Introduction

Antibiotics are over-prescribed in ambulatory care, increase the risk of adverse events, and promote the development of antibiotic-resistant bacteria [1–9]. Antibiotic stewardship interventions aim to reduce inappropriate antibiotic prescribing to improve patient safety and, more broadly, maintain the effectiveness of antibiotics [10, 11]. Many ambulatory antibiotic stewardship studies, measures, and recommendations examine or intervene on antibiotic prescribing at ambulatory in-person visits [10, 12–20].

However, antibiotics can be prescribed without in-person visits, for example when patients and clinicians communicate via telephone call or electronic patient portal messages. Antibiotic prescribing at these “not-in-person encounters” are overlooked in many stewardship initiatives and in research. Not-in-person encounters may also have lower quality because of an inability to do a physical examination. Prior publications have examined the appropriateness of outpatient antibiotic prescribing across all settings using claims data, but these studies did not distinguish between antibiotics prescribed with and without in-person encounters [5, 21–23].

Our objectives were, among all antibiotics prescribed within a large integrated health system using an electronic health record (EHR), to 1) assess the prevalence of not-in-person antibiotic prescriptions and 2) assess the appropriateness of antibiotic prescribing across both in-person and not-in-person encounters, thus providing a comprehensive picture of outpatient antibiotic prescribing [24]. We used EHR data because they have more detailed information on encounter type compared with claims data. Moreover, the indication for antibiotics can more readily be ascertained using EHR data, using problem lists and diagnosis codes.

Materials and methods

Results

In the four-year study period from 2016 to 2019, there were 714,057 antibiotic prescriptions ordered for 348,739 unique patients by 2,391 clinicians in 467 clinics. The most common antibiotic classes were penicillins (31%), macrolides (20%), tetracyclines (11%), and cephalosporins (10%). Patients were 61% female with a mean age of 41 (SD±22) years old; 52% had private insurance (Table 1). Over half (59%) of prescribing clinicians were physicians, 42% practiced in primary care, and 30% were > 0.5 ambulatory clinical full time equivalent.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Patient, clinician, and physician characteristics.

https://doi.org/10.1371/journal.pone.0289303.t001

In-person versus not-in-person prescriptions

Although the groups were not independent, patients who received any not-in-person antibiotic prescriptions were more likely to be older, women, married/partnered, have Medicare, take more medication, make more visits, have a primary care physician, and be hospitalized during the study period compared with patients who received any in-person antibiotic prescription (Table 1). Seventy three percent of clinicians prescribed both in-person and not-in-person antibiotics during the study period; 27% prescribed only one or the other.

Over the four-year study period, 81% (n = 578,130) antibiotics were prescribed with an in-person encounter and 19% (n = 135,927) with a not-in-person encounter (Table 2). The most common types of not-in-person encounters in which antibiotics were prescribed were telephone (10% of all antibiotics prescribed with not-in-person encounters), orders-only (5%), and refill (3%) encounters.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Antibiotic prescriptions by encounter type and diagnosis category (N = 714,057).

https://doi.org/10.1371/journal.pone.0289303.t002

Antibiotic appropriateness.

The 5 most common diagnoses in each appropriateness category are listed in the footnote of Table 2. Across all encounter types, chronic antibiotic use accounted for 16% of all antibiotic prescriptions, while antibiotic-appropriate and potentially antibiotic-appropriate prescriptions collectively accounted for 54% (Table 2). Non-antibiotic-appropriate diagnoses accounted for 22% of antibiotic prescriptions. Eight percent of antibiotic prescriptions were associated with no diagnosis. Thus, 30% of antibiotics were either non-antibiotic-appropriate or not associated with any diagnosis.

Antibiotics prescribed in not-in-person encounters were more likely to be chronic (20% versus 15% compared to antibiotics prescribed with in-person visits). Among not-in-person encounter types, antibiotic prescriptions were most likely to be for chronic use among refill (29%) and patient portal (22%) encounters. Not-in-person antibiotics were less likely to be associated with appropriate or potentially appropriate diagnoses (30% versus 59% for in-person visits). Not-in-person antibiotics, depending on the specific encounter type, were variably less likely to be associated with antibiotic-appropriate diagnoses. Not-in-person antibiotics were less likely associated with potentially-antibiotic-appropriate diagnoses (20% versus 43%). Except for the relatively rare “other” not-in-person encounters, not-in-person encounters were less likely to be associated with non-antibiotic appropriate diagnoses (8% versus 25%). Not-in-person encounters were more likely associated with no diagnosis code (42% versus <1%). Said another way, 58% of not-in-person encounters had a diagnosis whereas 99% of in-person encounters had a diagnosis.

Discussion

In this study of four years of antibiotic prescribing data within an integrated academic health system, we found that 19% of ambulatory antibiotics were prescribed with not-in-person encounters. Antibiotics prescribed without in-person visits were more likely for chronic diagnoses; less likely to be associated with appropriate or potentially appropriate diagnoses; and more likely to be associated with no diagnosis. These findings suggest that ambulatory stewardship interventions that focus only on in-person visits may miss a large proportion of antibiotic prescribing, inappropriate prescribing, and antibiotics prescribed in the absence of any diagnosis. The absence of a diagnosis may say more about inappropriate documentation than inappropriate antibiotic prescribing. However, if one considered the absence of a diagnosis to be inappropriate prescribing, 25% of in-person antibiotics were inappropriate and 50% of not-in-person antibiotics were inappropriate.

Our findings are particularly important considering the COVID-19 pandemic. During the pandemic, there were shifts in visit numbers: in-person visits have become less frequent and use of not-in-person encounters have become more common [27]. At the study health system, for example, there are now many more synchronous telephone and video visits that require a billing code. There were also pandemic shifts in antibiotic prescribing [28–30].

Familiarity with remote care and improved reimbursement for synchronous, remote visits has the potential to improve all-ambulatory-antibiotic stewardship. Alternatively, increased remote care has the potential to be associated with decreased antibiotic prescribing quality. Analyses of all-antibiotic prescribing, appropriateness, and visit type should be repeated and the results validated in the COVID pandemic and post-pandemic periods. Given the shifts in visit types and antibiotic prescribing, a continued focus only on antibiotic prescribing during in-person visits or solely on telehealth visits is less and less justifiable. Similarly, studies that only focus on telemedicine may only capture a fraction of overall antibiotic prescribing at a health system [31–33].

Our appropriateness findings are similar to what Fleming-Dutra and colleagues reported using nationally representative survey data from the United States. After sampling from ambulatory visits with oral antibiotic prescriptions they found an estimated annual antibiotic prescription rate of 506 per 1000 population and an estimated 353 (70%) were likely clinically appropriate [34]. We have conducted studies using claims data and found between 26% and 29% were not associated with a recent diagnosis code [5, 21], quite a bit higher than the 8% from the present study. Regarding not-in-person encounters, we previously found that about 30% of antibiotic fills were “non-visit-based” [22, 23] which is more than the 19% of not-in-person encounters in the present study. A key difference is that in the prior studies we had a more restrictive definition of chronic antibiotic prescribing, based on long-term dispensing. In the present study, to be more lenient to clinicians, we initially associated antibiotic prescriptions with chronic conditions based on 6 months of diagnoses. Claims and EHR data might differentially detect antibiotics: antibiotics might be prescribed in the EHR, but not generate a claim (i.e., the patient paid cash or did not pick up the antibiotic) or clinicians could circumvent EHR prescribing by calling a pharmacy directly. Compared to claims data, EHR data may be more likely to demonstrate the relationships among prescribers, antibiotics, sites of care, and diagnoses. Claims data may rely on different information sources for antibiotic claims, prescriber information, and diagnosis codes and overstate the “inappropriateness” of antibiotic prescribing relative to EHR data.

Our study is subject to several limitations beyond it being conducted prior to the pandemic. First, it was conducted within a single integrated health delivery system which may limit generalizability. Second, the data we analyzed were dependent on clinician diagnosis code selection within the EHR which is sometimes an incomplete reflection of actual care. Third, we were limited to prescribing data (e.g., clinician behavior) and did not have data regarding if or when patients filled and took their prescribed antibiotic (e.g., patient behavior).

Fourth, in determining appropriateness, we erred on the side of leniency. For example, we considered all antibiotic prescribing for chronic conditions to be acceptable (e.g., emphysema, chronic sinusitis, and chronic rhinitis) even though the prescribing may not have been clinically appropriate. We also forgivingly assumed that the most antibiotic-appropriate diagnosis was the reason for the antibiotic prescription, did not require the antibiotic class to be appropriate for a given condition, and could use a single diagnosis code to justify the appropriateness of multiple antibiotics. Thus, our study’s estimate of inappropriate prescribing is undoubtedly a lower bound.

Fifth, our approach assumed that antibiotic prescriptions on different days resulted from a separate episode of care, but this may not always be the case. For example, if a follow-up telephone visit by a patient or pharmacist to clarify an antibiotic written at an in-person visit resulted in a new antibiotic prescription, this would have been counted as two separate prescriptions. However, any resulting bias is likely minimal, as only 2% of antibiotic prescriptions had another antibiotic prescription in the subsequent 1 to 3 days. Sixth, it was beyond the scope of the current study to examine prescriber-level variability in antibiotic appropriateness, including whether certain prescribers disproportionately accounted for not-in-person antibiotic prescribing that lacked diagnosis codes or whether prescribers had different patterns of appropriateness across their in-person and not-in-person encounters. Seventh, direct comparisons of appropriateness between prescriptions written with in-person visits versus not-in-person encounters were limited by the frequent lack of diagnosis codes among prescriptions at not-in-person encounters. However, the lack of an associated diagnosis code could itself be viewed as inappropriate or a quality problem that would limit the effectiveness of ambulatory stewardship programs. Additional research is needed to examine the rationale of antibiotics prescribing not associated with a diagnosis code.

In summary, to decrease unnecessary antibiotic prescribing, improve patient safety, and slow the development of antibiotic-resistant bacteria, ambulatory antibiotic stewardship interventions should comprehensively measure all antibiotic prescribing, not just prescribing at in-person visits. The importance of considering prescribing across all encounters may only grow if ongoing shifts towards remote care continue.

Supporting information