-
PDF
- Split View
-
Views
-
Cite
Cite
Alisa Hamilton, Suprena Poleon, Jerald Cherian, Sara Cosgrove, Ramanan Laxminarayan, Eili Klein, COVID-19 and Outpatient Antibiotic Prescriptions in the United States: A County-Level Analysis, Open Forum Infectious Diseases, Volume 10, Issue 3, March 2023, ofad096, https://doi.org/10.1093/ofid/ofad096
- Share Icon Share
Abstract
Declines in outpatient antibiotic prescribing were reported during the beginning of the coronavirus disease 2019 (COVID-19) pandemic in the United States; however, the overall impact of COVID-19 cases on antibiotic prescribing remains unclear.
This was an ecological study using random-effects panel regression of monthly reported COVID-19 county case and antibiotic prescription data, controlling for seasonality, urbanicity, health care access, nonpharmaceutical interventions (NPIs), and sociodemographic factors.
Antibiotic prescribing fell 26.8% in 2020 compared with prior years. Each 1% increase in county-level monthly COVID-19 cases was associated with a 0.009% (95% CI, 0.007% to 0.012%; P < .01) increase in prescriptions per 100 000 population dispensed to all ages and a 0.012% (95% CI, −0.017% to −0.008%; P < .01) decrease in prescriptions per 100 000 children. Counties with schools open for in-person instruction were associated with a 0.044% (95% CI, 0.024% to 0.065%; P < .01) increase in prescriptions per 100 000 children compared with counties that closed schools. Internal movement restrictions and requiring facemasks were also associated with lower prescribing among children.
The positive association of COVID-19 cases with prescribing for all ages and the negative association for children indicate that increases in prescribing occurred primarily among adults. The rarity of bacterial coinfection in COVID-19 patients suggests that a fraction of these prescriptions may have been inappropriate. Facemasks and school closures were correlated with reductions in prescribing among children, possibly due to the prevention of other upper respiratory infections. The strongest predictors of prescribing were prior years’ prescribing trends, suggesting the possibility that behavioral norms are an important driver of prescribing practices.
Antibiotic prescribing rates in the United States dropped significantly in 2020 during the first several months of the coronavirus disease 2019 (COVID-19) pandemic, likely due to reduced health care–seeking behavior [1–3]. Throughout 2020, antibiotic prescribing continued to be lower than in the corresponding months in prior years [1–3]. There are 2 main reasons that COVID-19 cases may have altered prescribing. First, COVID-19 cases may have been inappropriately treated with antibiotics, even though antibiotic therapy was not indicated for severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2). Evidence indicates that <5% of hospitalized COVID-19 patients had confirmed bacterial coinfections in 2020, yet 40%–90% were prescribed antibacterial therapy [4–7]. Inappropriate prescribing was also common prepandemic, with an estimated 30% of outpatient antibiotic prescriptions considered inappropriate by guidelines [8]. Second, nonpharmaceutical interventions (NPIs) may have reduced transmission of other URIs (eg, the cold, influenza, and streptococcus). The winter of 2020 saw unseasonably low rates of these URIs [9, 10], peaks in which are typically associated with increased prescribing [11].
Here we estimate the relative importance of COVID-19 cases, NPIs, and sociodemographic factors in driving antibiotic prescribing. The results are important for understanding the differential and potential impact of measures to control the spread of infections, which can aid policy-makers in planning for seasonal epidemics. Additionally, ecological studies provide insight into the structural and geographic factors that drive variation in prescribing. Because of the causal link between antibiotic use and resistance, identifying potential targets to reduce prescribing can aid in public health efforts to mitigate antibiotic resistance.
METHODS
Data Sources
Data on systemic antibiotic prescriptions (ATC code J01) collected from retail pharmacies in the United States from January through December were obtained from IQVIA's Xponent database for the period from 2017 to 2020 [12]. The database contains the total number of outpatient prescriptions and the quantity of each antibiotic dispensed at the zip code and month level disaggregated by age group and sex. IQVIA data have been used extensively in prior studies of antibiotic prescribing [13, 14] and cover systemic antibiotics administered orally, intravenously, or via aerosol. We excluded topical agents that are not systemically absorbed (eg, Bacitracin) and medications not recommended for treatment of respiratory infections in the outpatient setting (eg, cefepime, vancomycin, and telavancin) (Supplementary Table 1). Zip code–level data were aggregated to the county level using the HUD USPS Zip Code Crosswalk [15]. In addition, because defined daily doses (DDDs) are a standardized international measure of drug consumption, they were included as a sensitivity analysis. DDDs were calculated from drug quantity, form, and strength using the Anatomical Therapeutic Chemical Classification System (ACT/DDD 2020) [16]. For drugs with no defined DDD, we estimated the common prescription dose from consumer packaging information (Supplementary Table 2).
Data on COVID-19 cases were obtained from the New York Times [17] via the Dartmouth Data Analytic Core (DAC) [18]. The New York Times provides county-level data with the exceptions of New York City and Kansas City. DAC's publicly available code apportions cases to separate counties, rather than grouping them, for these geographic anomalies. The monthly number of cases in each county was calculated as the difference between cumulative new cases at the start and end of each month. To adjust for differences in testing rates that can affect the number of reported positive cases, state-level COVID-19 testing data were obtained from the COVID Tracking Project [19] via the Johns Hopkins Coronavirus Resource Center [16]. We included the cumulative number of combined tests, which includes both antigen and viral tests, per 100 000 population at the end of each month.
To standardize case rates and control for county- and state-level differences in demographics and health care access [20–22], population data, including each county's total population [23], age group–stratified population, the percentage of the population of people of color [24], the percentage of the population living in poverty [25], and the number of physicians’ offices by county [26], were obtained from the United States Census Bureau. Counties were further classified by urbanization level based on the 2013 National Center for Health Statistics (NCHS) urban-rural scheme (Supplementary Table 3) [27].
To assess the impact of NPIs, state-level mask-wearing and internal movement restriction data were obtained from the Oxford COVID-19 Government Response Tracker (OxCGRT) [28]. OxCGRT collects publicly available data from news media and government press releases on several NPI indicators and was used to classify movement restrictions into 2 categories: “restrictions in place” and “no restrictions/recommended not travel between regions or cities.” For facial coverings, data were recategorized into 2 categories: “required in all places outside the home” and “no policy/recommended/required in some places.” Movement restriction and facial covering data were grouped at the state-month level as county-level data were unavailable.
County-level K-12 school data were obtained from MCH Strategic Data [29]. MCH school data classified each school district's 2020 opening date by teaching method (“on premises,” “online only,” “hybrid,” “unknown,” and “other”). The teaching method was recategorized into 3 groups—“open” (in-person instruction), “closed” (summer vacation or online only), and “hybrid/other/unknown”—then aggregated to the county-month level based on the teaching method with the highest enrollment for each county-month. All counties were assigned “closed” between March and the opening month. Forty-eight states closed schools for in-person instruction for the remainder of the school year after the spring 2020 outbreak [30]. As a sensitivity analysis, data on state-level school closures were obtained from EducationWeek [31], which tracked state mandates on K-12 in-person instruction from August 2020 through June 2021. EducationWeek classified school closures as “no order,” “full closure,” “partial closure,” “ordered open,” or “some grades ordered open.” Data were recategorized into 3 groups—“ordered open,” “closed,” or “hybrid/partially closed/no order”—and grouped by state-month based on the category most frequently reported for the month.
Analysis
We first compared national and state-level monthly number of prescriptions per 100 000 dispensed in 2020 with a baseline average from 2017 to 2019 and disaggregated by antibiotic class and age group. We then conducted a panel regression at the county level to assess associations between COVID-19 cases and prescribing. We used random effects to avoid collinearity among time-invariant variables with fixed effects, and we used 1-way analysis of variance to test between-group differences in mean 2020 monthly prescriptions per 100 000 and monthly COVID-19 cases per 100 000 for each categorical variable. We included state as a random-level intercept to account for geographic differences in pandemic response (eg, the extent of lockdowns, business closures, and caps on indoor capacity). Data on average 2017–2019 prescriptions by month were included to control for county-level prescribing behavior, and calendar month was included to capture seasonal differences in prescribing trends [13], climatological factors (eg, absolute humidity) associated with increases in URIs [32, 33], and temporal variations in government responses and pandemic-related behavioral changes. Because children likely had the greatest change in contact patterns during the pandemic with the closing of schools, we conducted a subanalysis with the same model except that the outcome variable was the number of prescriptions dispensed to children aged 0–9.
Data were processed using Python 3.8, visualizations were created in RStudio 1.3, and statistical analysis was conducted in Stata 16.1. As all data obtained were aggregated, de-identified, publicly available data, the analysis did not constitute human subjects research as defined by 45 CFR 46.102 and did not require institutional review board (IRB) review.
RESULTS
Overall Prescribing Changes Between 2020 and Prior Years
Total antibiotic prescriptions fell 26.8% between March and December 2020 compared with the same period from 2017 to 2019 (Figure 1A). Prescriptions fell significantly in April and May 2020. While there was some rebound in June and July, prescribing remained below average between July and December. States in the Southeast had the lowest percentage declines in prescribing from prior years (Figure 1B; Supplementary Table 4). Overall prescribing of quinolones dropped 28.9% from a mean of 6738.3 prescriptions per 100 000 in previous years to 4787.1 per 100 000 in 2020 (Figure 2A). In comparison, tetracyclines dropped 3.13%, while other antibiotics showed negligible changes. The largest change in prescriptions by age group was observed among children 0–2 years old, dropping 49.6% from 95 804.6 per 100 000 in previous years to 48 276.0 per 100 000 in 2020 (Figure 2B).

Number of antibiotic prescriptions dispensed per 100 000 population from 2017–2020. A, Total antibiotic prescriptions dispensed per 100 000 population by month for each year from 2017–2020. B, Total percent change in antibiotic prescriptions dispensed from the mean number of antibiotic prescriptions dispensed per 100 000 population in 2017–2019 compared with 2020 by state.

Number of antibiotic prescriptions dispensed per 100 000 population in 2020 vs 2017–2019 average by antibiotic class and age group. A, Mean number of antibiotic prescriptions dispensed per 100 000 population for the most widely used antibiotic classes from 2017–2019 compared with the number of antibiotic prescriptions dispensed per 100 000 population for the same classes in 2020. B, Mean number of antibiotic prescriptions dispensed per 100 000 population by age group from 2017–2019 compared with the number of antibiotic prescriptions dispensed per 100 000 population for the same age groups in 2020.
Descriptive Analysis
From January through December 2020, the mean monthly number of prescriptions per 100 000 county residents was 4266.9 (95% CI, 4235.4 to 4298.4) compared with 5452.4 (95% CI, 5415.2 to 5489.6) in previous years (Table 1). Noncore counties had the lowest mean monthly prescriptions per capita among all ages, while large fringe metro counties had the lowest among children (Supplementary Table 5). County-months where schools were open had a higher mean number of prescriptions compared with county-months where schools were closed (Supplementary Table 5). Lower mean prescriptions per capita were observed in county-months with internal movement restrictions and facial covering requirements in place, respectively (Supplementary Table 5).
Descriptive Statistics Using County-Level Number of Prescriptions Dispensed, United States, January–December 2020
Continuous Variables . | Mean (95% CI) . |
---|---|
Monthly COVID-19 cases per 100 000 population | 558.4 (549.4 to 567.4) |
Monthly COVID-19 tests per 100 000 state residents | 22 036.1 (21 771.3 to 22 301.0) |
2020 monthly TRX per 100 000 population | 4266.9 (4235.4 to 4298.4) |
2017–2019 monthly TRX per 100 000 population | 5452.4 (5415.2 to 5489.6) |
2020 monthly TRX per 100 000 children 0–9 | 3564.0 (3523.4 to 3604.5) |
2017–2019 monthly TRX per 100 000 children 0–9 | 6572.3 (6514.8 to 6629.8) |
Physicians’ offices per 100 000 population | 35.1 (34.7 to 35.5) |
Percentage of population in poverty | 14.4 (14.4 to 14.5) |
Percentage of population people of color | 15.6 (15.5 to 15.8) |
Categorical Variables | No. (%) |
School status | |
Closed | 18 672 (50.7) |
Hybrid/other/unknown | 8451 (23.0) |
Open | 9705 (26.4) |
Internal movement restrictions | |
No restrictions or recommended | 35 209 (93.4) |
Restrictions in place | 2507 (6.7) |
Facial coverings | |
No policy or recommended | 36 779 (97.5) |
Required in some/all places outside the home | 937 (2.5) |
Urbanization level | |
Large central metro | 816 (2.2) |
Large fringe metro | 4416 (11.7) |
Medium metro | 4476 (11.9) |
Small metro | 4296 (11.4) |
Micropolitan | 7692 (20.4) |
Noncore | 16 020 (42.5) |
Continuous Variables . | Mean (95% CI) . |
---|---|
Monthly COVID-19 cases per 100 000 population | 558.4 (549.4 to 567.4) |
Monthly COVID-19 tests per 100 000 state residents | 22 036.1 (21 771.3 to 22 301.0) |
2020 monthly TRX per 100 000 population | 4266.9 (4235.4 to 4298.4) |
2017–2019 monthly TRX per 100 000 population | 5452.4 (5415.2 to 5489.6) |
2020 monthly TRX per 100 000 children 0–9 | 3564.0 (3523.4 to 3604.5) |
2017–2019 monthly TRX per 100 000 children 0–9 | 6572.3 (6514.8 to 6629.8) |
Physicians’ offices per 100 000 population | 35.1 (34.7 to 35.5) |
Percentage of population in poverty | 14.4 (14.4 to 14.5) |
Percentage of population people of color | 15.6 (15.5 to 15.8) |
Categorical Variables | No. (%) |
School status | |
Closed | 18 672 (50.7) |
Hybrid/other/unknown | 8451 (23.0) |
Open | 9705 (26.4) |
Internal movement restrictions | |
No restrictions or recommended | 35 209 (93.4) |
Restrictions in place | 2507 (6.7) |
Facial coverings | |
No policy or recommended | 36 779 (97.5) |
Required in some/all places outside the home | 937 (2.5) |
Urbanization level | |
Large central metro | 816 (2.2) |
Large fringe metro | 4416 (11.7) |
Medium metro | 4476 (11.9) |
Small metro | 4296 (11.4) |
Micropolitan | 7692 (20.4) |
Noncore | 16 020 (42.5) |
Abbreviations: COVID-19, coronavirus disease 2019; TRX, number of prescriptions dispensed.
Descriptive Statistics Using County-Level Number of Prescriptions Dispensed, United States, January–December 2020
Continuous Variables . | Mean (95% CI) . |
---|---|
Monthly COVID-19 cases per 100 000 population | 558.4 (549.4 to 567.4) |
Monthly COVID-19 tests per 100 000 state residents | 22 036.1 (21 771.3 to 22 301.0) |
2020 monthly TRX per 100 000 population | 4266.9 (4235.4 to 4298.4) |
2017–2019 monthly TRX per 100 000 population | 5452.4 (5415.2 to 5489.6) |
2020 monthly TRX per 100 000 children 0–9 | 3564.0 (3523.4 to 3604.5) |
2017–2019 monthly TRX per 100 000 children 0–9 | 6572.3 (6514.8 to 6629.8) |
Physicians’ offices per 100 000 population | 35.1 (34.7 to 35.5) |
Percentage of population in poverty | 14.4 (14.4 to 14.5) |
Percentage of population people of color | 15.6 (15.5 to 15.8) |
Categorical Variables | No. (%) |
School status | |
Closed | 18 672 (50.7) |
Hybrid/other/unknown | 8451 (23.0) |
Open | 9705 (26.4) |
Internal movement restrictions | |
No restrictions or recommended | 35 209 (93.4) |
Restrictions in place | 2507 (6.7) |
Facial coverings | |
No policy or recommended | 36 779 (97.5) |
Required in some/all places outside the home | 937 (2.5) |
Urbanization level | |
Large central metro | 816 (2.2) |
Large fringe metro | 4416 (11.7) |
Medium metro | 4476 (11.9) |
Small metro | 4296 (11.4) |
Micropolitan | 7692 (20.4) |
Noncore | 16 020 (42.5) |
Continuous Variables . | Mean (95% CI) . |
---|---|
Monthly COVID-19 cases per 100 000 population | 558.4 (549.4 to 567.4) |
Monthly COVID-19 tests per 100 000 state residents | 22 036.1 (21 771.3 to 22 301.0) |
2020 monthly TRX per 100 000 population | 4266.9 (4235.4 to 4298.4) |
2017–2019 monthly TRX per 100 000 population | 5452.4 (5415.2 to 5489.6) |
2020 monthly TRX per 100 000 children 0–9 | 3564.0 (3523.4 to 3604.5) |
2017–2019 monthly TRX per 100 000 children 0–9 | 6572.3 (6514.8 to 6629.8) |
Physicians’ offices per 100 000 population | 35.1 (34.7 to 35.5) |
Percentage of population in poverty | 14.4 (14.4 to 14.5) |
Percentage of population people of color | 15.6 (15.5 to 15.8) |
Categorical Variables | No. (%) |
School status | |
Closed | 18 672 (50.7) |
Hybrid/other/unknown | 8451 (23.0) |
Open | 9705 (26.4) |
Internal movement restrictions | |
No restrictions or recommended | 35 209 (93.4) |
Restrictions in place | 2507 (6.7) |
Facial coverings | |
No policy or recommended | 36 779 (97.5) |
Required in some/all places outside the home | 937 (2.5) |
Urbanization level | |
Large central metro | 816 (2.2) |
Large fringe metro | 4416 (11.7) |
Medium metro | 4476 (11.9) |
Small metro | 4296 (11.4) |
Micropolitan | 7692 (20.4) |
Noncore | 16 020 (42.5) |
Abbreviations: COVID-19, coronavirus disease 2019; TRX, number of prescriptions dispensed.
Panel Regression
A total of 28 137 observations were included in the analysis of the association between monthly COVID-19 cases and antibiotic prescriptions dispensed to all ages, representing 3012 of 3143 counties in the United States. One hundred thirty-one counties (4.17%) were dropped due to missingness in either prescription or case data. The month of January was excluded as COVID-19 cases were not reported in any county until February.
We found a positive and significant association between monthly COVID-19 cases and monthly antibiotic prescriptions after controlling for other factors (Table 2; Supplementary Table 6). For each 1% increase in monthly COVID-19 cases per 100 000, there was an associated 0.009% increase in prescriptions per 100 000 (95% CI, 0.007% to 0.012%; P < .01). These results controlled for prior years’ monthly prescriptions, which were strongly associated with 2020 monthly prescriptions (0.647%; 95% CI, 0.633% to 0.660%; P < .01) as well as physicians’ offices per 100 000 (0.095%; 95% CI, 0.087% to 0.103%; P < .01). COVID-19 tests per 100 000 state residents was significant and negatively associated with prescribing (−0.011%; 95% CI, −0.022% to −0.001%; P < .05).
Effect of COVID-19 Cases on County-Level Number of Prescriptions Dispensed, United States, February–December 2020
. | Log of Monthly TRX per 100 000 Total Population . | Log of Monthly TRX per 100 000 Children 0–9 Years Old . |
---|---|---|
Coefficient (95% CI) . | Coefficient (95% CI) . | |
Log of monthly COVID-19 cases per 100 000 population | 0.009*** (0.007 to 0.012) | −0.012*** (−0.017 to −0.008) |
Log of monthly COVID-19 tests per 100 000 population | −0.011** (−0.022 to −0.001) | −0.039*** (−0.058 to −0.019) |
Log of 2017–2019 monthly TRX per 100 000 population | 0.647*** (0.633 to 0.660) | 0.556*** (0.541 to 0.570) |
Log of physician offices per 100 000 population | 0.095*** (0.087 to 0.103) | 0.102*** (0.092 to 0.113) |
Percentage of population in poverty | 0.008*** (0.006 to 0.011) | 0.011*** (0.007 to 0.015) |
Percentage of population of people of color | −0.002*** (−0.003 to −0.001) | −0.003*** (−0.005 to −0.002) |
School status (reference = closed) | ||
Hybrid/other/unknown | 0.006 (−0.003 to 0.014) | 0.030*** (0.013 to 0.046) |
Open | 0.004 (−0.007 to 0.015) | 0.044*** (0.024 to 0.065) |
Internal movement restrictions (reference = no restrictions/recommended) | ||
Restrictions in place | −0.003 (−0.012 to 0.006) | −0.030*** (−0.047 to −0.013) |
Facial coverings (reference = no policy/recommended/required in some places) | ||
Required in all places outside the home | −0.007 (−0.020 to 0.007) | −0.029** (−0.054 to −0.004) |
Urbanization level (reference = large central metro) | ||
Large fringe metro | −0.126*** (−0.208 to −0.044) | −0.163*** (−0.264 to −0.062) |
Medium metro | −0.101** (−0.183 to −0.018) | −0.093* (−0.195 to 0.009) |
Small metro | −0.056 (−0.139 to 0.028) | −0.072 (−0.175 to 0.032) |
Micropolitan | −0.068 (−0.150 to 0.014) | −0.026 (−0.127 to 0.075) |
Noncore | −0.088** (−0.170 to −0.005) | −0.078 (−0.179 to 0.024) |
Month (reference = February) | ||
March | 0.031 (−0.130 to 0.192) | 0.140 (−0.162 to 0.441) |
April | −0.411*** (−0.580 to −0.242) | −0.814*** (−1.131 to −0.497) |
May | −0.471*** (−0.646 to −0.297) | −0.890*** (−1.217 to −0.562) |
June | −0.251*** (−0.429 to −0.072) | −0.507*** (−0.841 to −0.172) |
July | −0.117 (−0.299 to 0.064) | −0.204 (−0.544 to 0.136) |
August | −0.201** (−0.384 to −0.018) | −0.298* (−0.641 to 0.046) |
September | −0.187** (−0.372 to −0.003) | −0.275 (−0.622 to 0.072) |
October | −0.202** (−0.388 to −0.016) | −0.326* (−0.676 to 0.025) |
November | −0.281*** (−0.469 to −0.093) | −0.469*** (−0.822 to −0.116) |
December | −0.281*** (−0.471 to −0.092) | −0.593*** (−0.949 to −0.237) |
Constant | 2.810*** (2.590 to 3.029) | 3.654*** (3.319 to 3.989) |
. | Log of Monthly TRX per 100 000 Total Population . | Log of Monthly TRX per 100 000 Children 0–9 Years Old . |
---|---|---|
Coefficient (95% CI) . | Coefficient (95% CI) . | |
Log of monthly COVID-19 cases per 100 000 population | 0.009*** (0.007 to 0.012) | −0.012*** (−0.017 to −0.008) |
Log of monthly COVID-19 tests per 100 000 population | −0.011** (−0.022 to −0.001) | −0.039*** (−0.058 to −0.019) |
Log of 2017–2019 monthly TRX per 100 000 population | 0.647*** (0.633 to 0.660) | 0.556*** (0.541 to 0.570) |
Log of physician offices per 100 000 population | 0.095*** (0.087 to 0.103) | 0.102*** (0.092 to 0.113) |
Percentage of population in poverty | 0.008*** (0.006 to 0.011) | 0.011*** (0.007 to 0.015) |
Percentage of population of people of color | −0.002*** (−0.003 to −0.001) | −0.003*** (−0.005 to −0.002) |
School status (reference = closed) | ||
Hybrid/other/unknown | 0.006 (−0.003 to 0.014) | 0.030*** (0.013 to 0.046) |
Open | 0.004 (−0.007 to 0.015) | 0.044*** (0.024 to 0.065) |
Internal movement restrictions (reference = no restrictions/recommended) | ||
Restrictions in place | −0.003 (−0.012 to 0.006) | −0.030*** (−0.047 to −0.013) |
Facial coverings (reference = no policy/recommended/required in some places) | ||
Required in all places outside the home | −0.007 (−0.020 to 0.007) | −0.029** (−0.054 to −0.004) |
Urbanization level (reference = large central metro) | ||
Large fringe metro | −0.126*** (−0.208 to −0.044) | −0.163*** (−0.264 to −0.062) |
Medium metro | −0.101** (−0.183 to −0.018) | −0.093* (−0.195 to 0.009) |
Small metro | −0.056 (−0.139 to 0.028) | −0.072 (−0.175 to 0.032) |
Micropolitan | −0.068 (−0.150 to 0.014) | −0.026 (−0.127 to 0.075) |
Noncore | −0.088** (−0.170 to −0.005) | −0.078 (−0.179 to 0.024) |
Month (reference = February) | ||
March | 0.031 (−0.130 to 0.192) | 0.140 (−0.162 to 0.441) |
April | −0.411*** (−0.580 to −0.242) | −0.814*** (−1.131 to −0.497) |
May | −0.471*** (−0.646 to −0.297) | −0.890*** (−1.217 to −0.562) |
June | −0.251*** (−0.429 to −0.072) | −0.507*** (−0.841 to −0.172) |
July | −0.117 (−0.299 to 0.064) | −0.204 (−0.544 to 0.136) |
August | −0.201** (−0.384 to −0.018) | −0.298* (−0.641 to 0.046) |
September | −0.187** (−0.372 to −0.003) | −0.275 (−0.622 to 0.072) |
October | −0.202** (−0.388 to −0.016) | −0.326* (−0.676 to 0.025) |
November | −0.281*** (−0.469 to −0.093) | −0.469*** (−0.822 to −0.116) |
December | −0.281*** (−0.471 to −0.092) | −0.593*** (−0.949 to −0.237) |
Constant | 2.810*** (2.590 to 3.029) | 3.654*** (3.319 to 3.989) |
Abbreviations: COVID-19, coronavirus disease 2019; TRX, number of prescriptions dispensed.
***P < .01; **P < .05; *P < 0.1.
Effect of COVID-19 Cases on County-Level Number of Prescriptions Dispensed, United States, February–December 2020
. | Log of Monthly TRX per 100 000 Total Population . | Log of Monthly TRX per 100 000 Children 0–9 Years Old . |
---|---|---|
Coefficient (95% CI) . | Coefficient (95% CI) . | |
Log of monthly COVID-19 cases per 100 000 population | 0.009*** (0.007 to 0.012) | −0.012*** (−0.017 to −0.008) |
Log of monthly COVID-19 tests per 100 000 population | −0.011** (−0.022 to −0.001) | −0.039*** (−0.058 to −0.019) |
Log of 2017–2019 monthly TRX per 100 000 population | 0.647*** (0.633 to 0.660) | 0.556*** (0.541 to 0.570) |
Log of physician offices per 100 000 population | 0.095*** (0.087 to 0.103) | 0.102*** (0.092 to 0.113) |
Percentage of population in poverty | 0.008*** (0.006 to 0.011) | 0.011*** (0.007 to 0.015) |
Percentage of population of people of color | −0.002*** (−0.003 to −0.001) | −0.003*** (−0.005 to −0.002) |
School status (reference = closed) | ||
Hybrid/other/unknown | 0.006 (−0.003 to 0.014) | 0.030*** (0.013 to 0.046) |
Open | 0.004 (−0.007 to 0.015) | 0.044*** (0.024 to 0.065) |
Internal movement restrictions (reference = no restrictions/recommended) | ||
Restrictions in place | −0.003 (−0.012 to 0.006) | −0.030*** (−0.047 to −0.013) |
Facial coverings (reference = no policy/recommended/required in some places) | ||
Required in all places outside the home | −0.007 (−0.020 to 0.007) | −0.029** (−0.054 to −0.004) |
Urbanization level (reference = large central metro) | ||
Large fringe metro | −0.126*** (−0.208 to −0.044) | −0.163*** (−0.264 to −0.062) |
Medium metro | −0.101** (−0.183 to −0.018) | −0.093* (−0.195 to 0.009) |
Small metro | −0.056 (−0.139 to 0.028) | −0.072 (−0.175 to 0.032) |
Micropolitan | −0.068 (−0.150 to 0.014) | −0.026 (−0.127 to 0.075) |
Noncore | −0.088** (−0.170 to −0.005) | −0.078 (−0.179 to 0.024) |
Month (reference = February) | ||
March | 0.031 (−0.130 to 0.192) | 0.140 (−0.162 to 0.441) |
April | −0.411*** (−0.580 to −0.242) | −0.814*** (−1.131 to −0.497) |
May | −0.471*** (−0.646 to −0.297) | −0.890*** (−1.217 to −0.562) |
June | −0.251*** (−0.429 to −0.072) | −0.507*** (−0.841 to −0.172) |
July | −0.117 (−0.299 to 0.064) | −0.204 (−0.544 to 0.136) |
August | −0.201** (−0.384 to −0.018) | −0.298* (−0.641 to 0.046) |
September | −0.187** (−0.372 to −0.003) | −0.275 (−0.622 to 0.072) |
October | −0.202** (−0.388 to −0.016) | −0.326* (−0.676 to 0.025) |
November | −0.281*** (−0.469 to −0.093) | −0.469*** (−0.822 to −0.116) |
December | −0.281*** (−0.471 to −0.092) | −0.593*** (−0.949 to −0.237) |
Constant | 2.810*** (2.590 to 3.029) | 3.654*** (3.319 to 3.989) |
. | Log of Monthly TRX per 100 000 Total Population . | Log of Monthly TRX per 100 000 Children 0–9 Years Old . |
---|---|---|
Coefficient (95% CI) . | Coefficient (95% CI) . | |
Log of monthly COVID-19 cases per 100 000 population | 0.009*** (0.007 to 0.012) | −0.012*** (−0.017 to −0.008) |
Log of monthly COVID-19 tests per 100 000 population | −0.011** (−0.022 to −0.001) | −0.039*** (−0.058 to −0.019) |
Log of 2017–2019 monthly TRX per 100 000 population | 0.647*** (0.633 to 0.660) | 0.556*** (0.541 to 0.570) |
Log of physician offices per 100 000 population | 0.095*** (0.087 to 0.103) | 0.102*** (0.092 to 0.113) |
Percentage of population in poverty | 0.008*** (0.006 to 0.011) | 0.011*** (0.007 to 0.015) |
Percentage of population of people of color | −0.002*** (−0.003 to −0.001) | −0.003*** (−0.005 to −0.002) |
School status (reference = closed) | ||
Hybrid/other/unknown | 0.006 (−0.003 to 0.014) | 0.030*** (0.013 to 0.046) |
Open | 0.004 (−0.007 to 0.015) | 0.044*** (0.024 to 0.065) |
Internal movement restrictions (reference = no restrictions/recommended) | ||
Restrictions in place | −0.003 (−0.012 to 0.006) | −0.030*** (−0.047 to −0.013) |
Facial coverings (reference = no policy/recommended/required in some places) | ||
Required in all places outside the home | −0.007 (−0.020 to 0.007) | −0.029** (−0.054 to −0.004) |
Urbanization level (reference = large central metro) | ||
Large fringe metro | −0.126*** (−0.208 to −0.044) | −0.163*** (−0.264 to −0.062) |
Medium metro | −0.101** (−0.183 to −0.018) | −0.093* (−0.195 to 0.009) |
Small metro | −0.056 (−0.139 to 0.028) | −0.072 (−0.175 to 0.032) |
Micropolitan | −0.068 (−0.150 to 0.014) | −0.026 (−0.127 to 0.075) |
Noncore | −0.088** (−0.170 to −0.005) | −0.078 (−0.179 to 0.024) |
Month (reference = February) | ||
March | 0.031 (−0.130 to 0.192) | 0.140 (−0.162 to 0.441) |
April | −0.411*** (−0.580 to −0.242) | −0.814*** (−1.131 to −0.497) |
May | −0.471*** (−0.646 to −0.297) | −0.890*** (−1.217 to −0.562) |
June | −0.251*** (−0.429 to −0.072) | −0.507*** (−0.841 to −0.172) |
July | −0.117 (−0.299 to 0.064) | −0.204 (−0.544 to 0.136) |
August | −0.201** (−0.384 to −0.018) | −0.298* (−0.641 to 0.046) |
September | −0.187** (−0.372 to −0.003) | −0.275 (−0.622 to 0.072) |
October | −0.202** (−0.388 to −0.016) | −0.326* (−0.676 to 0.025) |
November | −0.281*** (−0.469 to −0.093) | −0.469*** (−0.822 to −0.116) |
December | −0.281*** (−0.471 to −0.092) | −0.593*** (−0.949 to −0.237) |
Constant | 2.810*** (2.590 to 3.029) | 3.654*** (3.319 to 3.989) |
Abbreviations: COVID-19, coronavirus disease 2019; TRX, number of prescriptions dispensed.
***P < .01; **P < .05; *P < 0.1.
For each 1% increase in the proportion of the population living in poverty, there was an associated 0.008% increase in monthly prescriptions per 100 000 (95% CI, 0.006% to 0.011%; P < .01). However, the percentage of the population of people of color was negatively associated with antibiotic prescribing (−0.002%; 95% CI, −0.003% to −0.001%; P < .05). Large central metro counties had the strongest association with prescribing, while large fringe metro and noncore counties had the weakest association in comparison. Closing schools, internal movement restrictions, and requiring facemasks did not have a significant relationship with prescribing. There was a strong linear relationship between the number of prescriptions and DDDs (Supplementary Figure 1), and results were similar in the sensitivity analysis using DDDs (Supplementary Table 7).
In the subanalysis with children, a 1% increase in monthly COVID-19 cases was associated with a 0.012% decrease (95% CI, −0.017% to −0.008%; P < .01) in the number of monthly prescriptions (Table 2; Supplementary Table 6). Opening schools at the county level was positively correlated with prescriptions among children (0.044%; 95% CI, 0.024% to 0.065%; P < .01). Additionally, movement restrictions (−0.030%; 95% CI, −0.047% to −0.013%; P < .01) and requiring facemasks (−0.029%; 95% CI, −0.054% to −0.004%; P < .05) were associated with lower prescribing among children. In the sensitivity analysis using state-level school data, the effect of opening schools was stronger (0.092%; 95% CI, 0.056% to 0.129%; P < .01) and significant compared with closing schools in the regression using data among all ages (0.026%; 95% CI, 0.007% to 0.046%; P < .05) (Supplementary Table 8).
DISCUSSION
The COVID-19 pandemic altered patterns of care-seeking behavior as well as rates of other viral respiratory pathogens [1–3, 34]. Implementation and adherence to NPIs depended on the timing of surges and geographically specific attitudes toward disease prevention measures. These circumstances provided a natural experiment to evaluate the impact of COVID-19 cases as well as the importance of schools, NPIs, and sociodemographic factors in driving antibiotic prescribing in the outpatient setting, which makes up a significant proportion of overall antibiotic use [8]. A reduction in total antibiotic prescribing was observed from March through December 2020 compared with these months in prior years. Outpatient prescribing dispensed to all ages was positively correlated with monthly county-level COVID-19 cases; however, there was an inverse correlation between monthly cases and prescriptions dispensed to children. While the percentage increase per 100 000 population was not large at the county level, the total increase in the number of prescriptions attributable to COVID-19 cases was ∼1000 prescriptions for every 1% increase in COVID-19 cases.
These results suggest that surges in cases were primarily associated with increased prescribing among adults. IQVIA data do not include diagnostic information and do not allow for an understanding of prescribing appropriateness. As bacterial coinfection in COVID-19 patients is rare [4] and other studies have identified antibiotic overuse [6, 7], it is possible that a fraction of outpatient antibiotic prescribing in 2020 was inappropriate. While the evidence on telehealth and antibiotic stewardship is mixed, virtual visits during the pandemic may have also driven prescriptions if more physicians were prescribing antibiotics without diagnostic testing or in-person evaluation [35]. Alternatively, increases in prescribing could have been influenced by increases in health care–seeking behavior and health care access. This may be especially true in large urban areas, which associations between prescribing and urbanization level may suggest. The largest drop in dispensed antibiotics was observed among children, and COVID-19 cases were negatively associated with prescribing among children. These results suggest that closing schools for in-person instruction could have decreased transmission of other URIs among children and opening schools could have had the opposite effect. This is consistent with evidence of increased disease transmission of pathogens in school environments [36, 37].
Our analyses support prior findings of decreased outpatient antibiotic prescribing due to reduced health care–seeking behavior at the beginning of the pandemic [1–3, 38]. While the first quarter of 2020 followed expected trends in the monthly number of common antibiotic prescription fills, the number of patients dispensed antibiotics exceeded seasonally expected decreases by 33 percentage points from January to May 2020 [1]. Over the rest of the year, ambulatory prescribing rates for URIs may have declined by as much as 79% compared with the same period in prior years [3].
While we included structural factors, such as poverty rate, health care access, density (ie, urbanization level), and seasonal factors, the largest and most significant covariate was the county-level association with prior years’ prescribing rates. Areas with relatively higher prescribing rates continued to remain high and vice versa. Thus, while the pandemic altered many patterns of health care–seeking behavior, these results suggest that it did not alter the underlying factors that drive variation in prescribing. These factors appear to cluster at the state level, with prescribing generally highest in the Southeast region of the United States and lowest in the Western region (Supplementary Table 4, Supplementary Figure 2). The strong correlation with prior prescribing suggests that social norms could be the most important factors driving prescribing. Social norms describe the unwritten rules that govern interactions between individuals and can govern both health care–seeking behavior (eg, when and under what conditions patients seek care) and physician prescribing patterns. For example, there are differences in when individuals seek care that are not related to clinical factors [39]. Similarly, prescribing patterns are driven by comparative normative behavior (eg, how others in the practice setting are prescribing) [40]. Prescribing norms are also driven and enforced by patient expectations and actions. In the United States, patients often expect an antibiotic if they believe their symptoms warrant a visit to the doctor and may consult another physician if their expectations are not met. In other countries, such as the Scandinavian countries, patient expectations differ, and consequently the prescribing norms differ [41]. To date, there is no consensus on how to alter “prescribing norms” to reduce prescribing [42], but it is crucial for developing strategies and policies to reduce antibiotic prescribing in the long run and should be a focus of qualitative and quantitative research to improve our understanding.
Finally, there are important aspects of these data and results that should guide future policies for seasonal URI epidemics. It has been difficult to fully assess the effect of NPIs because while policies are implemented at the state or county level, they require individual adherence to be effective, and adherence likely changed when cases increased. While school closures, mask-wearing policies, and movement restrictions were not associated with a significant reduction in prescriptions among all ages in our analysis, they were significantly associated with decreases in prescribing among children. Given the relatively higher rate of prescribing in children for URIs, these results suggest that NPIs played a role in limiting the spread of diseases, including COVID-19, in the school-age population. Limited and targeted use of facemasks in particular settings (eg, schools, hospitals, and long-term care facilities) may help reduce the spread of all URIs and concomitantly antibiotic use, both appropriate and inappropriate. To achieve effectiveness, however, improved markers of rising infection are needed at a localized level, for example, wastewater surveillance for multiple pathogens at the individual school or health care facility level.
Limitations
We rigorously controlled for model specification and adjusted for state, year, and seasonality with carefully selected control variables; however, our analysis is subject to some limitations. First, results are correlative and not causative due to the ecological study design. Second, data only available at the state level (testing and NPIs) do not account for heterogeneity between and within counties. Furthermore, NPI data only represented reported policies, not actual behavior or individual compliance. Available data on school mandates did not allow for a detailed understanding of the heterogeneity of instruction within counties or states or among different age groups. Third, potential changes in county urban-rural classification since 2013 (the most recent classification scheme) or the small percentage of counties dropped from the analysis due to missingness could have biased results. Dropping January from the analysis likely did not bias results because there could not be a relationship between cases and prescriptions before the pandemic began. Fourth, use data came from the only source of large-scale antibiotic use in the United States, and data provided by IQVIA are an estimation of total use. The zip code reported by IQVIA represents where the prescription was filled, not where the patient lived. As mentioned, IQVIA data do not provide diagnostic decision information from the prescribing physician or the results of diagnostic tests, so we excluded antibiotics that are not typically used to treat respiratory infections in the outpatient setting. This exclusion reduced total antibiotic prescribing by <1%—and so did not qualitatively affect the results.
CONCLUSIONS
The COVID-19 pandemic dramatically changed antibiotic prescribing patterns presumably through changes in health care–seeking behavior and changes in transmission of non-COVID respiratory viruses. The results point to 2 interconnected issues for further analysis. First, providers may have inappropriately prescribed antibiotics for COVID-19, although more information on the prescribing physician's diagnosis or results of ordered tests is needed to conclude so definitively. While inappropriate treatment of viral infections is common, the assumption has been that reducing diagnostic uncertainty as to the etiology of infection would reduce antibiotic use. The evidence here suggests that antibiotic use may have continued even when infections were likely viral. Further research is needed to understand why clinicians continue to prescribe antibiotics in these situations and how to modify this behavior. Second, while inappropriate prescribing remains a common problem in the outpatient setting [8, 43], our results bolster similar studies, suggesting that there remain unknown factors—“prescribing norms”—driving variation in use. Norms may include differing awareness levels about the threat of antimicrobial resistance or differing views on its severity. Given that these norms appear to explain the largest difference in prescribing in most ecological research, studies are needed to describe the social norms that drive health care use and prescribing at the individual level and how interventions to alter prescribing norms can scale across geographic areas.
Supplementary Data
Supplementary materials are available at Open Forum Infectious Diseases online. Consisting of data provided by the authors to benefit the reader, the posted materials are not copyedited and are the sole responsibility of the authors, so questions or comments should be addressed to the corresponding author.
Acknowledgments
Financial support. This study was supported in part by the CDC MInD Healthcare Network (U01CK000589) and National Science Foundation Expeditions (grant CCF1917819).
Author contributions. A.H. processed the data, ran the analysis, and prepared the manuscript. S.P. generated visualizations. J.C., S.C., and R.L. provided subject matter expertise. E.K. conceived of the study and methods and provided subject matter and technical expertise.
Patient consent. This study did not require patient consent. As all data obtained were aggregated, de-identified, publicly available data, the analysis did not constitute human subjects research as defined at 45 CFR 46.102 and did not require IRB review.
References
Author notes
Potential conflicts of interest. All authors: no reported conflicts.
Comments