Discussion

The term “herd effect” is often applied to imply protection of nonimmune persons against a specific disease when most others in the population are immune (hence the usual term, “herd immunity”), either because of natural infection ^{Reference Ribeiro, Hamer, Diallo, Kitron, Ko and Weaver9} or of mass vaccination. ^{Reference Lahariya10,Reference Fine, Eames and Heymann11} This topic has been a central issue in COVID-19 epidemiology debates worldwide. ^{Reference Redwan12,Reference Tkachenko, Maslov, Elbanna, Wong, Weiner and Goldenfeld13} Less frequently, herd protection has been addressed in other interventions, such as improvements in sanitation and hygiene. ^{Reference Fuller and Eisenberg14} Presently, there is scarce (if any) evidence on herd effect in infection control or antimicrobial resistance studies, except for “one health” approaches of antimicrobial resistance in agriculture. ^{Reference Call, Davis and Sawant15} As stated by Lipsich and Samore, ^{Reference Lipsitch and Samore6} the “herd effect” of antimicrobial use refers to increasing risk of acquisition of resistant organisms by persons exposed to mass use of antibiotics. Those indirect effects may be relevant in preventing the emergence and spread of multidrug-resistant organisms.

We focused on that gap in the epidemiological approaches to antimicrobial resistance within healthcare settings. We were specifically interested in the indirect impact of antimicrobial use on the incidence of CR-GNB, which are hyperendemic in ICUs in Brazil. ^{Reference Labarca, Salles, Seas and Guzmán-Blanco16,Reference Sampaio and Gales17} As expected, ^{Reference Braga, Campos, Batistão, Gontijo Filho and Ribas18} we found not only a massive use of carbapenems but also a relevant incidence of CR-GNB, even among patients to whom carbapenems were not prescribed.

Our findings provide evidence of association of overall use of carbapenem in ICUs on the incidence of CR-GNB infections among patients not using those antimicrobials. Antimicrobial use seems to have an important indirect effect on carbapenem resistance, both from collective and individual perspectives. This finding is in line with current recommendations in epidemiological studies, which stress the relevance of ecologic studies supplemented with individual-level information. ^{Reference Morgenstern19,Reference Wakefield20} This relevant methodological aspect is best suited to analyze effects of exposures that are essentially collective.

The theoretical construct of indirect impacts on acquisition of antimicrobial resistance has been previously addressed regarding colonization pressure, described as a quantitative approach to the exposure to patients harboring multidrug-resistant organisms. ^{Reference Ajao, Harris and Roghmann21} Indeed, previous studies have demonstrated the effect of colonization pressure on the spread of CR-GNB. ^{Reference Castelo Branco Fortaleza, Moreira de Freitas and da Paz Lauterbach22,Reference DalBen, Basso and Garcia23} However, the current approach points to causality as a continuum of factors linking the exposures to the outcomes. ^{Reference Causality24} In a cause-directed acyclic graph, the colonization pressure may be an intermediary intersection between antimicrobial use and the acquisition of an antimicrobial-resistant organism by patients not exposed to the antibiotic of interest. ^{Reference Pearl25} This is the appropriate theoretical construct if we are to propose interventions based on antimicrobial prescription to prevent antimicrobial resistance. Those interventions should aim at the collective, focusing not only on antimicrobial formularies, but also at environmental determinants (eg, inanimate reservoirs, ^{Reference Morgan, Rogawski and Thom26} prior room occupants, ^{Reference Ajao, Johnson and Harris27} and even weather ^{Reference Fortaleza, Caldeira and Moreira28} ). Nonetheless, our results agree with previous studies pointing to the ecologic relevance of antimicrobial stewardship. ^{Reference Tacconelli1,Reference Davey, Marwick and Scott5,Reference Álvarez-Marín, López-Cerero and Guerrero-Sánchez29}

Our study had several limitations, mostly due to the observational, retrospective, and the partially ecologic approaches. ^{Reference Harris, Karchmer, Carmeli and Samore2,Reference Wakefield20} We did not directly address interventions on antimicrobial prescription formularies. Also, we did not analyze the impact of colonization pressure or environmental reservoirs. Finally, both the correlation coefficient in the ecological study and the incidence rate ratio in the cohort present modest (though statistically significant) effect. However, in an ICU with a monthly average 346 DOT of carbapenem use, decreasing a single DOT reduces the risk of CR-GNB acquisition among those not using carbapenems by ∼1%. Therefore, interventions aimed at reducing that use are likely to have a beneficial impact on that population (and on those for whom the carbapenem prescription would be prevented). Our study had several strengths as well. We performed an extensive chart review which allowed us to test the effect of “carbapenem pressure” adjusting for patients’ severity, time at risk, and use of other antimicrobials. Also, the simultaneous ecologic and individual-based approaches allowed us to infer causality from a population perspective. ^{Reference Lipsitch and Samore6}

In conclusion, total use of carbapenems in the ICU was independently associated with acquisition of CR-GNB by patients who did not use carbapenems. Although further studies (with mathematical models, quasi-experimental approaches or including other environmental confounding factors) are required to validate our findings, we found relevant evidence for a population (“herd-effect”-like) impact of antimicrobial use in the ICUs.