A comparison of five approaches for lithium dose and serum concentration prediction

Various methodologies are proposed in the literature to predict lithium dose or serum concentration. The authors compare the performance of a fuzzy system modeling algorithm that has been proposed with four other algorithms in terms of their performance based on serum lithium concentration prediction. The first method is the "Zetin method" proposed by M. Zetin et al. (1986), which is based on stepwise multiple linear regressions and designed specifically for lithium pharmacokinetics. Secondly, the more recent method proposed by T. Terao et al. (1999). The third method is a new formula developed from the data on hand by using stepwise multiple linear regressions. Fourth, a comparison is made with the well-known fuzzy system modeling algorithm proposed by M. Sugeno and T.A. Yasukawa (1993). The proposed method is the fifth alternative to be considered in this comparison. Published data from 30 patients (T. Terao et al. 1999) were used in the analysis. The performance of the algorithms with respect to precision as measured by the root mean square error are as follows: 0.54, 0.34, 0.36, 0.31 and 0.24 mmol/L, respectively for the Zetin method, Terao method, stepwise multiple linear regression, Sugeno-Yasukawa approach and the new algorithm.