Evaluation of antibiotic purchase data for ceftiofur and enrofloxacin and minimum inhibitory concentrations among Escherichia coli isolates from swine farms in the Midwestern United States using multiple statistical models

Antimicrobial resistance (AMR) is considered a threat to human and animal health on a global scale (World Health Organization, 2015). Numerous studies have investigated the relationship between antibiotic use and resistance in pathogens that could be transmitted between animals and humans (Chadwick et al., 1996, Hummel et al., 1986, Klare et al., 1995a, Klare et al., 1995b, Schouten et al., 1997, Witte, 2000). The suggestions to manage AMR through reduction of antibiotic use in food-producing animals is premised on the concept that harboring antimicrobial resistance genes exacts a fitness cost to the bacteria in the absence of antibiotic use; therefore, by reducing or removing antibiotic selection pressure, bacteria will gain a competitive advantage by losing resistance genes (Andersson, 2003, Penkova and Raymond, 2024). However, this fitness and resistance relationship is complex, and there are other factors that influence whether or not resistance is related to bacterial fitness (Anderson et al., 2023). Certainly, the relationship between antibiotic use and resistance is complicated and is expected to vary by drug class, resistance mechanism, and bacterial pathogen.

Escherichia coli bacteria are a highly diverse species of gram-negative enteric bacteria that can act as commensal, pathogenic, or zoonotic organisms under different contexts. E. coli and other Enterobacteriaceae can be associated with extraintestinal infections in humans and it is believed that at least some of these strains originally colonized the human host from a foodborne source (Manges and Johnson, 2012), even though the link between AMR genes in bacterial strains from livestock and pathogenic human strains remains unclear (Ludden et al., 2019). Nonetheless, E. coli harboring resistance against critically important antibiotics such as third-generation cephalosporins and fluoroquinolones were identified as a human health threat for extra-intestinal E. coli infections (World Health Organization, 2019). Furthermore, since non-pathogenic E. coli also carry circulating AMR genes, they are considered both a reservoir of, and an indicator for, AMR resistance in a given bacterial population (European Food Safety Authority and European Centre for Disease Prevention and Control, 2012). Therefore, in this paper we evaluated E. coli isolates as indicators of resistance in commercially raised pigs.

Antibiotic use and its relation to resistance is a highly researched topic without consistent results across all antimicrobial drugs or classes, including in many studies involving swine. One study evaluated the impact of injectable ceftiofur and penicillin use in pigs and found that use during the finishing period did not impact AMR gene distribution (Gaire et al., 2022). Another study showed that pigs seeded with an E. coli strain that had a plasmid bearing a β-lactamase resistance gene would shed more resistant E. coli strains after treatment with a cephalosporin (Fleury et al., 2015). Other studies showed use of antibiotics could increase resistance to some antibiotic classes but not others (Alali et al., 2009, Burow et al., 2019), or increase resistance in antibiotic classes not used, suggesting co-selection in the animal, but no difference in the AMR in isolates from the pork harvested (Lugsomya et al., 2018). Other population-based studies, captured a trend of decreasing antibiotic use and resistance in E. coli, but these authors also discuss that the data sets were small and that studies should be performed at the species level (Callens et al., 2018). Our study focuses on farm-level data rather than aggregate national or regional level data and will add considerable granularity to this discussion.

Antibiotic sensitivity tests that use broth microdilution returns the bacteria’s minimum inhibitory concentration (MIC) for an antibiotic. MIC data is hierarchical categorical data that do not necessarily have an underlying distribution. This makes choosing the appropriate statistical model challenging. Each model has various strengths and weaknesses in the generated results, such as misspecification bias and poor model fit (Michael et al., 2020, Aerts et al., 2022). Another study that used multiple statistical models to assess use and resistance, but that dichotomized isolates into resistant versus susceptible, could not define an ideal model and compared the various model results to identify trends (Noyes et al., 2016). An additional study evaluating multi-drug resistance in Salmonella isolates from poultry used generalized linear mixed models and accelerated failure time frailty models and found that the models returned different levels of sensitivity on the characteristics of the temporal trends associated with resistance and that censoring could make the models unstable (Bjork et al., 2015). Choice of statistical models is challenging when evaluating MIC data.

Globally, swine production has been estimated to represent the greatest mass of antimicrobial consumption among livestock species (Mulchandani et al., 2023), and also has the highest apparent consumption in the United States (Food and Drug Administration (FDA), 2022). Understanding the relationship between antimicrobial use (AMU) in animal and human hosts and resistance among bacteria is important and is shown to vary across bacterial species and by antibiotic class. There are few longitudinal studies in the literature evaluating the relationship and trends between AMR and AMU in commercially raised swine. The objective of this study was to assess the AMU and AMR relationship separately for ceftiofur and enrofloxacin, antibiotics commonly used in the US swine industry. This was done by evaluating multiple statistical models for agreement on significant explanatory variables of milligrams of purchased antibiotic over kilograms of pig produced, pig type (substandard versus sick), season, and farm type evaluated against the minimum inhibitory concentrations of enrofloxacin and ceftiofur from E. coli isolated from pooled pig samples.

1. Download: Download high-res image (277KB)
2. Download: Download full-size image

Fig. 1. Univariable Kaplan-Meier graphs comparing the growth of E. coli isolates from a farm and the quartile of ceftiofur antibiotics purchased among farms involved in the study. 1a: E. coli isolated from intestinal samples collected on breed-to-wean (BTW) farms (log rank test p-value<0.0001); 1b: E. coli isolated from fecal swabs collected on breed-to-wean (BTW) farms (log rank test p-value=0.0011); 1c: E. coli isolated from intestinal samples collected on wean-to-market (WTM) farms (log rank test p-value<0.0001); 1d: E. coli isolated from fecal swabs collected on wean-to-market (WTM) farms (log rank test p-value=0.9274).

1. Download: Download high-res image (250KB)
2. Download: Download full-size image

Fig. 2. Univariable Kaplan-Meier graphs comparing the growth of E. coli isolates from a farm and the quartile of enrofloxacin antibiotics purchased among farms involved in the study. 1a: E. coli isolated from intestinal samples collected on breed-to-wean (BTW) farms (log rank test p-value=0.062); 1b: E. coli isolated from fecal swabs collected on breed-to-wean (BTW) farms (log rank test p-value=0.014); 1c: E. coli isolated from intestinal samples collected on wean-to-market (WTM) farms (log rank test p-value=0.051); 1d: E. coli isolated from fecal swabs collected on wean-to-market (WTM) farms (log rank test p-value=0.4526).

In all models completed with ceftiofur purchases, the MIC of 0.25 µg/ml was the referent category. Although the different models assessed risk using different methods, thus resulting in different ratios or risks, the trends were similar. For the Cox proportional hazard model, the hazard ratios estimated the hazard of being susceptible (no growth), while all other models assessed the odds or risk of having an MIC value in a higher category (less susceptible to the antibiotic) than the referent category. As a result, the hazard ratios were inverted compared to all other models’ ratios and risks. For isolates from intestinal samples, in the Cox proportional hazard models with and without frailty, the BTW enterprises had a 1.25 (without frailty 95 % confidence interval (95 % CI): 1.08, 1.46; with frailty 95 % CI: 1.07, 1.46) greater hazard of susceptibility than WTM sites, and a one-unit increase in the mg/kg of purchased ceftiofur resulted in a decreased hazard ratio (HR; HR=0.87; with and without frailty 95 % CI: 0.83, 0.92) or decreased susceptibility. For isolates from fecal swabs, the same trend was seen for both models, with the BTW enterprise exhibiting a 1.26 (without frailty 95 % CI: 1.06, 1.50; with frailty 95 % CI: 1.06, 1.51) greater hazard of susceptibility than WTM, and a one-unit increase in mg/kg of purchased ceftiofur showing a lower hazard of susceptibility (HR=0.95; without and with frailty 95 % CI: 0.90, 0.99). The parametric survival models returned time ratios (TR), though MIC values were assessed rather than time. For both the parametric survival model with and without clustered standard errors (SE) on isolates from intestinal samples, BTW sites had median MIC values lower than WTM sites (TR=0.88; without clustered SE 95 % CI: 0.81, 0.95; with clustered SE 95 % CI: 0.81, 0.96). For a one unit increase in mg/kg of purchased ceftiofur, the MIC increased by 1.06 (without clustered SE 95 % CI: 1.04, 1.09; with clustered SE 95 % CI: 1.02, 1.11) as well. For the mixed effects parametric survival model, the TR for farm type (BTW versus WTM) was 0.82 (95 % CI: 0.70, 0.96) and for mg/kg of purchased ceftiofur the TR was 1.14 (95 % CI: 1.09, 1.19). For the fecal swab isolates, none of the parametric survival models without mixed effects had convergence. The mixed effects model had a TR of 0.76 (95 % CI: 0.62, 0.92) for the farm type comparison and 1.07 (95 % CI: 1.02, 1.13) for the mg/kg of purchased ceftiofur variable. For multinomial models, each level of the MIC from 0.5 to 8 µg/ml for ceftiofur was assessed against the 0.25 µg/ml MIC concentration category using a model with and without clustered standard errors. The range of significant relative risk ratios (RRR) is provided in Table 5. Among intestinal isolates, a one-unit increase in mg/kg of purchased ceftiofur resulted in higher MIC values when the MIC was 4 µg/ml (RRR=1.51; 95 % CI: 1.13, 2.00)) and the MIC was 8 µg/ml (RRR=1.31; 95 % CI: 1.07, 1.60), when compared to the referent 0.25 µg/ml MIC value. Farm type was also significantly associated with MIC values, with the model indicating a lower RRR for BTW relative to WTM enterprises at MIC values of 1 µg/ml, 2 µg/ml, and 4 µg/ml with RRRs of 0.31 (95 % CI: 0.14, 0.70), 0.09 (95 % CI: 0.02, 0.40), and 0.31 (95 % CI: 0.10, 0.98), respectively. Among fecal swab isolates, mg/kg of purchased ceftiofur was only statistically significant at the MIC of 8 µg/ml in the multinomial model without clustered SE (RRR=1.24; 95 % CI: 1.03, 1.50). The RRR comparing BTW to WTM sites was significant for both models for an MIC of 1 µg/ml (RRR=0.24; without clustered SE 95 % CI: 0.08, 0.74, with clustered SE 95 % SE: 0.09,0.67), and for an MIC of 2 µg/ml (RRR=0.18; 95 % CI: 0.05, 0.71) in the model with clustered SE. The mixed effects model did not converge when isolates from intestinal samples were assessed and no variables were statistically significant when the fecal swab isolates were assessed.

Among intestinal isolates, the ordinal regression model resulted in statistically significant results for BTW versus WTM farms (OR=0.34; 95 % CI: 0.24, 0.50)) and a one-unit increase in mg/kg of purchased enrofloxacin (OR=1.22; 95 % CI: 1.08, 1.39). For the fecal swab isolates, similar results were seen with BTW versus WTM farms (OR=0.26; 95 % CI: 0.18, 0.37) and a one-unit increase in mg/kg of purchased enrofloxacin (OR=1.27; 95 % CI: 1.15, 1.39). The proportional odds were violated when using both intestinal and fecal swab isolates for the ordinal continuous ratio model. Among the multinomial models, with and without clustered SEs used to assess intestinal isolates, farm type and mg/kg of purchased enrofloxacin were both statistically significant (P < 0.05). For both models, the MIC value of 0.12 µg/ml was the referent category. For farm type, the RRRs were the same for both models. BTW sites had a lower risk than WTM sites for MIC values of 0.25 µg/ml (RRR=0.27; without clustered SE 95 % CI: 0.10, 0.76; with clustered SE 95 % CI: 0.08, 0.90), 0.5 µg/ml (RRR=0.29; without clustered SE 95 % CI: 0.17, 0.51; with clustered SE 95 % CI: 0.15, 0.56), and 1 µg/ml (RRR=0.38; without clustered SE 95 % CI: 0.21, 0.68; with clustered SE 95 % CI: 0.20, 0.74). The mg/kg of purchased enrofloxacin variable was statistically significant only at the MIC value of 0.5 µg/ml and had an RRR of 1.33 (without clustered SE 95 % CI: 1.12, 1.58; with clustered SE 95 % CI: 1.14, 1.56) for both models. The mixed effects model did not converge when used to assess isolates from intestinal samples, and no results were available. Among the multinomial models, with and without clustered SEs used to assess fecal swab isolates, farm type and mg/kg of purchased enrofloxacin were also statistically significant both models yielded the same RRRs. Specifically, BTW farm sites compared to WTM sites had a lesser relative risk ratio than WTM sites at all MIC values, as follows: at an MIC of 0.25 µg/ml the RRR was 0.24 (without clustered SE 95 % CI: 0.11, 0.52; with clustered SE 95 % CI: 0.08, 0.66), at an MIC value of 0.5 µg/ml the RRR was 0.26 (without clustered SE 95 % CI: 0.15; 0.44, with clustered SE 95 % CI: 0.14, 0.47), at an MIC value of 1 µg/ml the RRR was 0.32 (without clustered SE 95 % CI: 0.19, 0.56; with clustered SE 95 % CI: 0.16, 0.65), and at an MIC value of 2 µg/ml the RRR was 0.12 (without clustered SE 95 % CI: 0.04, 0.37; with clustered SE 95 % CI: 0.03, 0.39). The mg/kg of purchased enrofloxacin variable was significant at an MIC of 0.5 µg/ml and an MIC of 1 µg/ml with RRRs of 1.26 (without clustered SE 95 % CI: 1.10, 1.45; with clustered SE 95 % CI: 1.07, 1.48) and 1.30 (without clustered SE 95 % CI: 1.14, 1.49; with clustered SE 95 % CI: 1.10, 1.54), respectively. The mixed effects model returned similar statistically significant results for isolates from fecal swabs, with farm type exhibiting statistically significant associations across all MIC values: an RRR of 0.17 (95 % CI: 0.05, 0.45; MIC=0.25 µg/ml), an RRR of 0.22 (95 % CI: 0.11, 0.47; MIC=0.5 µg/ml), an RRR of 0.27 (95 % CI: 0.13, 0.58; MIC=1 µg/ml), and an RRR of 0.09 (95 % CI: 0.02, 0.38; MIC=2 µg/ml). The omnibus test of mg/kg or purchased enrofloxacin was not statistically significant in the mixed effects model, but the variable was forced into the model and was significant at an MIC of 0.5 µg/ml (RRR=1.32; 95 % CI: 1.07, 1.62) and an MIC of 1 µg/ml (RRR=1.37; 95 % CI: 1.12, 1.67). Finally, the generalized ordinal regression model with and without clustered standard errors also indicated that farm type and mg/kg of purchased enrofloxacin were statistically significantly associated with MIC values, The ORs were the same for each model (with and without clustered SEs) for both the intestinal isolates and the fecal swab isolates. For the intestinal isolates, the BTW versus WTM comparison was statistically significant at the MIC values of 0.25, 0.5 and 1 µg/ml with ORs of 0.33 (without clustered SE 95 % CI: 0.23, 0.48; with clustered SE 95 % CI: 0.20, 0.54), 0.35 (without clustered SE 95 % CI: 0.24, 0.52; with clustered SE 95 % CI: 0.22, 0.57), and 0.46 (without clustered SE 95 % CI: 0.28, 0.75; with clustered SE 95 % CI: 0.26, 0.81), respectively. The mg/kg of purchased enrofloxacin variable was significant at MIC values of 0.25 and 0.5 µg/ml with ORs of 1.25 (without clustered SE 95 % CI: 1.09, 1.42; with clustered SE 95 % CI: 1.06, 1.46) and 1.24 (without clustered SE 95 % CI: 1.08, 1.42; with clustered SE 95 % CI: 1.06, 1.44), respectively. For the fecal swab isolates, the farm type variable (BTW versus WTM) was statistically significant at all levels of MIC assessed (0.25, 0.5, 1, and 2 µg/ml) with ORs of 0.25 (without clustered SE 95 % CI: 0.18, 0.36; with clustered SE 95 % CI: 0.16, 0.41), 0.27 (without clustered SE 95 % CI: 0.18, 0.40; with clustered SE 95 % CI: 0.16, 0.45), 0.30 (without clustered SE 95 % CI: 0.18, 0.51; with clustered SE 95 % CI: 0.15, 0.59), and 0.14 (without clustered SE 95 % CI: 0.04, 0.51; with clustered SE 95 % CI: 0.03, 0.62), respectively. The mg/kg variable was statistically significant at all values except MIC of 2 µg/ml. For an MIC of 0.25 µg/ml the OR was 1.26 (without clustered SE 95 % CI: 1.14, 1.39; with clustered SE 95 % CI: 1.08, 1.47), an MIC of 0.5 µg/ml the OR was 1.30 (without clustered SE 95 % CI: 1.17, 1.44; with clustered SE 95 % CI: 1.12, 1.51), and an MIC of 1 µg/ml the OR was 1.29 (without clustered SE 95 % CI: 1.12, 1.48; with clustered SE 95 % CI: 1.06, 1.57).

In this study, we summarized the relationship of presumptive antibiotic use, based on purchase records and number of pigs produced, to antibiotic MIC values for enrofloxacin and ceftiofur from pig E. coli isolates while controlling for farm type, season, and pig health status through evaluation of the agreement between different multivariable modeling approaches.

By evaluating multiple statistical models, we discovered a consistent association between farm type and the amount of ceftiofur and enrofloxacin purchased at an enterprise level, and phenotypic E. coli resistance as measured via the MIC. Interestingly, BTW isolates had a lower odds, hazard, or risk of having higher MIC values compared to WTM isolates, despite the fact that BTW sites had higher median mg/kg of antibiotic purchases as well as more isolates from farms in the fourth quartile of antibiotic purchases per kg of pig produced.

The relationship detected between mg/kg of antibiotic purchased and MIC was similar to another experimental study that evaluated therapeutic and subtherapeutic doses of enrofloxacin and ceftiofur on E. coli resistance 10 days after the last antibiotic exposure. The study reported that resistant E. coli populations were more prevalent in the subtherapeutic group and therapeutic enrofloxacin and ceftiofur exposure groups relative to controls that were not exposed (Lin et al., 2017). Yet, that study only looked at a short-time period of antibiotic exposure, with a short follow-up period. Another study evaluating the effect of ceftiofur treatment on fecal E. coli recovered from cattle revealed a temporal effect on the likelihood of recovering a resistant isolate, which lasted from 13 to 15 days post-exposure (Lowrance et al., 2007). The study described herein did not capture any data about the time lapse between antibiotic exposure and sample collection, which would have improved results interpretation. More robust longitudinal sampling could improve understanding of antibiotic exposure effects on bacteria isolated from commercial pigs.

Clearly, antimicrobial resistance is complex with many variables to consider. Our study looked at weaning-aged pigs (i.e., 21–28 days of age) at BTW sites and in older-aged pigs at WTM sites. WTM sites have the widest potential range of pig ages from 21 days to 6 months; however, pig age was not consistently captured at the time of sampling. It is well-documented that neonatal, suckling pigs had the highest E. coli resistant to antibiotics across age groups in longitudinal studies (Cameron-Veas et al., 2016, Dohmen et al., 2017, Gaire et al., 2021). This study evaluated pigs at the weaning age only (BTW sites) and that may be one reason the WTM sites resulted in having higher MIC values overall. The trend of having higher level of antimicrobial resistance in neonatal pigs may still exist, but for these reasons, were not seen here. In addition, recent and current antibiotic treatments that shortly preceded or occurred contemporaneously with sampling were not captured, which could also potentially bias the results, particularly if treatments have a short-term effect on resistance prevalence. Another potentially confounding factor was the inclusion of sick and substandard pigs in the May 2020 through August 2022 period (period 1), but only healthy pigs in the July 2022 through October 2023 period (period 2). The inclusion of sick pigs in period 1 could have made pathogenic E. coli more likely to be recovered in the corresponding isolates, which could have skewed the resistance profile of the isolate set as compared to period 2. The strain-level profile of recovered E. coli is likely an important factor in phenotypic resistance profiles and could explain the significant difference in MIC values observed between intestinal isolates recovered from sick versus substandard pigs. However, commensal E. coli are considered an indicator for AMR gene presence, so the findings here likely represent the bacterial populations present (European Food Safety Authority and European Centre for Disease Prevention and Control, 2012). Our sampling design did not equally space sampling over time; instead, it was at the convenience of the veterinarian with encouragement to sample every six months. The practicality of needed veterinary services resulted in sites being sampled at intervals greater than or less than six months. Future studies would do well to capture farm type, use, age, and duration since last treatment at sampling to better understand these relationships. Standard sampling time points should also be set to ensure representative sampling over time.

Furthermore, using purchased data as a proxy for antibiotic use contained challenges. Purchase data is an aggregate metric at the enterprise or barn-level, but not at the pig-level. Many antibiotics are given as water medication to the entire barn, which can make it difficult to quantify the final amount a pig received without further assumptions and estimations. and are Injectable antibiotics, such as enrofloxacin and ceftiofur, are typically given to sick pigs. Finally, in BTW sites, the antibiotics account for the antibiotics given to sows and piglets (only piglets were sampled) while in WTM sites they account for the antibiotics given solely to the pigs sampled. This is another reason that farm type was likely a significant confounder in the model as well. Isolates were taken from pooled samples of pigs, but a limitation to this study is that it assumed a shared environment impacted pig-level resistance patterns.

This study provides further evidence of the relationship between use and resistance. Yet, the relationship is complex and may have been confounded in this study by missing data on pig age and time since last treatment, as discussed above. It is more likely that sick pigs received treatments of enrofloxacin and ceftiofur as these drugs are given by injection in swine medicine and not in the feed or water. Healthy pigs would not receive these treatments. During period 1, substandard pigs were compared to sick pigs in this study, substandard pigs were not treated, yet, they had a lesser hazard of being susceptible (0.64) and an increased time ratio (1.27) for MIC values when enrofloxacin purchases were assessed compared to sick pigs when using the Cox proportional hazard model and the parametric proportional hazard model, but not for other models. There was no such relationship when ceftiofur was evaluated. It is also important to note that BTW sites had greater mg/kg purchases of antibiotics relative to WTM sites, and that BTW sites still had lower MIC values compared to WTM sites, which initially implied that higher use was not related to MIC values. Clearly, the production stage matters and needs to be included in statistical models as a confounder. An additional challenge with these findings is that mg/kg antibiotic metric itself. On WTM sites, the population of pigs being treated with the purchased drug in the numerator is the same as the group being marketed in the denominator. On BTW sites, a larger population is treated with the purchased drugs (female pigs being raised for breeding, breeding female pigs, and male pigs used to detect estrus) than are in the denominator (weaned pigs produced). It is very difficult to separate out what animals were treated off paper records and, since piglets are not marketed for food, antibiotic use in that age group is not recorded. Yet, as breeding pigs are treated, the piglets that become weaned pigs are also exposed either through nursing or the placenta, so excluding sow treatments can also be troublesome. Veterinarians should be aware of the evidence that there is a relationship between use and resistance for enrofloxacin and ceftiofur, but the research community should continue to clarify the details of the relationship.

Antibiotic resistance among enteric bacteria to ceftiofur and enrofloxacin is of concern as these drugs are members of the highest priority and critically important drug classes used to treat humans with E. coli and other gram-negative bacterial infections such as salmonellosis (World Health Organization, 2019). The results of this study suggest that enterprises that purchase and, thus, potentially use, more ceftiofur and enrofloxacin have higher MIC values for ceftiofur and enrofloxacin among fecal E. coli isolates obtained from pigs raised in commercial barns, which could be suggestive of higher resistance. Future work should control for time from treatment and the age of the animal as well to ensure the observed associations are consistent across the production stages, and especially at slaughter when any risk to the food supply would be greatest.

Funding for this work was provided by the US Department of Agriculture through a cooperative agreement (Funding opportunity number: USDA-APHIS-10025-VSSP0000–22–0025) with the SDSU ADRDL, the Foundation for Food and Agriculture (Grant IDs: ICASA-0000000016, RDS-0000000001, and 22–000515), Pipestone Veterinary Services, and the National Pork Board (Grant numbers 19–239 and 21–018).

This study was approved by the Pipestone Research Institutional Animal Care and Use Committee. Protocol #2020–01.

Taylor Spronk: Project administration, Methodology, Investigation, Data curation, Conceptualization. Scott Dee: Writing – review & editing, Project administration, Methodology, Investigation, Funding acquisition, Conceptualization. Noelle Noyes: Writing – review & editing, Methodology, Conceptualization. Laura B Goodman: Writing – review & editing, Methodology, Conceptualization. H Morgan Scott: Writing – review & editing, Methodology, Conceptualization. Karyn A Havas: Writing – original draft, Validation, Project administration, Formal analysis, Data curation. Laura Ruesch: Writing – review & editing, Project administration, Methodology, Investigation, Data curation. Roy Edler: Writing – review & editing, Project administration, Methodology, Data curation. Joel Nerem: Writing – review & editing, Supervision, Methodology, Conceptualization. Marlee Braun: Writing – review & editing, Investigation.