We found that each unit of improvement in the household hygiene index was associated with a ≈30–50% reduction in the likelihood of detecting antimicrobial-resistant bacteria within people. Importantly, as hygiene improved, the effects of using antibiotics became increasingly apparent. That is, poor hygiene modifies the effects of antibiotic use, but this modification is not a strong issue when hygiene conditions are very good. This is consistent with studies in diverse contexts indicating that hygiene may play an important role in the distribution and persistence of antibiotic resistant bacteria within communities in low- and middle-income countries5,6,16,17,27. An immediate implication of this interaction is that efforts to improve antibiotic stewardship, including control of unregulated access to antibiotics, may have little immediate impact on the overall prevalence of antimicrobial resistant bacteria when poor hygiene conditions are prevalent.
Given the considerable increase in odds in detecting antimicrobial-resistant bacteria across our household hygiene scale, we surmise that bacterial transmission is the primary mechanism influencing the prevalence of antimicrobial-resistant bacteria. We also found that individuals who reported boiling their raw milk before consumption exhibited a significantly lower prevalence of bacteria that were resistant to amoxicillin, ceftazidime, chloramphenicol, ciprofloxacin and tetracycline compared to those who consumed raw milk, drank powdered or packaged milk, and those that reported not consuming any milk. These results, specifically in relation to households who consumed raw milk, are consistent with an earlier report showing that Maasai pastoralists in Tanzania who boiled their milk exhibited reduced prevalence of antimicrobial-resistant bacteria17. Hygiene practices may also account for the higher predicted resistances for those who used powdered or packaged milk. About one-third of households reported not treating their water and mixing this with powdered milk would increase the risk of transmission and sickness. Packaged milk was reported to be stored for longer periods compared to fresh milk, which could also increase the risk of transmission and sickness if sub-standard storage practices (e.g., lack of refrigeration) are used. The higher prevalence of resistance in households reporting no milk consumption is more challenging to explain. It may be that households reporting no milk consumption are consuming other higher risk alternatives which could also be related to the ability to purchase milk. To examine this further, we generated a correlation matrix and found a weak positive correlation between milk consumption and wealth (r = 0.13) and between milk consumption and household hygiene (r = 0.13) (see supplement Table S3). Further study is likely needed to understand the constraints and substitutions that impact milk consumption in these communities. While the relationship between milk consumption and resistance is clearly complex, the different environments and cultures for which milk handling practices appear important (e.g., Guatemalan and Tanzanian communities), suggests that milk hygiene practices may play an important role in the transmission and persistence of antimicrobial resistance within low-income communities. However, further study is needed to determine the extent of improvement that might be achieved through mitigation of milk hygiene practices.
As with other studies conducted in low- and middle-income countries, we document a positive association between living in more urban areas and antimicrobial resistance. And as with other low- and middle-income countries, poor access to clean water, poor hygiene and sanitation conditions are evident in Guatemala although the extent of these issues differ based on urban and rural settings28,29. For the present study, compared to rural households, individuals living in urban areas had a ~ 170% increase in the odds of harboring bacteria resistant to ampicillin, amoxicillin, ceftazidime, chloramphenicol, ciprofloxacin and the MDR phenotype. It is worth emphasizing that the relationship between urban living and the odds of detecting antimicrobial-resistant bacteria only emerges after controlling for antibiotic use, hygiene and sanitation, and age differences. Without controlling for these differences, the prevalence of resistant bacteria is seemingly higher for most antibiotics in rural areas than in urban areas (see Fig. 2). After controlling for hygiene and antimicrobial use, this relationship becomes inverted, suggesting that there are likely other factors contributing to a higher prevalence of antimicrobial resistance in urban areas.
Antibiotic use had little measurable impact in sub-optimal hygiene conditions. The overall prevalence of antibiotic-resistance phenotypes is consistent with the relative availability and cost of antibiotics sold without a prescription in these communities (e.g., sold at shops called tiendas). Medications in Guatemala are subsidized by the government through the Ministry of Health system and the National Social Security health care system, yet frequent stock-outs force most Guatemalans to purchase medications out-of-pocket at private pharmacies29. In these establishments, medications can be up to 20 times the international listed price [e.g., costing up to 15 days wages for third-generation cephalosporins30]. In contrast, amoxicillin and tetracycline are considerably more affordable and widely available in tiendas, perhaps reflecting antibiotic use in these communities where the average prevalence of resistance to amoxicillin and tetracycline was approximately 40%.
We also detected several cases where factors including antibiotic use, hygiene, and diarrheal episodes were correlated with both an increase and a decrease in the odds of harboring detectable levels of antimicrobial-resistant bacteria. For example, while antibiotic use was mostly correlated with higher odds of detecting resistance bacteria, it was correlated with lower odds of detecting bacteria resistant to Kanamycin (see Table 4). There are likely two mechanisms underlying these observations. First, some resistance traits may have increased in prevalence despite the absence of commensurate use of corresponding antibiotics (e.g., diarrheal episodes and resistance to chloramphenicol). These changes always occurred in the context of similar changes in other antibiotic-resistance phenotypes, and one likely explanation is co-selection that occurs when the genes encoding these resistance traits are genetically linked. In essence, selection for one resistance phenotype co-selects for any linked traits. In some cases, it is also possible that antibiotic use can “filter” a population, by favoring strains that have the associated resistance gene and this would simultaneously increase the prevalence of any other genetic resistance genes found with these strains (i.e., “co-selection”). At the same time there would be a decrease in the prevalence of strains that do not harbor a resistance gene for the antibiotic being used. A potential example of this is the relationship between the likelihood of detecting antimicrobial resistance with recent episodes of diarrhea, where prevalence of resistance to ceftazidime, kanamycin and sulfamethoxazole increased, but prevalence of resistance to chloramphenicol and streptomycin decreased.
As with any epidemiological study, we are limited to identifying correlated variables for largely uncontrolled systems, making clear cause-and-effect relationships difficult to identify due to confounded variables. An example of this is the apparent higher prevalence of resistance in rural households shown in the univariate comparison (Fig. 2) vs. the statistically higher likelihood of detecting resistant bacteria in urban households once several other variables are controlled. Furthermore, our analysis of antibiotic use may be compromised by limitations of participant recall31, and the commensurate limitations on the ability to gather accurate data about the magnitude and frequency of antibiotic use. In addition, while we modeled household hygiene and sanitation as a linear variable, hygiene and sanitation are clearly complex phenomena that include many interacting factors so that an increase in one factor (handwashing frequency) likely does not reflect the same impact on the prevalence of antimicrobial resistance as another (e.g., improved toilet). Nevertheless, we argue our composite measure represents a measure of of household hygiene and sanitation with changes in the scale reflective of general increases and decreases in household hygiene and sanitation.
The robust relationship between hygiene and resistance in the sampled Guatemalan communities, along with the interaction between hygiene and antibiotic use, provides important implications for the efficacy of stewardship efforts globally when aggregate hygiene levels are compromised. In such cases, investment in infrastructure to improve hygiene can be easily justified as a tool to limit the proliferation of antimicrobial resistance in communities across the globe. As this study highlights, assigning priorities and subsequent development of targeted strategies will require analysis of a greater spectrum of living conditions, using cross-cultural investigations developed and implemented by interdisciplinary teams from the natural and social sciences.
Ethics committee approval
The study protocol was approved by Washington State University in Pullman Institutional Review Board (15895-001), the Universidad del Valle de Guatemala-Center for Health Studies Ethics committee (159-01-2017), and the Guatemalan Ministry of Health Ethics Committee (10-2017).