NONPARAMETRIC METAANALYSIS OF PUBLISHED DATA ON KIDNEY-FUNCTION DEPENDENCE OF PHARMACOKINETIC PARAMETERS FOR THE AMINOGLYCOSIDE NETILMICIN

The distribution and elimination of various drugs depend on kidney fun ction. This dependence is published either as a linear regression equa tion or as the discrete extreme values for normal kidney function and anuria. A meta-analysis of the published pharmacokinetic data is requi red to build up a knowledge-based computer system for dosage adjustmen t in renal failure. A sample comparison of 4 statistical methods for m eta-analysis was performed by applying them to 13 publications about t he 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 extr eme values for anuria. The parametric meta-analytical method III is ba sed on means of extreme values (standard 2-stage approach) and does no t permit a decision as to whether linear interpolation of a parameter (e.g. volume of distribution) can be used for all degrees of renal ins ufficiency. In contrast, the nonparametric median (meta-analytical met hod IV) is based on the extreme values calculated from regression equa tions and empirical extreme values combined into 1 group of data on no rmal kidney function and another on anuria. For netilmicin, the meta-a nalytical median with the 95% confidence interval (95% CI) yields a si gnificant 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 o f 65kg, the volume of distribution also increases significantly from 1 3L (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 e limination parameters, since they can both be independently altered in renal failure. We conclude that the most robust meta-analysis of thes e alterations is achieved with the nonparametric median of extreme val ues.