Comparison of three a-priori models in the prediction of serum lithium concentration

Abstract

Context:Mathematical models are valuable for optimizing drug dose and dosing regimens.Aims:To compare the precision and bias of three a-priori methods in the prediction of serum level of lithium in patients with bipolar disorder, and to determine their sensitivity and specificity in detecting serum lithium levels outside the therapeutic range.Settings and Design:Hospital-based, retrospective study.Materials and Methods:In a retrospective study of 31 in-patients, the serum level of lithium was calculated using three different a-priori methods. Mean Prediction Error was used as a measure of bias while Mean Absolute Error and Root Mean Squared Error were used as a measure of precision. The sensitivity and specificity of the methods was calculated.Results:All three models underestimated serum lithium level. Precision was best with the model described by Pepin et al., while bias of prediction was the least with the method of Abou Auda et al. The formula by Pepin et al. was able to predict serum lithium level with a mean error of 36.57%. The sensitivity and specificity of the models in identifying serum lithium levels outside the therapeutic range was 80% and 76.19% for Pepin et al., 90% and 74.19% for Zetin et al., and 90% and 66.67% for Abou-Auda et al., respectively.Conclusion:The study demonstrates the difference in precision and bias of three a-priori methods, with no one method being superior to the other in the prediction of serum concentration.

Prediction Model of Serum Lithium Concentrations

“…To address these issues, several strategies for predicting serum lithium concentrations from oral doses, which are classified into 2 categories (i. e., a prior predictive method and test-dose method), have been proposed [5]. However, the root mean squared prediction error, which is a proxy for accuracy of prediction with a higher value indicating lower prediction performance, was reported to be as high as 0.37, 0.59, and 0.57 mmol/L in the prediction models by Pepin et al [6], Zetin et al [7], and Abou Auda et al [8], respectively [9]. This insufficient predictive power of those prediction models hampers the use of them in routine clinical practice.…”

Lithium Dosing Equations

“…5 The Pepin et al 6 and Jermain et al 12 equations make multiple assumptions regarding clearance, elimination halflife, and volume of distribution. [5][6][7][8][9] These assumptions may increase the potential for error in each equation.…”

“…Pepin and colleagues initially developed an equation based on first-order, one-compartment, patient-specific pharmacokinetic parameters that have been evaluated in several small studies. [5][6][7][8][9] The complex computation required by this equation may limit its utility. Zetin et al 2 similarly developed an equation based on patient-specific parameters using multiple linear regression analysis for derivation 2,5,[7][8][9][10][11] ; however, the authors potentially omitted an influential variable.…”

Calculation of Lithium Clearance for Clinical use Based on Renal Excretion in Japanese Patients

“…In addition, they reported that the calculation of Li-CL and estimation of lithium dosage using their method was more reliable compared to the Jermain, Pepin and Terao methods [19]. Moreover, Radhakrishnan et al compared the Pepin, Jermain and Abou-Auda methods for Css prediction, and reported that the Pepin method was the most precise, where as the Abou-Auda method was the least biased [20].…”

“…However, this consideration may be negligible in the population of the present study, due to good adherence to therapy schedules as well as generally stable condition of their disease. [6,14,[20][21][22], standardized prediction methods using Li-CL and Css remain to be elucidated.…”