Antimicrobial Use and Antimicrobial Resistance: A Population Perspective

Marc Lipsitch, Matthew H. Samore


Emerging Infectious Diseases. 2002;8(4) 

In This Article

Role of Mathematical Models

Transmission-dynamic modeling can also play an important role in bridging the gap between individual- and group-level effects [20,21,48,49,50]. These models take information about individual-level effects as parameters and make predictions about the response of the population to changes in such parameters as transmission risk or antibiotic usage. Although models cannot substitute for empirical intervention studies, they can be particularly valuable in at least four ways: 1) generating hypotheses about the relationship between antibiotic use and resistance that can be used in designing and prioritizing empirical studies; 2) defining the conditions under which a particular intervention is likely to work, thereby suggesting how empirical results can (and cannot) be extrapolated to other settings; 3) providing explanations for phenomena that have been observed but whose causes were uncertain; and 4) identifying biological mechanisms that, while important, remain poorly understood.

An example of models for generating hypotheses comes from the question of antimicrobial rotation or “cycling.” Cycling of antimicrobial classes in hospitals has been suggested and is currently being evaluated for its ability to curtail resistance in major nosocomial pathogens [5,51,52,53,54]. One mathematical model of this process has suggested that using a mixture of different drug classes simultaneously (e.g., if two drug classes are available for empiric therapy of certain infections, treat half of the patients with one drug class and half with the other) will reduce resistance more effectively than cycling under a broad range of conditions [19]. This suggests that such mixed regimens would be good candidates for comparison with cycling in controlled trials.

As a second example, levels of resistance in hospital-acquired pathogens may change rapidly within a matter of weeks or months after changes in antimicrobial use. By contrast, studies of reductions in antimicrobial use in communities have shown slow and equivocal effects on resistance in community-acquired pathogens [55]. Mathematical models suggest that, in communities, the key factor driving the change in resistance levels may be the "fitness cost" of resistance, i.e., resistance will decline after a reduction in antimicrobial use if resistant organisms in untreated patients are at a disadvantage for transmission or persistence [19,50,56,57]. This cost may be small in many bacteria, accounting for the slow response [55,58]. In contrast, a model indicates that, in hospitals, changes in resistance may be driven primarily by the admission of new patients who often bring with them drug-susceptible flora, and this may rapidly "dilute" levels of resistance in the absence of continuing selection by antibiotics [59]. If correct, this explanation suggests that the success of antimicrobial control measures should be evaluated differently for hospitals and for communities.

The use of mathematical models, and more generally the attempt to predict the relative merits of different interventions, will depend on an improved understanding of the mechanisms of antibiotic selection in particular organisms. For example, two recently published models for the nosocomial spread of resistant pathogens made contrasting assumptions about whether antimicrobial treatment increased an patient’s susceptibility to colonization only during treatment [60] or for a period following treatment [59], and about the importance of colonization with drug-susceptible strains in protecting against acquisition of resistant ones. As a result of these differences in assumptions, predictions differed in important ways: one model suggested that reduction of antibiotic use would be a comparatively poor intervention when endemic transmission is high and that resistant organisms could persist endemically even in the absence of input from admitted patients or antibiotic selection [60]. The other model predicted rapid declines in the level of resistance when use is reduced, and a more complicated relationship between the effectiveness of interventions and the level of transmission within the hospital [59]. Testable predictions will permit the evaluation of different models for particular settings and provide a basis for refining the assumptions of these models.


Comments on Medscape are moderated and should be professional in tone and on topic. You must declare any conflicts of interest related to your comments and responses. Please see our Commenting Guide for further information. We reserve the right to remove posts at our sole discretion.