Abstract
Lithium has been used in the treatment of bipolar disorder for several decades. Treatment optimization is recommended for this drug, due to its narrow therapeutic range and a large pharmacokinetics (PK) variability. In addition to therapeutic drug monitoring, attempts have been made to predict individual lithium doses using population pharmacokinetics (popPK) models. This study aims to assess the clinical applicability of published lithium popPK models by testing their predictive performance on two different external datasets. Available PopPK models were identified and their predictive performance was determined using a clinical dataset (46 patients/samples) and the literature dataset (89 patients/samples). The median prediction error (PE) and median absolute PE were used to assess bias and inaccuracy. The potential factors influencing model predictability were also investigated, and the results of both external evaluations compared. Only one model met the acceptability criteria for both datasets. Overall, there was a lack of predictability of models; median PE and median absolute PE, respectively, ranged from -6.6% to 111.2% and from 24.4% to 111.2% for the literature dataset, and from -4.5% to 137.6% and from 24.9% to 137.6% for the clinical dataset. Most models underpredicted the observed concentrations (7 out of 10 models presented a negative bias). Renal status was included as a covariate of lithium's clearance in only two models. To conclude, most of lithium's PopPK models had limited predictive performances related to the absence of covariates of interest included, such as renal status. A solution to this problem could be to improve the models with methodologies such as metamodeling. This could be useful in the perspective of model-informed precision dosing.