Evaluation of the Basis and Performance of Simple Allometry for Predictions of Clearance and Oral Clearance in Humans and Improvement Potential with Added In Silico Predictions

Abstract

AO_SCPLOWBSTRACTC_SCPLOWO_ST_ABSBackgroundC_ST_ABSAllometry is traditionally used in drug development in order to extrapolate and predict pharmacokinetic (PK) parameter estimates, such as steady-state volume of distribution (Vss), clearance (CL), oral clearance (CL/F, where F is the oral bioavailability) in animal species to humans. Recent results show that in silico prediction methodology can improve and outperform laboratory data-based methods for predictions of Vss and F. The main objectives were to evaluate the simple allometry principle for CL and CL/F, how well simple allometry predicts CL and CL/F in humans, and whether a combination of simple allometry and in silico predictions can improve predictions of CL and CL/F in humans. MethodsThe literature was searched for CL and CL/F-data in animal species and man. Data from at least 2 species, and preferably 3 or 4, and humans, for each compound (only small drugs) were used for the evaluation. The software ANDROMEDA by Prosilico was used for in silico predictions. A rule-based approach, based on laboratory and in silico data, and 2D PK-Space, were used to localize and predict compounds with largest allometric errors. Results and DiscussionThe evaluation shows limited support for the theoretical basis and empirical evidence and applicability of simple allometry for the prediction of CL and CL/F. There are many deviations from the simple allometric relationship (such as relatively low liver weight and blood flow in humans and 140-fold underprediction to 5,800-fold overprediction), skewness (including general overprediction, cases where humans deviate, and relatively high CL and low F in rats) and limited interspecies relationships (R2=0.07-0.19 for CL in animals vs humans). With 43 qualified compounds, R2 for CL and CL/F with simple allometry reached ca 1/3 (log scales). With the combination of simple allometry and in silico it was possible to improve the predictive performance. 31 to 124 % (average 68 %) improvement were found for R2, <2-fold error and median error, whereas maximum errors were reduced to 1/13 to 1/6. In silico predictions alone were even better - 19 to 176 % (average 96 %) and 1/157 to 1/25, respectively. 3 and 4 different PK-rules for large allometric prediction errors were found for CL and CL/F, respectively. These also had distinct positioning in 2D PK-Space. With in silico predictions, errors for the compounds with largest allometric prediction errors were decreased by 7-fold for CL and 4-fold for CL/F (median). ConclusionSimple allometry is associated with limited theoretical and empirical support for predictions of CL and CL/F in humans, and can be clearly improved when combined with (or replaced by) new in silico prediction methodology, rules and 2D PK-Space positioning. This is in line with the ambition to reduce and replace animal testing in drug development and need for methodological improvement.