Discussion on Within-cycle Correction of Markov Model in Pharmacoeconomic Evaluation / 中国药房

ABSTRACT

OBJECTIVE：To pro vide reference for reducing the error of Markov model in pharmacoeconomic evaluation. METHODS：Referring to foreign literatures ，the errors in Markov model were explained. The commonly used correction methods were introduced ：half-cycle correction ，trapezoidal rule ，Simpson’s 1/3 rule，Simpson’s 3/8 method and life table method and their implementation in Excel and TreeAge software. RESULTS & CONCLUSIONS ：The error of Markov model was caused by the discretization process and could be corrected by the within-cycle correction methods. Half-cycle correction ，was the most commonly used correction method and could be corrected by adding half of the results of the first cycle and subtracting half of the results of the last cycle. By trapezoidal rule ，the interval was represented by the mean value of the first end point value of the interval ，and the area of right trapezoid was the cummulative result. Simpson’s 1/3 rule and Simpson ’s 3/8 rule took another point in the interval on the basis of trapezoidal correction and the continuous curve where these three points were represented the whole curve. By life table method ，the person-years of survival was the product of the group distance of the age group and the mean value of the number of survivors of the age group and the next age group ，and the total person-years of survival was the sum of person-years of survival of each age group. In Excel ，the fuction was set according to the method principle to realize the correction ；in Tree Age software，the function expression of Init Rwd ，Inor Rwd and Final Rwd were set to realize the correction. When using Markov model for pharmacoeconomic evaluation ，if the selection of periodic correction method is based on the ease of modeling and the universality of application ，the trapezoid rule method is recommended ；based on the accuracy of results ，it is suggested to use Simpson’s 1/3 method，so as to improve the accuracy of Markov model.