Antimicrobial Drug Development Common Practices in PK-PD Model Selection and Common Misconceptions
Parviz Ghahramani, Tatiana Khariton, Joannellyn Chiu
Inncelerex, Jersey City, NJ, USA
Introduction: For antimicrobial drug development, relationship between efficacy and indices of exposure (e.g. AUC, Cmax ,%Time above MIC) are often examined to select the most relevant driver of efficacy. Coefficient of determination (R2) is often used for evaluation of target attainment and determination of PK driver of antimicrobial efficacy [1]. This abstract presents current extent of application and limitations of the approach.
Objectives: To assess the extent to which R2 is used as a model selection criterion in the development of antimicrobial agents, and to provide recommendations for alternative methods when selecting PK drivers of antimicrobial efficacy.
Methods: Published literature between 2006-2016 was searched using a range of terms associated with R2 and PK indices in the Journal of Antimicrobial Agents and Chemotherapy and the Journal of Antimicrobial Chemotherapy. Each publication was scrutinized for methodology used to evaluate the PK-PD relationship. Simulations were also conducted to compare performance of R2 vs. Residual Standard Error (RSE) statistics and to examine the effect of: 1) non-linearity, 2) variability.
Results: There were a total of 15313 published articles (2006-2016) in the two journals. Of these, 63 met the search criteria of which, 26 publications were not relevant for the objectives of this work, the remaining 37 publications all used R2 as the statistics to choose among non-linear models and select the best PK parameter as the predictor of antimicrobial efficacy. In addition, none of the articles attempted combination of PK parameters as a predictor of antimicrobial efficacy. With increasing non-linearity for a given variability, R2 may provide incorrect parameter selection (>25% of cases), whereas RSE statistics is not affected by non-linearity.
Conclusions: R2 is widely used in selection of PK parameters driving efficacy of antimicrobial agents regardless of the nature of the relationship [1]. Majority of articles apply R2 to non-linear models which is associated with major inadequacies and may result in erroneous model/parameter selection. Other statistics such as RSE provide more robust method in the context of antimicrobial PK-PD. In addition, current practices assess mainly one PK index as predictor of efficacy and no attempt is made to examine predictive value of the combined parameters. An example of application of appropriate alternative methodology using RSE and combination of parameters is presented.
References:
[1] Ambrose PG, Bhavnani SM, Rubino CM, Louie A, Gumbo T, Forrest A, Drusano GL. Pharmacokinetics-Pharmacodynamics of Antimicrobial Therapy: It's Not Just for Mice Anymore. Clinical Infectious Diseases 2007, 44(1):79-86.