Leveraging the COVID-19 pandemic as a natural experiment to assess changes in antibiotic use and antibiotic-resistant E. coli carriage in semi-rural Ecuador

The coronavirus 2019 (COVID-19) pandemic has had significant impacts on health systems, population dynamics, public health awareness, and antibiotic stewardship, which could affect antibiotic resistant bacteria (ARB) emergence and transmission. In this study, we aimed to compare knowledge, attitudes, and practices (KAP) of antibiotic use and ARB carriage in Ecuadorian communities before versus after the COVID-19 pandemic began. We leveraged data collected for a repeated measures observational study of third-generation cephalosporin-resistant E. coli (3GCR-EC) carriage among children in semi-rural communities in Quito, Ecuador between July 2018 and September 2021. We included 241 households that participated in surveys and child stool sample collection in 2019, before the pandemic, and in 2021, after the pandemic began. We estimated adjusted Prevalence Ratios (aPR) and 95% Confidence Intervals (CI) using logistic and Poisson regression models. Child antibiotic use in the last 3 months declined from 17% pre-pandemic to 5% in 2021 (aPR: 0.30; 95% CI 0.15, 0.61) and 3GCR-EC carriage among children declined from 40 to 23% (aPR: 0.48; 95% CI 0.32, 0.73). Multi-drug resistance declined from 86 to 70% (aPR: 0.32; 95% CI 0.13; 0.79), the average number of antibiotic resistance genes (ARGs) per 3GCR-EC isolate declined from 9.9 to 7.8 (aPR of 0.79; 95% CI 0.65, 0.96), and the diversity of ARGs was lower in 2021. In the context of Ecuador, where COVID-19 prevention and control measures were strictly enforced after its major cities experienced some of the world’s the highest mortality rates from SARS-CoV-2 infections, antibiotic use and ARB carriage declined in semi-rural communities of Quito from 2019 to 2021.

of this study was to compare caregiver knowledge, attitudes, and practices (KAP) around antibiotic use and antibiotic-resistant E. coli carriage in healthy children before versus after the onset of the COVID-19 pandemic.

Methods
We analyzed a subset of data collected for a repeated measures observational study conducted in semi-rural communities east of Quito, Ecuador between July 2018 and September 2021.The parent study enrolled 605 households with at least one child between 6 months and 5 years old at the time of enrollment throughout the study period.A single child was enrolled per household.Child fecal samples and survey data on child and household characteristics, including caregiver-reported child illness, antibiotic use, and KAP around antibiotics, were collected at up to five time points per household throughout the study period (Supplemental Materials, Fig. S1).After the third data collection cycle in July through November of 2019, the fourth of five data collection cycles began in January 2020 and ended prematurely in March 2020 due to initial lockdowns at the beginning of the COVID-19 pandemic.Data collection resumed for the fifth cycle in April 2021 through September 2021.
Details of the parent study design and methodology are described elsewhere (Amato et al. 2023 38 , Accepted, PLOS Medicine).
Patient and public involvement.Antibiotic resistance is an issue that our communities were unfamiliar with prior to beginning the study, especially because it can spread asymptomatically.Thus, we worked with a community organizer to identify community needs we could fulfill while simultaneously achieving research priorities of microbiology and public health researchers at the Universidad San Francisco de Quito and University of California, Berkeley.For example, we provided individual results on child intestinal parasites to caregivers, which were of interest to the community.We also modified the study incentive based on community feedback; we planned to distribute hygiene kits but instead provided basic items like rice and beans, as requested by community members.
For the design of the study, we involved a community organizer who lives in the community and meets regularly with community residents.Based on feedback from residents communicated to us via the community organizer, we dropped some survey questions that were contextually irrelevant or inappropriate.We also improved our process for delivering study results back to households by working with a community organizer; we reduced turnaround time for communicating child intestinal parasite data to caregivers, along with recommendations for seeking treatment.Community residents were also included as part of the study team.These individuals have long standing relationships with the community and have worked on several public health campaigns by the government and by local NGOs.We relied heavily on these members of our research team who live in the community to guide us and change protocols and adjust to timelines that were acceptable to the community.
We have hosted two community-wide events (all study participants invited) to share and discuss results with the community.We are also working to design health promotion materials that are collaboratively designed with feedback from the community.The community events were a great success and helped build a stronger understanding of zoonotic infectious diseases and antibiotic resistance (see short video of one event here: https:// vimeo.com/ 65431 5270).

Study site.
The study area is approximately 320 km 2 spanning seven semi-rural parishes east of Quito, Ecuador.Three parishes have high-intensity commercial food-animal production (> 30 operations) and approximately 30% of households in the parent study owned backyard food-animals, known risk factors for antibioticresistant infections (Amato et al.In Review).Household-level access to water, sanitation, and hygiene infrastructure is very high in this study area.In the parent study, almost all households had handwashing stations with both soap and water, flush toilets, and piped drinking water inside the home.There are no wastewater treatment plants in the study area; wastewater is discharged directly into surface water.

Microbiological methods.
Child fecal samples were plated on MacConkey agar (Difco, Sparks, Maryland) with 2 mg/L of ceftriaxone and incubated for 18 h at 37 °C to screen for 3GCR-EC (Salinas et al., 2019).One to five isolates phenotypically matching E. coli were selected from each fecal sample and preserved at − 80 °C in Trypticase Soy Broth medium (Difco, Sparks, MD) with 20% glycerol.Isolates were thawed and regrown on MacConkey agar at 37 °C for 24 h for evaluation of antibiotic susceptibility by the disk diffusion method (Kirby Bauer test) on Mueller-Hinton agar (Difco, Sparks, Maryland).Colonies were inoculated onto Chromocult® coliform agar to confirm presumptive E. coli growth (Merck, Darmstadt, Germany).
Statistical analyses.For this before and after analysis, we compared caregiver KAP, child health and antibiotic use, and 3GCR-EC carriage in children from data collected in the third cycle in 2019 (before COVID-19 began) to data collected in the fifth cycle in 2021 (after COVID-19 began) (Fig. S1).We used a within subjects design, including only data from households that participated in data collection in both 2019 and 2021, so each household served as its own control in the comparison.Sensitivity analyses.To assess whether the observed changes in our outcomes of interest were possibly due to other secular trends, we conducted sensitivity analyses using the differences-in-differences approach, a quasi-experimental, controlled before-and-after design 50 .We compared the changes from data collection cycle one in 2018 to cycle three in 2019 (comparison 1: secular trends pre-COVID-19) with the changes from cycle three in 2019 to cycle five in 2021 (comparison 2: pre versus post pandemic onset) (Fig. S1).We used multivariable logistic and Poisson regression models with pre-specified covariates (child age, child sex, caregiver education, household wealth, and parish), plus an interaction term for the time and comparison (i.e., treatment) group.For this sensitivity analysis, we included all child fecal samples and multiple isolates per fecal sample when applicable, regardless of their participation in each data collection cycle.

Ethics approval and consent to participate.
The study was approved by the Office for Protection of Human Subjects (OPHS) at the University of California, Berkeley (IRB# 2019-02-11803) and by the Bioethics Committee at the Universidad San Francisco de Quito (#2017-178 M), and the Ecuadorian Health Ministry (#MSPCURI000243-3).All study participants provided informed consent prior to participation.Caregivers provided written assent prior to their child's participation in the study.

Results
For the main analysis, we included 241 children from households that participated in both the third (pre-pandemic, 2019) and fifth (post-pandemic onset, 2021) data collection cycles.Forty-two percent (42%) of children were female.In 2019, 71% of children were at least two years of age and the mean age was 3.4 (Standard Deviation (SD): 1.6).The mean age of caregivers who responded to household surveys was 32.8 (SD: 12.0), and 73% of caregivers had at least a high school or college education.Of note, 74% of children had contact with pets in the last three months in both 2019 and 2021 (Table S3).Twelve percent (12%) of children attended daycare prior to the pandemic, while only 4% attended daycare in 2021 after the pandemic began (Table S3).
Caregiver knowledge, attitudes, and practices.Caregiver knowledge that antibiotics do not kill viruses increased from 27% in 2019 to 38% in 2021 (Fig. 1, Table S1).Fewer caregivers responded "yes" or "don't know" in 2021 when asked if antibiotics kill viruses (adjusted Prevalence Ratio (aPR): 0.54; 95% Confidence Interval (CI) 0.35, 0.86) (Table 1).In 2021, fewer caregivers responded "don't know" and more caregivers correctly responded "yes" when asked if antibiotics kill bacteria compared to 2019, before the pandemic (Fig. 1).However, the change in the proportion of incorrect responses ("no" or "don't know") did not change significantly (aPR: 0.73; 95% CI 0.48, 1.10) (Table 1).In sensitivity analyses of a total of 590 households, the changes in knowledge about antibiotic use observed before COVID-19 (from 2018 to 2019) were not significantly different from the changes observed after vs. before COVID-19 began (Table S6).However, the proportion of caregivers who correctly stated that antibiotics do not kill viruses rose from 25 to 28% in 2018 and 2019, then jumped to 36% in 2021 (Table S5).Caregiver attitudes and practices around antibiotic use changed, becoming more nuanced after the COVID-19 pandemic began.The proportion of caregivers responding "don't know" increased after COVID-19 across all questions about attitudes, practices, and social norms around antibiotic use (Fig. 1).When asked about situations in which the caregiver or their child should take antibiotics, whether they can expect to purchase antibiotics at the pharmacy or receive a prescription from their doctor, and whether taking antibiotics will make them feel better, caregivers responded "always" or "never" less frequently in 2021 compared to in 2019 (Fig. 1).For example, 44% of caregivers believed that giving their child antibiotics for a cold would "always" make them better before the pandemic; this dropped to 27% after the pandemic (aPR: 0.42; 95% CI 0.27, 0.65) (Table 1).For this same question, 29% believed antibiotics would "sometimes" make their child feel better before the pandemic, which increased to 51% after the pandemic (Fig. 1).Changes in attitudes and practices around antibiotic use were only observed in before and after COVID-19 comparisons, while they remained largely unchanged from 2018 to 2019 before COVID-19 began based on sensitivity analyses (Table S5, Table S6).Similar patterns were observed for caregiver-perceived social norms around antibiotic use.Before COVID-19, 29% of caregivers believed that most of their friends think they should "always" give their child antibiotics for a cold, while 16% believed their friends think they should "never" give their child antibiotics for a cold (Fig. 1).These responses declined to 7% and 10%, respectively, after the pandemic (aPR: 0.51; 95% CI 0.34, 0.77), while the proportion of caregivers who responded "sometimes" increased from 34 to 53% (Table 1, Fig. 1).Notably, the  proportion of caregivers who believed most of their friends "always" buy their child antibiotics without a prescription decreased from 24% before COVID-19 to 4% in 2021 (Fig. 1).In sensitivity analyses, the difference between the changes in perceptions of social norms around antibiotic use from 2019 to 2021 and changes observed from 2018 to 2019 was not statistically significant (Table S6).However, the ratios of aPRs remained < 1, indicating a greater decrease in "always" and "never" responses from 2019 to 2021 compared to pre-COVID-19 trends.For example, caregivers included in the sensitivity analysis said they believed most of their friends should "always" give their child antibiotics for a cold 24% of the time in 2018, 27% of the time in 2019, and only 6% of the time in 2021 (Table S5).This trend was observed in sensitivity analyses for all variables on caregiver perceived social norms around antibiotic use.

Discussion
This study leveraged data collected before and after the COVID-19 pandemic as a natural experiment to compare antibiotic use and ARB carriage among healthy individuals in semi-rural communities of Quito, Ecuador.Overall, we observed significant declines in antibiotic use (68% reduction), community-acquired 3GCR-EC (42% reduction), MDR 3GCR-EC (18% reduction), and total ARGs (24% reduction) from 2019 to 2021 in semi-rural Quito, Ecuador.Several factors related to population mobility and infection prevention and control measures may have contributed to the observed reduction in 3GCR-EC carriage among children.Restrictions on international travel may have limited the introduction of ESBL-E and MDR bacteria from individuals coming from subtropical regions, and restrictions in movement within communities could limit opportunities for ARB transmission 51 .International travel to low-income countries in subtropical regions-especially South Asia & Sub-Saharan Africa-is a known risk factor for ESBL-E infections, especially given antibiotic use during travel [52][53][54] .International travel restrictions and local quarantines in March-April 2020, the beginning of the COVID-19 pandemic, may have contributed to similar declines in ESBL-E detected in clinical isolates from community-acquired infections in France 55 .www.nature.com/scientificreports/Daycare attendance and hospital admissions are also known risk factors for ARB carriage in children 56 .Fewer children in our study site attended a daycare in 2021 compared to 2018 and 2019, preventing exposures to unhygienic or crowded daycare environments and likely reducing ARB transmission.There were also fewer children who received medical treatment in 2021 compared to pre-pandemic years.Caregivers may have decided not to seek care or schedule elective or non-urgent medical procedures for their child to avoid exposure to COVID-19 patients in hospitals and other health care centers.Telemedicine also became widely available in Ecuador during the pandemic, reducing parents' need to visit a hospital for medical advice 57 .
Public health messaging to encourage frequent handwashing, social distancing, and use of face masks may have further reduced community transmission of infectious pathogens, including ESBL-E and other ARB 31,32 .However, child hand washing after contact with animals decreased from 2019 to 2021 in our study site.Other measures to interrupt COVID-19 transmission like travel restrictions, curfews, and school or business closures likely played a stronger role in interrupting ARB transmission in semi-rural Ecuador.Adherence to infection prevention measures may have been higher in places with strict mandates or in areas more severely impacted by the pandemic, suggesting that these areas may have seen a greater reduction in both COVID-19 and communityacquired ARB [58][59][60] .This may have been the case in our study site in Quito, Ecuador, where the government enforced strict pandemic mitigation policies.
We observed 64% and 68% reductions in the proportion of children who received medical treatment and antibiotics in the last 3 months, respectively.The reduction in antibiotic use during COVID-19 is consistent with findings on antibiotic prescribing for community-acquired infections in the United States, Europe, and Australia [61][62][63][64][65][66] .While several studies have qualitatively described antibiotic use in LMIC communities during COVID-19 [10][11][12]67 , few have quantitatively estimated the change in community antibiotic use in LMICs since the pandemic began 68 . A ingle interrupted time series analysis estimated that the private health sector in India sold 216 million excess doses of antibiotics between June and September of 2020, though antibiotic sales declined from September to December of 2020 and total sales in 2020 were lower than in 2018 or 2019 68 .Despite evidence of declines in antibiotic use in our study site, community antibiotic use may rebound over time.Clinically relevant antibiotics remain unregulated and easily accessible for self-medication, which is common in poor communities in Ecuador and elsewhere [6][7][8] . Ovruse of antibiotics among children in LMICs remains an urgent threat to children's health, and policies restricting OTC access to antibiotics without a prescription are urgently needed to curb unnecessary or inappropriate antibiotic use 9 .
There are some limitations to this study.The parent study was not designed to test the hypothesis that there was a change in antibiotic use and resistance before vs. after the onset of the COVID-19 pandemic, increasing the potential risk of bias.We attempted to reduce bias from unmeasured confounding by only including households and children that participated in both data collection cycles before and after the start of the pandemic in 2020, allowing each household to serve as its own control.Another limitation is the use of caregiver-reported outcomes like antibiotic use and diarrhea, which increases the risk of response bias.We aimed to reduce bias in the parent study by using optimal recall periods for diarrhea (7 days) and antibiotic use (3 months), which have been shown to be reliable 69,70 .Finally, we induced selection bias by selecting for 3GCR-EC; the observed changes in phenotypic resistance to individual antibiotics may not be representative of the prevalence of resistance among all enteric E. coli, let alone other Enterobacterales.Periodic surveillance or additional studies of community-acquired ARB in Ecuador and other LMICs will be necessary to more comprehensively describe the prevalence of resistance.
This analysis, which leveraged an existing dataset from a longitudinal observational study, also has several strengths.We conducted the first (to our knowledge) before and after analysis of ARB in the context of the COVID-19 pandemic in a LMIC.While the before and after analysis is clear-cut and easily interpretable, we further tested our hypotheses with a quasi-experimental sensitivity analysis via the difference in differences approach.The difference in difference method controls for longitudinal trends before COVID-19, strengthening and contextualizing the findings from the before and after analysis.Results from the main analysis were largely supported by the results of sensitivity analyses, suggesting that response bias (e.g., social-desirability bias) from repeated household visits and other secular trends did not significantly affect our findings.Finally, we included a sub-analysis of 3GCR-EC whole genomes to evaluate changes in ARGs.By adding this sequencing analysis, we were able to highlight the public health significance of 3GCR-EC carriage in this setting by identifying clinically relevant E. coli STs and ARGs encoding ESBL production.
Among children colonized with 3GCR-EC, several were infected with E. coli strains of clinical relevance.We identified several major extraintestinal pathogenic E. coli (ExPEC) STs including ST131, ST10, ST69, ST38, ST354, ST117, ST88, and ST617, which are responsible for a significant proportion of urinary tract and bloodstream infections globally 71 .We found a high prevalence of blaCTX-M and other ESBL genes among 3GCR-EC isolates, and the diversity of ESBL genes seemed to decrease after the pandemic began.CTX-M-encoding genes bla CTX-M-55 , bla CTX-M-65 , and bla CTX-M-15 were the most frequently detected bla genes in our study site in Ecuador, and are frequently found in meat products and food animals such as chickens, pigs, and cattle across Asia, North America, and Europe [72][73][74][75][76] .bla CTX-M-15 is also frequently detected in ExPEC, specifically the virulent and often MDR strain, ST131 77 .Previously published findings from the parent study suggest both clonal spread and horizontal gene transfer of 3GCR-EC between children, backyard chickens, and dogs in the study site, where animal waste management and handling practices were generally poor 43,78 .Commercial food animal production is also common in this study site, and living in high-intensity production areas was identified as a risk factor for carriage of ESBL-producing 3GCR-EC among children in this study site (Amato et al. 38 In Review).While detection of bla CTX-M-15 , bla CTX-M-55 , and bla CTX-M-8 (another ESBL gene found in broiler chickens 79 ) increased after COVID-19 began, detection of all other bla CTX-M genes decreased.The decrease in diversity of ESBL genes after the pandemic began may suggest that children were exposed to fewer sources of ESBL-producing bacteria, ESBL genes, and mobile genetic elements, especially during quarantine and lockdown periods.Still, the high www.nature.com/scientificreports/prevalence (31% overall) of ESBL-producing 3GCR-EC-including virulent and MDR strains-among children in Ecuadorian communities is concerning.Comprehensive strategies such as vaccine development and largescale human and animal waste management are necessary to tackle the spread of MDR pathogenic E. coli 80,81 .

Conclusions
The COVID-19 pandemic provided an unprecedented opportunity to assess the impacts of restricted population mobility and intensive infection control and prevention measures on community transmission of resistant bacteria.Researchers with relevant data that was collected before and during the pandemic should consider conducting similar analyses to broaden our understanding of ARB transmission dynamics in various contexts.Antibiotic use and community spread of ARB during the pandemic may vary locally based on socioeconomic status, population density, proximity to urban centers, social and cultural norms, and intensity of COVID-19 control and prevention policies.Our findings suggest that in the context of severe outbreaks and strict policies to reduce COVID-19 transmission in Ecuador, community-acquired ARB decreased from 2019 to 2021.Periodic surveillance of community-acquired ARB infections will be necessary to track long-term changes in ARB transmission as pandemic-related policies continue to evolve.Knowledge, attitudes, and practices around antibiotic use may also evolve, and further improvements are needed to curb excessive and inappropriate antibiotic use.
As antibiotic resistance remains an urgent threat to public health, governments should leverage public health communications systems bolstered during the pandemic to encourage and enforce antibiotic stewardship among community members as well as pharmacists, physicians, veterinarians, and food-animal producers.Improved hygiene practices (e.g., frequent handwashing, regular cleaning of surfaces, limited capacity/crowding) adapted to prevent the spread of COVID-19 should be maintained in public places, in schools/daycares, and in households to prevent ARB transmission.Finally, given the role of domestic and commercially produced food animals in community spread of ARB, public health policies and communications efforts must use a One Health approach to effectively address inappropriate antibiotic use and limit the selection for and exposures to resistant bacteria.

DNA sequencing and analysis. We
40tracted purified DNA from the isolates using Wizard® Genomic DNA Purification (Promega) kits and QIAGEN© DNEasy Blood & Tissue Kits according to the manufacturer's instructions.Whole-genome sequencing of E. coli isolates was carried out at the University of Minnesota using Illumina MiSeq with Nextera XT libraries.Raw reads were quality-trimmed and adapter-trimmed using trimmomatic40and assembled using SPAdes (Bankevich et al. 2012).Antibiotic resistance genes (ARGs) were detected using ABRicate (version 0.813) and a curated version of the ResFinder database (Zankari et al. 2012).

Table 1 .
Prevalence ratios of caregiver knowledge, attitudes, and practices around antibiotic use in 2021 after the COVID-19 began compared to 2019 before the pandemic.*Adjusted Prevalence Ratios (aPR) estimated with logistic regressions using Generalized Estimating Equations to adjust for repeated measures, and adjusted for confounding by including the following covariates: child age, child sex, household wealth (asset score), and caregiver education.95% CI 95% Confidence Intervals.

Table 2 .
Prevalence ratios of 3GCR-EC carriage and phenotypic and genotypic antibiotic resistance among 3GCR-EC isolates from child fecal samples collected in 2021 after the COVID-19 pandemic began compared to samples collected in 2019 before the pandemic.AM Ampicillin; CAZ Ceftazidime; CIP Ciprofloxacin; CTX Cefotaxime; CZ Cefazolin; FEP Cefepime; GM Gentamicin; IPM Imipenem; SXT Trimethoprim/ sulfamethoxazole; TE Tetracycline.MDR Multidrug resistant (3 + classes).*Adjusted Prevalence Ratios (aPR) estimated with logistic regressions (Poisson for Total Classes and Total ARGs) using Generalized Estimating Equations to adjust for repeated measures, and adjusted for confounding by including the following covariates: child age, child sex, household wealth (asset score), and caregiver education.(-indicates model did not converge due to small numbers.)95% CI 95% Confidence Intervals.