Antibiotic Dosing—Does One Size Fit All?

In this issue of THE JOURNAL, Preston and colleagues¹ argue that pharmacokinetic and pharmacodynamic modeling applied prospectively to a clinical trial of a fluoroquinolone antibiotic aids in understanding the efficacy and safety of therapy. The investigators used modeling to link bacterial eradication to the individual minimum inhibitory concentration (MIC) values of the bacteria, and the individual serum concentrations (ie, peak concentration or the area under the curve [AUC]) of the antibiotic in the treated patient. The article presents considerable evidence that the prospective use of these modeling methods in the course of a multicenter antibiotic trial can identify a dose that will ensure efficacy and can predict when a regimen will not work or will select resistant subpopulations of organisms.

The modeling methods and techniques appear complex, and their output consists of complicated graphs and charts. There are few practitioners of this craft, and finding experts to answer basic questions can be difficult. While experts argue incessantly over simple decisions such as which pharmacokinetic parameters are best to use, there is far more agreement than disagreement in this discipline. The methods and parameters are being standardized, and the majority of studies point to predictability among parameters and models in vitro, in animal models, and, most important, in patients.^{1 - 6}

The prospective use of these methods in a clinical trial necessitates that study investigators collect blood samples, identify the causative organism, and measure its MIC. Fortunately, large dividends arise from the additional work. Chief among them is the ability to characterize fully the antibiotic response in a relatively small number of patients. This is pivotal if the correct dose is to be chosen. The study by Preston et al¹ shows that a posttherapy culture of the infection site will reliably link the blood levels and the MICs with eradication of the organism. Serial cultures of the infection site would have determined the speed of organism eradication² but were not collected in this trial. Definitive dosing guidelines derived from a population of fewer than 120 patients are a major dividend, particularly when these guidelines can be determined early in the course of new antibiotic development. In fact, considerable evidence suggests that common parameters, such as the AUC/MIC ratio, can predict the outcome of many clinical situations with a wide variety of antibiotics.^{2 - 6}

The success of the modeling technique used by Preston et al rests on the revival of several straightforward principles, some of which have been neglected in clinical trials of antibiotic therapy. Chief among them is the consideration of bacterial eradication as the end point of antibiotic action in vivo. Antibiotics do not cure disease, but under proper conditions of pharmacokinetics optimized to MIC, they eradicate bacteria. Returning to bacterial eradication as an end point of antibiotic efficacy does not take away from the importance of end points such as clinical cure, but does argue that doses of antibiotics can be adjusted to eliminate bacteria in a predictable manner.

The antibiotic development industry has drifted toward reliance on clinical cure as an end point of antibiotic therapy, in line with clinical practice guidelines that also emphasize cure. Clinical cure is a more convincing end point for the clinician, but a far more difficult end point to use in a model of antibiotic activity against bacteria. Whether the patient is cured depends on 2 successive processes, bacterial eradication and disease resolution. Disease resolution alone, especially in the absence of a search for pathogens, may be less closely linked to the action of the antibiotic or its relationship to pharmacokinetics than might be anticipated. Studies of infections that demonstrate close links between bacterial eradication and disease resolution, such as nosocomial pneumonia or bacteremia, reveal close agreement between eradication and clinical cure.^{2 ,4} But what should be done about conditions like otitis media, for which there is little opportunity to obtain cultures, and a 60% clinical cure rate can be achieved without antibiotics?^{3 ,7} When conditions have been subjected to modeling exercises,³ the occasional treatment failure can also be predicted.

The speed of bacterial killing also is becoming more useful as an end point for clinical studies. Animal studies have shown that some antibiotics kill bacteria faster than others,⁸ but only recently has faster bacterial killing been shown in humans, and with some antibiotics more rapidly than others.⁴ Serial and even daily culturing² is the key breakthrough, and in this model, the patient sample collection techniques are following the time-killing study model that has been used to assess antibiotic actions in vitro for more than 40 years.⁹ It was encouraging to see that in the study by Preston et al, 1 model predicted the results in the other. The agreement between in vitro models of time killing and serial cultures in the patient is only possible when the data of each of these situations are converted to a relationship between MIC and exposure concentration.

Most clinicians readily accept the principle that blood levels must exceed the MIC of the pathogen, but few seem to appreciate how often there is only a marginal amount of time during which this occurs in patients. Even fewer clinicians directly connect this marginal coverage of the MIC to failure to eradicate the pathogen,² and fewer still connect marginal coverage (ie, dosing) to the development of bacterial resistance.⁵ Yet, this phenomenon has been observed using the in vitro models of antibiotic action.¹⁰ The problem is one of natural variability in the determinants of antibacterial outcomes. The MIC can vary by more than 1000-fold even between susceptible organisms of the same species,¹¹ and a wide range of pharmacokinetic variability exists between patients. The pharmacokinetics of a new antibiotic are well characterized in normal volunteers, but only recently have actual patient pharmacokinetics been directly identified as an explanation for variations in bacterial eradication in patients with active infection.^{1 - 2 ,4 - 5} The failure to eradicate pathogens emerges as a mismatch between the pharmacokinetics in the patient and the MIC of the pathogen. Patient factors certainly play a role, but appear to be less important than the role of the antibiotic AUC and the MIC of the pathogen under treatment.^{1 - 2 ,4 - 5}

The variability that occurs in both pharmacokinetics and in MIC in a typical fixed-dose antibiotic clinical trial usually works directly against the attempts to find the correct dose. For instance, each of the various disease groups may have some patients who harbor organisms with relatively high or very low MICs, and each group may have some patients with relatively high or low drug concentrations. In a very large study, the groups may yield no clear evidence of a dose response, and the trial may be deemed "uninformative." In contrast, with the improvements that modeling can make in the same trial if MIC is known, dose is converted to individual patient pharmacokinetics, and these 2 determinants are modeled as a composite parameter (such as AUC/MIC or peak/MIC ratios) vs bacterial killing. Accordingly, even a much smaller trial than that of Preston et al¹ can be much more informative as to dose that works, because the modeling exercise identifies the occasional failures as a predictable result of higher than normal MIC, lower than expected blood levels, or both. This outcome is even more frequent in the clinical setting than in clinical trials, and it is equally predictable if clinicians could be convinced to look.

Although the mathematics of these modeling techniques can be formidable, the approach is amenable to computer programs that store the equations for AUC and estimate the AUC/MIC ratios for most of the antibiotics. Assisted by bedside or palmtop computer software, the process may prove to be no more complex than the routine dose calculation for aminoglycosides or vancomycin. If subsequent studies demonstrate that doses targeted to organism MICs minimize treatment failure and reduce resistance, health care institutions most likely will consider the adoption of these methods for their cost-saving value.

All this takes advantage of the considerable, but apparently predictable, variability between patients who seemingly receive the same antibiotic dose but have different outcomes. Presumably, either their pharmacokinetics are different or their organisms are less susceptible. The current alternative, at least in the world of antibiotic development, is to overwhelm the variability with large numbers of patients and a crude end point, such as clinical cure. The article by Preston et al¹ provides more evidence of the ability to take advantage of this built-in variability to improve the information content of even rather small, prospective clinical trials. The best use of these modeling techniques is to explain the exceptions (ie, failures or resistance) to the "one dose for all" paradigm.

In the antibiotic resistance–laden late 1990s, it is becoming ever more important to attend to factors that can control treatment failure and limit antibiotic resistance. Improvements in clinical expertise arise from understanding the exceptions and being able to anticipate and avoid them from readily available data and observations. Since few miracle antibiotics are on the immediate horizon and patients are no less ill, the only available method at the moment appears to be accurate dosing, or "one dose for each."⁶ With antibiotic therapy, the data say it is time to discuss clinical implementation.