosa, likely due to the larger number of antibiotic exposures incurred by E. coli during asymptomatic carriage.

This scenario also allows us to observe the impact of differences in disease incidence, antibiotic use patterns, and carriage prevalence across age groups on the proportion of avertable exposures. For example, M. catarrhalis is asymptomatically carried in children much more frequently than in adults. Since cystitis occurs primarily in adults, the proportion of avertable exposures is often lower for M. catarrhalis compared to other species due to bystander exposures in children that are largely unaffected by modified prescribing practices for cystitis. This age group effect can be substantial for antibiotics that are commonly given to children, especially when the antibiotic is also used to treat a condition caused by M. catarrhalis (e.g. cefdinir and otitis media), but is less influential for antibiotics that are infrequently prescribed to children, such as ciprofloxacin and levofloxacin.

Overall, the proportion of avertable exposures is driven by pathogen, syndrome, and prescribing characteristics. Because our organisms of interest experience the vast majority of their antibiotic exposures as bystanders, the metric is largely dependent on the proportion of antibiotic use that is associated with the conditions affected in the given scenario. In a scenario where all antibiotic use for affected conditions is eliminated, this proportion would be equal to the proportion of avertable exposures for organisms exposed to antibiotics only during carriage (never pathogens). Being a causative pathogen for any of the affected conditions increases the proportion of avertable exposures. In contrast, being a causative pathogen for unaffected conditions increases unaffected antibiotic exposures and thus decreases the proportion of avertable exposures. For pathogens responsible for multiple syndromes, both factors may be at play. Finally, asymptomatic carriage prevalence is positively associated with both affected and unaffected antibiotic exposures, and, as a result, its overall effect on the proportion of avertable exposures is dependent on the antibiotic-species combination of interest.

Discussion

We quantify the species-level impact of changes in antibiotic consumption as the proportion of antibiotic exposures experienced by common bacterial pathogens that could be averted under four hypothetical scenarios. In the scenario where unnecessary antibiotic use for all outpatient conditions is eliminated, we find that up to 48% of exposures (of H. influenzae, E. coli, and P. aeruginosa to azithromycin) could be avoided. In addition, impact of the intervention across antibiotics and species is widespread, with half of antibiotic-species pairs expected to experience a reduction in exposures of over one-fifth (22%). For conditions which require antibiotic treatment, such as UTIs, switching to antibiotics with decreased collateral damage to the microbiome, such as nitrofurantoin, may be an effective strategy for reducing antibiotic exposures across species.

In three out of four situations we assess, antibiotics are considered entirely unnecessary for some or all cases. In the future, this method may also be extended to measure net changes in exposures resulting from more nuanced scenarios where one antibiotic is substituted for another. We did not assess such changes for nitrofurantoin, as its effects on the microbiome outside of the bladder are thought to be minimal (Stewardson et al., 2015). The reductions in use presented here may be practically infeasible due to challenges including similar clinical presentation of viral and bacterial infections, laboratory processing times that prevent identification of causative pathogens during outpatient visits, individual considerations such as allergies (Sakoulas et al., 2019) or heightened risk of adverse events, misdiagnosis (Filice et al., 2015; Tomas et al., 2015), and patient-driven demand (Vanden Eng et al., 2003). Even if elimination of unnecessary use were fully realized, our results imply that the majority of species’ antibiotic exposures occur in the context of ‘necessary’ antibiotic use. These findings underscore the importance of considering bystander effects and the need for a multi-pronged approach to programs aimed at controlling antibiotic resistance. Rapid diagnostics, diagnostics that accurately discriminate between bacterial and non-bacterial causes, patient education, and improved decision-making tools and other interventions to motivate changes in clinician behavior can enhance responsible antibiotic consumption and reduce unnecessary antibiotic use; these should be implemented simultaneously with prevention measures such as infection control, access to clean water and sanitation, safe sex interventions, and vaccination, which aim to reduce infection incidence and thus any antibiotic use.

At the time of this analysis, NHAMCS data from hospital outpatient departments was not available for 2015. In the 2010–2011 NAMCS/NHAMCS data, visits to hospital outpatient departments accounted for 36% of included sampled visits, but just 9% of all antibiotic mentions. Since the proportions of avertable exposures including and excluding these visits for 2010–2011 were similar for Scenario 1 (Figure 1—figure supplement 3), we assumed that the existing 2015 data without hospital outpatient department visits was representative of the NAMCS/NHAMCS population for the purposes of our analysis. Certain outpatient encounters are also outside the scope of NAMCS/NHAMCS, including federal facility visits, telephone contacts, house calls, long-term care stays, urgent care, retail clinics, and hospital discharge prescriptions.

Additional limitations of the included datasets and method for estimating antibiotic exposures have been previously enumerated (Tedijanto et al., 2018). Notably, the NAMCS/NHAMCS does not provide information to link medications with diagnoses, so we adopted a published tiered system to pair antibiotic use at each visit with the most likely indication (Fleming-Dutra et al., 2016). For visits with multiple diagnoses, all antibiotic use was attributed to the single most-likely indication. This method tends to overestimate antibiotic use for conditions for which antibiotics are almost always indicated, which may lead to clinically unusual diagnosis-treatment linkages. In the same way, antibiotic prescriptions are underestimated for diagnoses for which antibiotic use is not indicated, potentially leading to downward bias in avertable exposures for these conditions. Additionally, NAMCS/NHAMCS antibiotic use data does not include prescription details such as duration or dose. Future studies with more granular data may extend the methods presented here to account for such factors.

At the time of this analysis, Fleming-Dutra et al. remained the most up-to-date source of unnecessary antibiotic prescribing by condition. Recent work with slightly different methods found a lower proportion of inappropriate antibiotic prescriptions overall (23.2% compared to 30%) but did not report their estimates by condition (Chua et al., 2019). Although comparable estimates of unnecessary prescribing over time are unavailable, multiple studies have shown that declines in antibiotic use from approximately 2011 to 2016 in the United States have been primarily due to pediatric prescribing, implying that, at least for adults, levels of unnecessary prescribing likely remained similar (Durkin et al., 2018; King et al., 2020; Klevens et al., 2019; Olesen et al., 2018b). We also assume that the proportion of unnecessary use is constant across all antibiotic use for the same condition as it is difficult to identify specific antibiotic prescriptions that were unwarranted without detailed chart review. A study among outpatients in the Veterans Affairs medical system found that among prescriptions for community-acquired pneumonia, sinusitis, and acute exacerbations of chronic bronchitis, the highest proportion of macrolide use was inappropriate (27%), followed by penicillins (22%) and quinolones (12%) (Tobia et al., 2008). Similar studies are needed to understand which antibiotics are frequently used inappropriately for other indications and settings.

Finally, it is important to note that the reduction in antibiotic exposures estimated here does not translate to the same reduction in the prevalence of resistance or in the morbidity and mortality attributable to resistance. Although population-level antibiotic consumption has been positively correlated with levels of antibiotic resistance, the impact of changing consumption is not well-understood and is likely to vary widely by antibiotic-species combination (or even antibiotic-strain combination) based on resistance mechanisms, fitness costs, and co-selection, among other factors (Andersson and Hughes, 2010; Pouwels et al., 2017). Antibiotic consumption is also highly heterogeneous and the impact of stewardship on resistance may depend on the affected patient populations (Olesen et al., 2018a). Additionally, in this analysis, we give each exposure equal weight. However, the selective pressure imposed by a single exposure depends on a number of variables, including pharmacokinetics, pharmacodynamics, distribution of bacteria across body sites, and bacterial population size. For example, we might expect that the probability of resistance scales with population size, and thus that an exposure received by an individual with higher bacterial load will have a larger impact on resistance. Further research, integrating knowledge from clinical, ecological, and evolutionary spheres, is needed to elucidate the relationship between antibiotic use and selective pressures and ultimately between use and resistance at the population level (MacLean and San Millan, 2019).

Reductions in antibiotic consumption are necessary to preserve the potency of these drugs. Quantifying changes in species-level exposures due to stewardship programs is one more step towards understanding how changes in antibiotic use correspond to antibiotic resistance. The methods presented here may be easily extended to incorporate other data sources, such as claims, or to assess more specific stewardship programs. We show that while improved prescribing practices have the potential to prevent antibiotic exposures experienced by bacterial species throughout the microbiome, complementary efforts to facilitate appropriate antibiotic consumption and decrease overall infection incidence are required to substantially avert exposures.

Acknowledgements

We thank Dr. Lauri Hicks for her helpful comments on this manuscript.

Funding Statement

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Contributor Information

Christine Tedijanto, Email: ctedijanto@g.harvard.edu.

Neil M Ferguson, Imperial College London, United Kingdom.

Miles P Davenport, University of New South Wales, Australia.

Funding Information

This paper was supported by the following grants:

  • National Institute of General Medical Sciences U54GM088558 to Marc Lipsitch.

  • National Institute of Allergy and Infectious Diseases R01AI132606 to Yonatan H Grad.

  • Centers for Disease Control and Prevention CK000538-01 to Marc Lipsitch.

  • Doris Duke Charitable Foundation to Yonatan H Grad.

  • National Institute of Allergy and Infectious Diseases T32AI007535 to Christine Tedijanto.

Additional information

Competing interests

Reviewing editor, eLife.

No competing interests declared.

Has received consulting income from Merck and GlaxoSmithKline.

Author contributions

Conceptualization, Formal analysis, Visualization, Methodology.

Supervision, Funding acquisition, Methodology.

Conceptualization, Supervision, Funding acquisition, Methodology.

Additional files

Transparent reporting form

Data availability

Data from the 2015 National Ambulatory Medical Care Survey (NAMCS) and National Hospital Ambulatory Medical Care Survey (NHAMCS) are publicly available from the National Center for Health Statistics. This study also uses data from published literature, including the Human Microbiome Project and other studies summarized in Figure 1—source data 1.

The following previously published datasets were used:

National Center for Health Statistics. 2015. National Hospital Ambulatory Medical Care Survey. National Center for Heath Statistics. ED2015-stata

National Center for Health Statistics. 2015. National Ambulatory Medical Care Survey. National Center for Heath Statistics. namcs2015-stata

References

  1. Aitken M, Taylor JA. Prevalence of clinical sinusitis in young children followed up by primary care pediatricians. Archives of Pediatrics & Adolescent Medicine. 1998;152:244–248. doi: 10.1001/archpedi.152.3.244. [DOI] [PubMed] [Google Scholar]
  2. Andersson DI, Hughes D. Antibiotic resistance and its cost: is it possible to reverse resistance? Nature Reviews Microbiology. 2010;8:260–271. doi: 10.1038/nrmicro2319. [DOI] [PubMed] [Google Scholar]
  3. Bäckhed F, Roswall J, Peng Y, Feng Q, Jia H, Kovatcheva-Datchary P, Li Y, Xia Y, Xie H, Zhong H, Khan MT, Zhang J, Li J, Xiao L, Al-Aama J, Zhang D, Lee YS, Kotowska D, Colding C, Tremaroli V, Yin Y, Bergman S, Xu X, Madsen L, Kristiansen K, Dahlgren J, Wang J, Jun W. Dynamics and stabilization of the human gut microbiome during the first year of life. Cell Host & Microbe. 2015;17:690–703. doi: 10.1016/j.chom.2015.04.004. [DOI] [PubMed] [Google Scholar]
  4. Benninger MS, Holzer SES, Lau J. Diagnosis and treatment of uncomplicated acute bacterial rhinosinusitis: summary of the agency for health care policy and research evidence-based report. Otolaryngology–Head and Neck Surgery. 2000;122:1–7. doi: 10.1016/S0194-5998(00)70135-5. [DOI] [PubMed] [Google Scholar]
  5. Bluestone CD, Stephenson JS, Martin LM. Ten-year review of otitis media pathogens. The Pediatric Infectious Disease Journal. 1992;11:7–11. doi: 10.1097/00006454-199208001-00002. [DOI] [PubMed] [Google Scholar]
  6. Bogaert D, Keijser B, Huse S, Rossen J, Veenhoven R, van Gils E, Bruin J, Montijn R, Bonten M, Sanders E. Variability and diversity of nasopharyngeal Microbiota in children: a metagenomic analysis. PLOS ONE. 2011;6:e17035. doi: 10.1371/journal.pone.0017035. [DOI] [PMC free article] [PubMed] [Google Scholar]
  7. Brook I, Thompson DH, Frazier EH. Microbiology and management of chronic maxillary sinusitis. Archives of Otolaryngology - Head and Neck Surgery. 1994;120:1317–1320. doi: 10.1001/archotol.1994.01880360015003. [DOI] [PubMed] [Google Scholar]
  8. Brook I. Microbiology of chronic rhinosinusitis. European Journal of Clinical Microbiology & Infectious Diseases. 2016;35:1059–1068. doi: 10.1007/s10096-016-2640-x. [DOI] [PubMed] [Google Scholar]
  9. Buffie CG, Pamer EG. Microbiota-mediated colonization resistance against intestinal pathogens. Nature Reviews Immunology. 2013;13:790–801. doi: 10.1038/nri3535. [DOI] [PMC free article] [PubMed] [Google Scholar]
  10. Burgstaller JM, Steurer J, Holzmann D, Geiges G, Soyka MB. Antibiotic efficacy in patients with a moderate probability of acute rhinosinusitis: a systematic review. European Archives of Oto-Rhino-Laryngology. 2016;273:1067–1077. doi: 10.1007/s00405-015-3506-z. [DOI] [PubMed] [Google Scholar]
  11. Celin SE, Bluestone CD, Stephenson J, Yilmaz HM, Collins JJ. Bacteriology of acute otitis media in adults. JAMA: The Journal of the American Medical Association. 1991;266:2249–2252. doi: 10.1001/jama.1991.03470160081036. [DOI] [PubMed] [Google Scholar]
  12. Chira S, Miller LG. Staphylococcus aureus is the most common identified cause of cellulitis: a systematic review. Epidemiology and Infection. 2010;138:313–317. doi: 10.1017/S0950268809990483. [DOI] [PubMed] [Google Scholar]
  13. Chua KP, Fischer MA, Linder JA. Appropriateness of outpatient antibiotic prescribing among privately insured US patients: icd-10-cm based cross sectional study. BMJ. 2019;364:k5092. doi: 10.1136/bmj.k5092. [DOI] [PMC free article] [PubMed] [Google Scholar]
  14. Durkin MJ, Jafarzadeh SR, Hsueh K, Sallah YH, Munshi KD, Henderson RR, Fraser VJ. Outpatient antibiotic prescription trends in the united states: a national cohort study. Infection Control & Hospital Epidemiology. 2018;39:584–589. doi: 10.1017/ice.2018.26. [DOI] [PMC free article] [PubMed] [Google Scholar]
  15. Edlin RS, Shapiro DJ, Hersh AL, Copp HL. Antibiotic resistance patterns of outpatient pediatric urinary tract infections. Journal of Urology. 2013;190:222–227. doi: 10.1016/j.juro.2013.01.069. [DOI] [PMC free article] [PubMed] [Google Scholar]
  16. European Centre for Disease Prevention and Control . Antimicrobial Consumption: Annual Epidemiological Report for 2017. Stockholm: Annual Epidemiological Report on Communicable Diseases in Europe; 2018. [Google Scholar]
  17. Filice GA, Drekonja DM, Thurn JR, Hamann GM, Masoud BT, Johnson JR. Diagnostic errors that lead to inappropriate antimicrobial use. Infection Control & Hospital Epidemiology. 2015;36:949–956. doi: 10.1017/ice.2015.113. [DOI] [PubMed] [Google Scholar]
  18. Fleming-Dutra KE, Hersh AL, Shapiro DJ, Bartoces M, Enns EA, File TM, Finkelstein JA, Gerber JS, Hyun DY, Linder JA, Lynfield R, Margolis DJ, May LS, Merenstein D, Metlay JP, Newland JG, Piccirillo JF, Roberts RM, Sanchez GV, Suda KJ, Thomas A, Woo TM, Zetts RM, Hicks LA. Prevalence of inappropriate antibiotic prescriptions among US ambulatory care visits, 2010-2011. Jama. 2016;315:1864–1873. doi: 10.1001/jama.2016.4151. [DOI] [PubMed] [Google Scholar]
  19. Fokkens W, Lund V, Mullol J, European Position Paper on Rhinosinusitis and Nasal Polyps Group EP3OS 2007: european position paper on rhinosinusitis and nasal polyps 2007. A summary for otorhinolaryngologists. Rhinology. 2007;45:97–101. [PubMed] [Google Scholar]
  20. Ginsburg CM, McCracken GH, Crow SD, Dildy BR, Morchower G, Steinberg JB, Lancaster K. Seroepidemiology of the group-A streptococcal carriage state in a private pediatric practice. American Journal of Diseases of Children. 1985;139:614–617. doi: 10.1001/archpedi.1985.02140080084039. [DOI] [PubMed] [Google Scholar]
  21. Gunnarsson RK, Holm SE, Söderström M. The prevalence of beta-haemolytic streptococci in throat specimens from healthy children and adults implications for the clinical value of throat cultures. Scandinavian Journal of Primary Health Care. 1997;15:149–155. doi: 10.3109/02813439709018506. [DOI] [PubMed] [Google Scholar]
  22. Gupta K, Scholes D, Stamm WE. Increasing prevalence of antimicrobial resistance among uropathogens causing acute uncomplicated cystitis in women. Jama. 1999;281:736–738. doi: 10.1001/jama.281.8.736. [DOI] [PubMed] [Google Scholar]
  23. Gupta K, Hooton TM, Naber KG, Wullt B, Colgan R, Miller LG, Moran GJ, Nicolle LE, Raz R, Schaeffer AJ, Soper DE, Infectious Diseases Society of America, European Society for Microbiology and Infectious Diseases International clinical practice guidelines for the treatment of acute uncomplicated cystitis and pyelonephritis in women: a 2010 update by the infectious diseases society of America and the european society for microbiology and infectious diseases. Clinical Infectious Diseases. 2011;52:e103–e120. doi: 10.1093/cid/ciq257. [DOI] [PubMed] [Google Scholar]
  24. Gwaltney JM, Scheld WM, Sande MA, Sydnor A. The microbial etiology and antimicrobial therapy of adults with acute community-acquired sinusitis: a fifteen-year experience at the university of Virginia and review of other selected studies. Journal of Allergy and Clinical Immunology. 1992;90:457–462. doi: 10.1016/0091-6749(92)90169-3. [DOI] [PubMed] [Google Scholar]
  25. Hammitt LL, Bruden DL, Butler JC, Baggett HC, Hurlburt DA, Reasonover A, Hennessy TW. Indirect effect of conjugate vaccine on adult carriage of Streptococcus pneumoniae: an explanation of trends in invasive pneumococcal disease. The Journal of Infectious Diseases. 2006;193:1487–1494. doi: 10.1086/503805. [DOI] [PubMed] [Google Scholar]
  26. Hooton TM, Gupta K. Acute simple cystitis in women. 2019 https://www.uptodate.com/contents/acute-simple-cystitis-in-women
  27. Huang SS, Hinrichsen VL, Stevenson AE, Rifas-Shiman SL, Kleinman K, Pelton SI, Lipsitch M, Hanage WP, Lee GM, Finkelstein JA. Continued impact of pneumococcal conjugate vaccine on carriage in young children. Pediatrics. 2009;124:e1–e11. doi: 10.1542/peds.2008-3099. [DOI] [PMC free article] [PubMed] [Google Scholar]
  28. Human Microbiome Project Consortium A framework for human microbiome research. Nature. 2012;486:215–221. doi: 10.1038/nature11209. [DOI] [PMC free article] [PubMed] [Google Scholar]
  29. Huttenhower C, Gevers D, Knight R, Abubucker S, Badger JH, Chinwalla AT, White O, Human Microbiome Project Consortium Structure, function and diversity of the healthy human microbiome. Nature. 2012;486:207–214. doi: 10.1038/nature11234. [DOI] [PMC free article] [PubMed] [Google Scholar]
  30. Jain S, Self WH, Wunderink RG, Fakhran S, Balk R, Bramley AM, Reed C, Grijalva CG, Anderson EJ, Courtney DM, Chappell JD, Qi C, Hart EM, Carroll F, Trabue C, Donnelly HK, Williams DJ, Zhu Y, Arnold SR, Ampofo K, Waterer GW, Levine M, Lindstrom S, Winchell JM, Katz JM, Erdman D, Schneider E, Hicks LA, McCullers JA, Pavia AT, Edwards KM, Finelli L, CDC EPIC Study Team Community-Acquired pneumonia requiring hospitalization among U.S adults. New England Journal of Medicine. 2015;373:415–427. doi: 10.1056/NEJMoa1500245. [DOI] [PMC free article] [PubMed] [Google Scholar]
  31. Kabbani S, Hersh AL, Shapiro DJ, Fleming-Dutra KE, Pavia AT, Hicks LA. Opportunities to improve fluoroquinolone prescribing in the united states for adult ambulatory care visits. Clinical Infectious Diseases. 2018;67:134–136. doi: 10.1093/cid/ciy035. [DOI] [PMC free article] [PubMed] [Google Scholar]
  32. Khosravi A, Mazmanian SK. Disruption of the gut microbiome as a risk factor for microbial infections. Current Opinion in Microbiology. 2013;16:221–227. doi: 10.1016/j.mib.2013.03.009. [DOI] [PMC free article] [PubMed] [Google Scholar]
  33. King LM, Bartoces M, Fleming-Dutra KE, Roberts RM, Hicks LA. Changes in US outpatient antibiotic prescriptions from 2011–2016 . Clinical Infectious Diseases. 2020;316:370–377. doi: 10.1093/cid/ciz225. [DOI] [PMC free article] [PubMed] [Google Scholar]
  34. Klevens RM, Caten E, Olesen SW, DeMaria A, Troppy S, Grad YH. Outpatient antibiotic prescribing in Massachusetts, 2011-2015. Open Forum Infectious Diseases. 2019;6:ofz169. doi: 10.1093/ofid/ofz169. [DOI] [PMC free article] [PubMed] [Google Scholar]
  35. Lif Holgerson P, Öhman C, Rönnlund A, Johansson I. Maturation of oral Microbiota in children with or without dental caries. PLOS ONE. 2015;10:e0128534. doi: 10.1371/journal.pone.0128534. [DOI] [PMC free article] [PubMed] [Google Scholar]
  36. Linder JA. Editorial commentary: antibiotics for treatment of acute respiratory tract infections: decreasing benefit, increasing risk, and the irrelevance of antimicrobial resistance. Clinical Infectious Diseases. 2008;47:744–746. doi: 10.1086/591149. [DOI] [PubMed] [Google Scholar]
  37. Lipsitch M. The rise and fall of antimicrobial resistance. Trends in Microbiology. 2001;9:438–444. doi: 10.1016/S0966-842X(01)02130-8. [DOI] [PubMed] [Google Scholar]
  38. MacLean RC, San Millan A. The evolution of antibiotic resistance. Science. 2019;365:1082–1083. doi: 10.1126/science.aax3879. [DOI] [PubMed] [Google Scholar]
  39. Mainous AG, Hueston WJ, Everett CJ, Diaz VA. Nasal carriage of Staphylococcus aureus and methicillin-resistant S aureus in the united states, 2001-2002. The Annals of Family Medicine. 2006;4:132–137. doi: 10.1370/afm.526. [DOI] [PMC free article] [PubMed] [Google Scholar]
  40. National Center for Health Statistics National ambulatory medical care Survey—Public-use data file and documentation. [September 5, 2017];2015 ftp://ftp.cdc.gov/pub/Health_Statistics/NCHS/Datasets/NAMCS
  41. Olesen SW, Barnett ML, MacFadden DR, Brownstein JS, Hernández-Díaz S, Lipsitch M, Grad YH. The distribution of antibiotic use and its association with antibiotic resistance. eLife. 2018a;7:e39435. doi: 10.7554/eLife.39435. [DOI] [PMC free article] [PubMed] [Google Scholar]
  42. Olesen SW, Barnett ML, MacFadden DR, Lipsitch M, Grad YH. Trends in outpatient antibiotic prescribing practice among US older adults, 2011-2015: an observational study. BMJ. 2018b;362:k3155. doi: 10.1136/bmj.k3155. [DOI] [PMC free article] [PubMed] [Google Scholar]
  43. Pettigrew MM, Laufer AS, Gent JF, Kong Y, Fennie KP, Metlay JP. Upper respiratory tract microbial communities, acute otitis media pathogens, and antibiotic use in healthy and sick children. Applied and Environmental Microbiology. 2012;78:6262–6270. doi: 10.1128/AEM.01051-12. [DOI] [PMC free article] [PubMed] [Google Scholar]
  44. Pouwels KB, Batra R, Patel A, Edgeworth JD, Robotham JV, Smieszek T. Will co-trimoxazole resistance rates ever go down? resistance rates remain high despite decades of reduced co-trimoxazole consumption. Journal of Global Antimicrobial Resistance. 2017;11:71–74. doi: 10.1016/j.jgar.2017.07.013. [DOI] [PubMed] [Google Scholar]
  45. Public Health Agency of Canada . Canadian Antimicrobial Resistance Surveillance System 2017 Report. Public Health Agency of Canada; 2018. [Google Scholar]
  46. Public Health England . English Surveillance Programme for Antimicrobial Utilisation and Resistance (ESPAUR) 2017. Public Health England; 2017. [Google Scholar]
  47. Regev-Yochay G, Dagan R, Raz M, Carmeli Y, Shainberg B, Derazne E, Rahav G, Rubinstein E. Association between carriage of Streptococcus pneumoniae and Staphylococcus aureus in children. Jama. 2004;292:716–720. doi: 10.1001/jama.292.6.716. [DOI] [PubMed] [Google Scholar]
  48. Rosenfeld RM, Piccirillo JF, Chandrasekhar SS, Brook I, Ashok Kumar K, Kramper M, Orlandi RR, Palmer JN, Patel ZM, Peters A, Walsh SA, Corrigan MD. Clinical practice guideline (update): adult sinusitis. Otolaryngology–Head and Neck Surgery. 2015;152:S1–S39. doi: 10.1177/0194599815572097. [DOI] [PubMed] [Google Scholar]
  49. Sakoulas G, Geriak M, Nizet V. Is a reported penicillin allergy sufficient grounds to forgo the multidimensional antimicrobial benefits of β-Lactam antibiotics? Clinical Infectious Diseases. 2019;68:157–164. doi: 10.1093/cid/ciy557. [DOI] [PMC free article] [PubMed] [Google Scholar]
  50. Sande MA, Gwaltney JM. Acute community-acquired bacterial sinusitis: continuing challenges and current management. Clinical Infectious Diseases. 2004;39 Suppl 3:S151–S158. doi: 10.1086/421353. [DOI] [PubMed] [Google Scholar]
  51. Shehab N, Patel PR, Srinivasan A, Budnitz DS. Emergency department visits for antibiotic-associated adverse events. Clinical Infectious Diseases. 2008;47:735–743. doi: 10.1086/591126. [DOI] [PubMed] [Google Scholar]
  52. Smith SS, Evans CT, Tan BK, Chandra RK, Smith SB, Kern RC. National burden of antibiotic use for adult rhinosinusitis. Journal of Allergy and Clinical Immunology. 2013;132:1230–1232. doi: 10.1016/j.jaci.2013.07.009. [DOI] [PMC free article] [PubMed] [Google Scholar]
  53. Snow V, Mottur-Pilson C, Hickner JM, American Academy of Family Physicians, American College of Physicians-American Society of Internal Medicine, Centers for Disease Control, Infectious Diseases Society of America Principles of appropriate antibiotic use for acute sinusitis in adults. Annals of Internal Medicine. 2001;134:495. doi: 10.7326/0003-4819-134-6-200103200-00016. [DOI] [PubMed] [Google Scholar]
  54. Stewardson AJ, Gaïa N, François P, Malhotra-Kumar S, Delémont C, Martinez de Tejada B, Schrenzel J, Harbarth S, Lazarevic V. Collateral damage from oral ciprofloxacin versus nitrofurantoin in outpatients with urinary tract infections: a culture-free analysis of gut Microbiota. Clinical Microbiology and Infection. 2015;21:344.e1–34344. doi: 10.1016/j.cmi.2014.11.016. [DOI] [PubMed] [Google Scholar]
  55. Swedres-Svarm . Consumption of Antibiotics and Occurrence of Antibiotic Resistance in Sweden. Public Health Agency of Sweden and National Veterinary Institute; 2017. [Google Scholar]
  56. Tedijanto C, Olesen SW, Grad YH, Lipsitch M. Estimating the proportion of bystander selection for antibiotic resistance among potentially pathogenic bacterial flora. PNAS. 2018;115:E11988–E11995. doi: 10.1073/pnas.1810840115. [DOI] [PMC free article] [PubMed] [Google Scholar]
  57. Tobia CC, Aspinall SL, Good CB, Fine MJ, Hanlon JT. Appropriateness of antibiotic prescribing in veterans with community-acquired pneumonia, sinusitis, or acute exacerbations of chronic bronchitis: a cross-sectional study. Clinical Therapeutics. 2008;30:1135–1144. doi: 10.1016/j.clinthera.2008.06.009. [DOI] [PMC free article] [PubMed] [Google Scholar]
  58. Tomas ME, Getman D, Donskey CJ, Hecker MT. Overdiagnosis of urinary tract infection and underdiagnosis of sexually transmitted infection in adult women presenting to an emergency department. Journal of Clinical Microbiology. 2015;53:2686–2692. doi: 10.1128/JCM.00670-15. [DOI] [PMC free article] [PubMed] [Google Scholar]
  59. Vanden Eng J, Marcus R, Hadler JL, Imhoff B, Vugia DJ, Cieslak PR, Zell E, Deneen V, McCombs KG, Zansky SM, Hawkins MA, Besser RE. Consumer attitudes and use of antibiotics. Emerging Infectious Diseases. 2003;9:1128–1135. doi: 10.3201/eid0909.020591. [DOI] [PMC free article] [PubMed] [Google Scholar]
  60. Verhaegh SJ, Lebon A, Saarloos JA, Verbrugh HA, Jaddoe VW, Hofman A, Hays JP, Moll HA, van Belkum A. Determinants of Moraxella catarrhalis colonization in healthy dutch children during the first 14 months of life. Clinical Microbiology and Infection. 2010;16:992–997. doi: 10.1111/j.1469-0691.2009.03008.x. [DOI] [PubMed] [Google Scholar]
  61. Wald ER, Guerra N, Byers C. Upper respiratory tract infections in young children: duration of and frequency of complications. Pediatrics. 1991;87:129–133. [PubMed] [Google Scholar]
  62. Williams JW, Simel DL, Roberts L, Samsa GP. Clinical evaluation for sinusitis. Making the diagnosis by history and physical examination. Annals of Internal Medicine. 1992;117:705–710. doi: 10.7326/0003-4819-117-9-705. [DOI] [PubMed] [Google Scholar]
  63. Wubbel L, Muniz L, Ahmed A, Trujillo M, Carubelli C, McCoig C, Abramo T, Leinonen M, McCracken GH. Etiology and treatment of community-acquired pneumonia in ambulatory children. The Pediatric Infectious Disease Journal. 1999;18:98–104. doi: 10.1097/00006454-199902000-00004. [DOI] [PubMed] [Google Scholar]
  64. Yassour M, Vatanen T, Siljander H, Hämäläinen AM, Härkönen T, Ryhänen SJ, Franzosa EA, Vlamakis H, Huttenhower C, Gevers D, Lander ES, Knip M, Xavier RJ, DIABIMMUNE Study Group Natural history of the infant gut microbiome and impact of antibiotic treatment on bacterial strain diversity and stability. Science Translational Medicine. 2016;8:343ra81. doi: 10.1126/scitranslmed.aad0917. [DOI] [PMC free article] [PubMed] [Google Scholar]

Decision letter

Editor: Miles P Davenport1
Reviewed by: Miles P Davenport2, Marc Bonten

In the interests of transparency, eLife publishes the most substantive revision requests and the accompanying author responses.

Acceptance summary:

This manuscript provides a quantitative framework for assessing the impact of potential interventions to reduce unnecessary antibiotic use. It combines data on the prevalence of both infection-associated and colonizing bacterial populations with publicly available prescription data to assess the frequency of intentional vs. bystander antibiotic exposure in clinical practice. It then applies a modelling approach to understand how reduction in antibiotic prescriptions will affect overall antibiotic exposure across a variety of antibiotic-pathogen pairs. This extends previous work to provide further insights into the potential effectiveness of different antibiotic stewardship regimes.

Decision letter after peer review:

Thank you for submitting your article "Potential impact of outpatient stewardship interventions on antibiotic exposures of bacterial pathogens" for consideration by eLife. Your article has been reviewed by four peer reviewers, including Miles Davenport as the Reviewing Editor and Reviewer #1, and the evaluation has been overseen by a Reviewing Editor and Neil Ferguson as the Senior Editor. The following individuals involved in review of your submission have agreed to reveal their identity: Marc Bonten (Reviewer #3).

The reviewers have discussed the reviews with one another, and the Reviewing Editor has drafted this decision to help you prepare a revised submission.

Summary:

This manuscript looks at strategies to reduce antibiotic prescribing, and the role of 'bystander' exposure of commensals. The authors present a theoretical framework and nice thought experiment to quantify the potential benefits of changing antibiotic prescription in the outpatient setting. They have used "publicly available data to quantify potentially avertable exposures of bacterial species to commonly used antibiotics under hypothetical changes in prescribing practice." The four scenarios are relevant and "All scenarios demonstrate that stewardship interventions, facilitated by changes in clinician behaviour and improved diagnostics, have the opportunity to broadly reduce antibiotic exposures across a range of potential pathogens." The methodology builds upon previous work of the authors on the effects of antibiotics in the open population and the methods are clearly described. The limitations of the data regarding to infection coding and true antibiotic exposure and the uncertainty that any of the four strategies can be realized in real life are addressed appropriately. And despite the – somewhat – obvious conclusions ("less is more, but sometimes less more") this work presents a very thoughtful and interesting framework for future analyses.

The take home message is that exposure is reduced but would remain substantial, so outpatient stewardship alone is not enough.

Essential revisions:

1) The impact of changes prescribing strategy is interesting, but actually only relevant to the particular circumstance of bacteria that are associated with infection and either treated or not treated appropriately (see (a) below). This does not discount the importance of the insights – but in fact it feels that the manuscript presented numbers more than insights (see (b) below). Thus, the manuscript would be greatly strengthened and appeal to a much broader audience if the current results could be supplemented by some more intuitive explanations and ways to think about the problem.

a) While giving a percentage reduction is interesting, providing an intuition of how we might think about this and how it applies to different bacteria/drug pathogens combinations would be even more helpful. This is described to some extent in the Results section – but this was not an easy passage to follow unless you have a clear mental picture of pathogen/syndrome/antibiotic combinations, and could be explained much better in terms of 'classes' of pathogens. For example, if eliminating unnecessary use of penicillin for respiratory infection reduced overall penicillin prescriptions by 30%, then we can consider how this impact on three different classes of pathogen. (i) For unrelated pathogens (not specifically associated with respiratory infection), exposure is reduced by 30%. (ii) For pathogens that cause respiratory infection – but for which prescription is unnecessary and avoided by the strategy – they will have a >30% reduction (eg: overall reduction, + presence in infection * prescription rate (or some simple intuitive formula)). (iii) For pathogens that cause respiratory infection and for which penicillin is indicated, then they will have a lower reduction than 30% (ie: 30% – correct prescription rate).

This is not saying this issue is not described at all here (in formulas, and in the Results section) – just that it was not immediately clear that this fell out so easily, and the reader has to think about it. A flowchart-type figure of these classes and effects would make it very clear how the numbers in the tables were arrived at.

b) One thing that is not considered here is pathogen population size/density. For example, one might expect that the number of bacteria present in the microbiome may be many orders of magnitude lower than that present in the context of infection caused by the bacterium. Since the probability of resistance (and transmission of resistance) no doubt scales with population size, this seems important to discuss. This could be simply summarised as a 'ratio' even if not known precisely. That is, if the results were presented in the intuitive way described in (a) above, the issue of ratio could be introduced even without actual numbers of bacteria in microbiome/infection being known. For example, discussion of the impact of 'density' on the different scenarios above (eg: in scenario (ii) above, exposure per bacterium is reduced more than expected (because the treatments avoided are high density ones), whereas for scenario (iii) the reduction would be lower than calculated. (indeed, in the simple formula proposed above for a figure – the ratio of density in commensal carriage vs. infection could be added).

2) An additional paragraph on the relationship between antibiotic exposure and antibiotic resistance would be valuable. I realize the authors are avoiding that because it is not at all straightforward, but I think worth stating that a 46% exposure does not mean a 46% reduction in the frequency of resistance (let alone a 46% reduction in the morbidity/mortality associated with AMR). It might mean more or less; we have no idea. There is often a hidden assumption in the stewardship literature that if we can half exposure, we half the problem. The next important frontier is to figure out the relationship between the exposure and the resistance problem.

3) The four scenarios chosen are interesting and clinically relevant. However, they are all fairly different for one another in terms of clinical behavior change and potential implications. Therefore, it could be helpful to the reader if results for each scenario were presented separately in the results and discussion, with additional section tying all findings together and evaluating impact of all four scenarios together.

4) Additionally, for Scenario 1 it may be helpful to briefly define what is meant by unnecessary antibiotic use for readers unfamiliar with Fleming-Dutra et al. as it may be easy to confuse that with Scenario 2.

5) There is an issue of assuming "20/20 hindsight": the 'incorrect prescription' rate is based upon post-hoc analysis of the proportion of infections caused by a given pathogen. This information is not known to the clinician in advance. Therefore, there is no way to reach 'optimal targeting' in real-time prescribing. This should be discussed in more detail.

6) For acute sinusitis, it needs to be clarified that antibiotic treatment of acute bacterial sinusitis is currently guideline-recommended. Therefore, until guideline recommendations change, this scenario is unlikely to come to fruition. It may be helpful for the authors to consider a sensitivity analysis for acute sinusitis looking at (1) if only unnecessary use was eliminated (using Fleming-Dutra et al. definition) and (2) if all treatment was eliminated (current analysis). Some of this already exists in the supplement (Figure 1—figure supplement 2) and main text, so little additional analysis would be needed.

7) Consider being more explicit about codes/diagnoses are being considered in Scenarios 1, 2, and 4. Perhaps including a table in the supplement would be a nice way to summarize this.

8) The incorporation of sensitivity analyses and discussion of outpatient departments is unnecessary and may be confusing to readers. Many outpatient settings are excluded from NAMCS/NHAMCS. Just mention that in the methods and limitations.

eLife. 2020 Feb 5;9:e52307. doi: 10.7554/eLife.52307.sa2

Author response

Essential revisions:

1) The impact of changes is prescribing strategy is interesting, but actually only relevant to the particular circumstance of bacteria that are associated with infection and either treated or not treated appropriately (see (a) below). This does not discount the importance of the insights – but in fact it feels that the manuscript presented numbers more than insights (see (b) below). Thus, the manuscript would be greatly strengthened and appeal to a much broader audience if the current results could be supplemented by some more intuitive explanations and ways to think about the problem.

a) While giving a percentage reduction is interesting, providing an intuition of how we might think about this and how it applies to different bacteria/drug pathogens combinations would be even more helpful. This is described to some extent in the Results section – but this was not an easy passage to follow unless you have a clear mental picture of pathogen/syndrome/antibiotic combinations, and could be explained much better in terms of 'classes' of pathogens. For example, if eliminating unnecessary use of penicillin for respiratory infection reduced overall penicillin prescriptions by 30%, then we can consider how this impact on three different classes of pathogen. (i) For unrelated pathogens (not specifically associated with respiratory infection), exposure is reduced by 30%. (ii) For pathogens that cause respiratory infection – but for which prescription is unnecessary and avoided by the strategy – they will have a >30% reduction (eg: overall reduction, + presence in infection * prescription rate (or some simple intuitive formula)). (iii) For pathogens that cause respiratory infection and for which penicillin is indicated, then they will have a lower reduction than 30% (ie: 30% – correct prescription rate).

This is not saying this issue is not described at all here (in formulas, and in the Results section) – just that it was not immediately clear that this fell out so easily, and the reader has to think about it. A flowchart-type figure of these classes and effects would make it very clear how the numbers in the tables were arrived at.

We thank the reviewers for their constructive feedback and agree that broader, more intuitive statements may be useful for a wider audience. Describing pathogen classes is slightly more complicated than outlined above because exposures depend not only on etiology for the condition of interest (e.g. respiratory infection) and carriage, but also on the propensity of an organism to cause other infections. In addition, whether antibiotics are or are not indicated does not depend entirely on the causative pathogen. These complexities make it difficult to put forth simple formulas for the proportion of avertable exposures by pathogen “class”, but we have included a set of more general statements at the end of the Results section. We hope that this approach will provide readers with the desired intuition.

We have also added an additional source data file (Figure 1—source data 3) including tables of minimally processed data inputs to help facilitate better understanding of the underlying calculations and enable further exploration/manipulation by readers.

b) One thing that is not considered here is pathogen population size/density. For example, one might expect that the number of bacteria present in the microbiome may be many orders of magnitude lower than that present in the context of infection caused by the bacterium. Since the probability of resistance (and transmission of resistance) no doubt scales with population size, this seems important to discuss. This could be simply summarised as a 'ratio' even if not known precisely. That is, if the results were presented in the intuitive way described in (a) above, the issue of ratio could be introduced even without actual numbers of bacteria in microbiome/infection being known. For example, discussion of the impact of 'density' on the different scenarios above (eg: in scenario (ii) above, exposure per bacterium is reduced more than expected (because the treatments avoided are high density ones), whereas for scenario (iii) the reduction would be lower than calculated. (indeed, in the simple formula proposed above for a figure – the ratio of density in commensal carriage vs. infection could be added).

We thank the reviewers for their comment and acknowledge that bacterial density could play a role in propagation of resistance. However, we do not believe there is enough data to support the idea that an exposure during treatment is more important for resistance than an exposure experienced during colonization. There are also reasons to believe the opposite – for example, infections reflect bacteria that found their way into normally sterile sites, and resistant strains or their genes may find it harder to transmit from these sites (e.g. the bloodstream is thought to be a “dead end for evolution” (Levin and Bull, 1994). There is insufficient evidence to conclude whether infection is associated with higher loads of the causative pathogen throughout the body. In addition, many other variables beyond density may modulate the interaction between antibiotics and infecting and colonizing populations of bacteria. For example, pharmacokinetics, pharmacodynamics, the nature of the particular infection (e.g. abscess that is hard for antibiotics to penetrate) and the non-uniform distribution of bacteria across body sites may also play a role.

We have added several lines to the Discussion section to acknowledge that population size/density, along with other factors mentioned above, may impact our findings.

2) An additional paragraph on the relationship between antibiotic exposure and antibiotic resistance would be valuable. I realize the authors are avoiding that because it is not at all straightforward, but I think worth stating that a 46% exposure does not mean a 46% reduction in the frequency of resistance (let alone a 46% reduction in the morbidity/mortality associated with AMR). It might mean more or less; we have no idea. There is often a hidden assumption in the stewardship literature that if we can half exposure, we half the problem. The next important frontier is to figure out the relationship between the exposure and the resistance problem.

We thank the reviewers for raising this important point. We have added a paragraph on this issue to the Discussion section.

3) The four scenarios chosen are interesting and clinically relevant. However, they are all fairly different for one another in terms of clinical behavior change and potential implications. Therefore, it could be helpful to the reader if results for each scenario were presented separately in the results and discussion, with additional section tying all findings together and evaluating impact of all four scenarios together.

As recommended, the Results section has been separated by scenario. We hope this will help improve clarity for readers. Most of the Discussion section is broadly applicable and serves to tie the scenarios together.

4) Additionally, for Scenario 1 it may be helpful to briefly define what is meant by unnecessary antibiotic use for readers unfamiliar with Fleming-Dutra et al. as it may be easy to confuse that with Scenario 2.

We thank the reviewers for raising this potential point of confusion. A brief paragraph has been added to the Materials and methods section to define unnecessary antibiotic use as estimated by Fleming-Dutra et al.

5) There is an issue of assuming "20/20 hindsight": the 'incorrect prescription' rate is based upon post-hoc analysis of the proportion of infections caused by a given pathogen. This information is not known to the clinician in advance. Therefore, there is no way to reach 'optimal targeting' in real-time prescribing. This should be discussed in more detail.

We thank the reviewers for this important point. We have added “laboratory processing times that prevent identification of causative pathogens during outpatient visits” to the list of issues that preclude optimal prescribing (Discussion section); another related point that is already listed is “similar clinical presentation of viral and bacterial infections”. We also include rapid diagnostics as a key tool for reducing unnecessary antibiotic use (Discussion section).

6) For acute sinusitis, it needs to be clarified that antibiotic treatment of acute bacterial sinusitis is currently guideline-recommended. Therefore, until guideline recommendations change, this scenario is unlikely to come to fruition. It may be helpful for the authors to consider a sensitivity analysis for acute sinusitis looking at (1) if only unnecessary use was eliminated (using Fleming-Dutra et al. definition) and (2) if all treatment was eliminated (current analysis). Some of this already exists in the supplement (Figure 1—figure supplement 2) and main text, so little additional analysis would be needed.

We have added this analysis as a third scenario in Figure 1—figure supplement 2 and clarified that antibiotic treatment of acute bacterial sinusitis is currently guideline-recommended in the text (subsection “Scenarios of interest”). Based on the Fleming-Dutra et al. paper, we estimate that 18% (0-19 years old), 100% (20-64 years old), and 34% (>64 years old) of antibiotic prescriptions given for acute sinusitis are unnecessary. Note that these estimates do not align directly with the numbers from the Fleming-Dutra paper (9% for 0-19 years old, 51% for 20-64 years old, and 16% for >64 years old) because the Fleming-Dutra paper groups together prescriptions for acute and chronic sinusitis. Although chronic sinusitis arises from multiple pathophysiologic processes, antibiotics may be indicated for chronic sinusitis patients with acute exacerbations and/or signs of infection. Thus, we back-calculate the modified proportions of unnecessary prescriptions for acute sinusitis alone under the simplifying assumption that all prescriptions for chronic sinusitis are necessary (see Figure 1—source data 2 for more details).

7) Consider being more explicit about codes/diagnoses are being considered in scenarios 1, 2, and 4. Perhaps including a table in the supplement would be a nice way to summarize this.

Explicit ICD-9-CM codes for each condition in Scenarios 1 and 2 can be found in eTable 2 of Fleming-Dutra et al., 2016. We describe in the text that we have directly applied their coding methodology: “We used methodology developed by Fleming-Dutra and colleagues, and applied in other studies, to group diagnosis codes into conditions…” (subsection “Data sources and methodology”). Codes for Scenario 4 are written out in the text in the Materials and methods section.

8) The incorporation of sensitivity analyses and discussion of outpatient departments is unnecessary and may be confusing to readers. Many outpatient settings are excluded from NAMCS/NHAMCS. Just mention that in the methods and limitations.

We apologize for any confusion this sensitivity analysis and discussion caused. The purpose of the sensitivity analysis in Figure 1—figure supplement 3 was not explicitly to explore the inclusion/exclusion of specific outpatient departments, but rather to explore validity of using the more recent 2015 data which is missing hospital outpatient department visits (over one-third of all sampled visits in the 2010-2011 data). Because we observed that the 2010-2011 results were similar with and without this segment of visits, we assumed the antibiotic use and diagnosis data from 2015 were representative of the entire NAMCS/NHAMCS population for the purposes of our metric.

To simplify, we have maintained only Scenario 1 in the sensitivity analysis and have added more detail in the Discussion section to explain this Figure supplement. We have also briefly mentioned the limitation of missing outpatient settings in NAMCS/NHAMCS (Discussion section).

Associated Data

    This section collects any data citations, data availability statements, or supplementary materials included in this article.

    Data Citations

    1. National Center for Health Statistics. 2015. National Hospital Ambulatory Medical Care Survey. National Center for Heath Statistics. ED2015-stata
    2. National Center for Health Statistics. 2015. National Ambulatory Medical Care Survey. National Center for Heath Statistics. namcs2015-stata

    Supplementary Materials

    Figure 1—source data 1. Summary of etiology studies.
    elife-52307-fig1-data1.docx (34.2KB, docx)
    Figure 1—source data 2. Modifications to proportions of unnecessary antibiotic prescriptions based on published etiology studies and antibiotic use in NAMCS/NHAMCS.

    Modifications were made if the reported proportion of bacterial cases for a given condition exceeded the estimated proportion of appropriate prescriptions reported in Fleming-Dutra et al. (2016).

    elife-52307-fig1-data2.docx (20.9KB, docx)
    Figure 1—source data 3. Estimated antibiotic class exposures by exposed and causative species.
    elife-52307-fig1-data3.docx (1.2MB, docx)
    Transparent reporting form
    elife-52307-transrepform.pdf (323.9KB, pdf)

    Data Availability Statement

    Data from the 2015 National Ambulatory Medical Care Survey (NAMCS) and National Hospital Ambulatory Medical Care Survey (NHAMCS) are publicly available from the National Center for Health Statistics. This study also uses data from published literature, including the Human Microbiome Project and other studies summarized in Figure 1—source data 1.

    The following previously published datasets were used:

    National Center for Health Statistics. 2015. National Hospital Ambulatory Medical Care Survey. National Center for Heath Statistics. ED2015-stata

    National Center for Health Statistics. 2015. National Ambulatory Medical Care Survey. National Center for Heath Statistics. namcs2015-stata

    Articles from eLife are provided here courtesy of eLife Sciences Publications, Ltd

    RESOURCES