Nonparametric meta-analysis of published data on kidney-function dependence of pharmacokinetic parameters for the aminoglycoside netilmicin

Abstract

The distribution and elimination of various drugs depend on kidney function. This dependence is published either as a linear regression equation or as the discrete extreme values for normal kidney function and anuria. A meta-analysis of the published pharmacokinetic data is required to build up a knowledge-based computer system for dosage adjustment in renal failure. A sample comparison of 4 statistical methods for meta-analysis was performed by applying them to 13 publications about the aminoglycoside netilmicin. Parametric meta-analytical methods I and II are based on regression equations alone (Z-transformation, maximum likelihood) and yield unreliable data, especially with regard to extreme values for anuria. The parametric meta-analytical method III is based on means of extreme values (standard 2-stage approach) and does not permit a decision as to whether linear interpolation of a parameter (e.g. volume of distribution) can be used for all degrees of renal insufficiency. In contrast, the nonparametric median (meta-analytical method IV) is based on the extreme values calculated from regression equations and empirical extreme values combined into 1 group of data on normal kidney function and another on anuria. For netilmicin, the meta-analytical median with the 95% confidence interval (95% CI) yields a significant increase in the dominant elimination half-life from 2h (95% CI 1.9h, 2.6h) in patients with normal kidney function to 45h (95% CI 41h, 301h) in those with anuria (p = 0.001). For a normal bodyweight of 65kg, the volume of distribution also increases significantly from 13L (95% CI 9L, 15L) to 20L (95% CI 14L, 21L) in patients with anuria (p = 0.04). Thus, drug dosage adjustment according to therapeutic peak and trough concentrations requires knowledge of the distribution and elimination parameters, since they can both be independently altered in renal failure. We conclude that the most robust meta-analysis of these alterations is achieved with the nonparametric median of extreme values.