Variation in mechanisms of resistance has implications for the choice of antimicrobial therapy and the evaluation of strategies to minimize resistance. Adopting the individual and population-level perspective informs therapeutic decision-making, clinical study design, and public policy.
In TB, preventing the emergence of resistance in a treated host is a sound policy for averting the emergence of resistance at the population level as well (although once resistant strains have emerged, special measures are required to contain them [30
]). With respect to antimicrobial resistance, what is good for the patient is good for the population.
In contrast, for other types of resistance, antimicrobial treatment may exert individual-level effects that are substantially different in magnitude or even opposite in direction to that of population-level effects. Treatment with a beta-lactam may produce only a small, short-lived increase in the treated patient's odds of carrying or being infected by a resistant pneumococcus (7)
. In some cases, treatment may actually eradicate carriage of a resistant organism, thereby reducing the individual’s risk of resistant carriage. Small or unobservable effects on individual risk have been observed in other cases as well, such as vancomycin use for VRE (8
) and the use of various antibiotics for infections with resitant gram-negative rods (32)
. In these cases, preventing resistance in the treated patient may not be the central goal of a prudent antibiotic use policy; instead, treatment should seek to minimize the advantage it provides to resistant organisms in the community or the hospital as a whole, subject to the constraint of providing effective treatment for the patient.
The considerations of the distinctive biologic and epidemiologic mechanisms of antibiotic resistance in different pathogens lead to several broad suggestions for future studies. First, the optimal study design to estimate individual-level effects of antibiotics on colonizing organisms such as VRE and beta-lactam resistant S. pneumoniae is to measure acquisition and loss rates in an observational cohort or experimental study where subjects are serially cultured before, during, and after antibiotic therapy (23
). Time-to-event statistical models (e.g., Cox proportional hazards regression) are appropriate analytic methods for these kinds of studies (23
). This design allows investigators to distinguish between the effects of antimicrobials on the risk for acquisition (colonization) and their effects on the risk for clinical infection once an patient has been colonized with a resistant organism.
As a consequence of the mechanisms we have described, the magnitude of an antibiotic’s effect on a patient’s risk of resistant colonization or infection may be dependent on his or her exposure to potential transmission of resistant organisms (14)
. Stated differently, the frequency of contact with others carrying the resistant organisms is likely an important effect-modifier of antibiotic effects for pathogens that do not follow the simple model of emergence of resistance exhibited by organisms such as M. tuberculosis. Individual-level antibiotic effects mediated by alterations in colonization resistance or killing of susceptible bacteria may be greater in settings of high exposure to resistant organisms, for example, during outbreaks (7)
. Controlling for transmission risk or measuring effects conditional on a specified level of transmission risk is advised, when possible. Standard analytic approaches make the assumption that outcomes in different subject are independent, but this assumption is violated in the case of infectious diseases. Use of one of these strategies to model exposure to transmission will help to account for this non-independence of outcomes in different persons in the same study (13
One practical result of quantifying direct, individual-level antibiotic effects is to provide information on the short-term risk of infection with a resistant organism to a person about to initiate antibiotic treatment. This hazard needs to be taken into account when weighing the risks and benefits of use of antimicrobial agents in individual patients. However, analogous to the evaluation of vaccine programs, combined direct and indirect antibiotic effects carry increased importance from the public health and policy management perspective (38
). The measurement of population-level effects of antimicrobials also has educational value in demonstrating to clinicians and patients the extent to which individual antibiotic use choices have negative consequences for the population as a whole. Such a conflict between individual benefit and the population’s harm is an example of what economists term an “externality” or what environmentalists have called the ‘Tragedy of the Commons” (40)
To estimate overall antibiotic effects from data requires group-level studies. Observational group-level studies may lack sufficient data to avoid confounding and other causes of ecologic bias (41)
. For this reason, studies that estimate the effects of individual- and group-level antimicrobial use are generally preferable to ones that contain group-level data alone. Depending on the context, the appropriate group(s) may include the family, the community, the hospital, or the hospital unit or department (42
). Further research is necessary to evaluate hierarchical regression methods and compare results obtained from different levels of analysis (44)
For the most accurate measurement of overall antibiotic effect on resistance in communities, a cluster-randomized intervention trial is appropriate (45)
. In cluster-randomized trials, the unit of randomization is a group such as a community or a hospital, and multiple units (sometimes as few as six, but often more) are assigned to each of two (or more) treatment arms. We are not aware of published studies using this design to evaluate antibiotic resistance, although we know of two in progress (R. Platt, pers. comm.) (12)
. However, this design has been used in other areas of infectious disease epidemiology for which group level effects are important (such as vaccination programs), and it is considered the standard design for investigations of the effects of insecticide-impregnated bednets in preventing malaria (45
). In the context of antimicrobial resistance, cluster-randomized trials have two key advantages. First, unlike studies that gather individual-level data alone, they provide the opportunity to observe the indirect effects of treatment on resistance. Second, they provide a clean way to avoid the statistical problems of nonindependence between patients in a study that may reduce the power or increase the false-positive rate of observational studies. In cluster-randomized studies of antimicrobial resistance, both the incidence rate of infection with resistant organisms in the population and the ratio of resistant to susceptible (or proportion of total organisms that are resistant) would be appropriate study endpoints.