贝叶斯法与多元线性回归法预测华法林剂量模型的比较研究

目的：比较缺失华法林相关基因型检测结果的情况下，基于贝叶斯法的Warfarin Dose Calculator（WDC）和基于多元线性回归法（MLR）的Warfarin Dosing、IWPC华法林剂量计算器等三种华法林剂量预测辅助工具的预测性能，为无条件进行华法林相关基因型检测地区的患者使用华法林提供准确、适于临床推广的预测工具。方法：计算三种预测工具的平均预测误差（MPE）、均方根误差（RMSE）和预测误差在实际维持剂量±20%范围内的患者百分比，比较它们的华法林剂量预测性能。所有统计学分析均采用SPSS Statistics 26.0软件进行。结果：WDC的Posteriori模块的MPE和RMSE最低且预测误差在实际维持剂量±20%范围内的患者百分比最高，此外WDC的Priori和Posteriori模块的预测精确度显著高于Warfarin Dosing和IWPC华法林剂量计算器（P<0.05）。结论：临床常用的三种华法林剂量预测辅助工具中，基于贝叶斯法的Warfarin Dose Calculator可能更适合于无条件进行华法林相关基因型检测地区患者的华法林剂量预测。

Comparison of Bayesian method and multiple linear regression method in predicting warfarin dose model

AIM: To evaluate the predictive performance of Warfarin Dose Calculator (WDC) based on Bayesian method, Warfarin Dosing and International Warfarin Pharmacogenetics Consortium (IWPC) Warfarin Dose Calculator based on Multiple Linear Regression (MLR) in the absence of warfarin related genotypes, in order to provide help for patients using warfarin in areas where there is unconditional warfarin related genotypes detection. METHODS: The predicted performance of each tool was evaluated by calculating the mean prediction error (MPE), root mean square error (RMSE) and the percentage of patients whose prediction error was within ±20% of the actual maintenance dose. All statistical analyses were performed by SPSS statistics 26.0 software. RESULTS: The Posteriori of the WDC had the lowest MPE, RMSE and highest percentage of patients whose prediction errors within ±20% of the actual maintenance dose. In addition, the predictive accuracy of the Priori and Posteriori of WDC is significantly higher than that of Warfarin Dosing and IWPC warfarin dose calculator. CONCLUSION: Among the three assistant tools for warfarin dose prediction, Warfarin Dose Calculator based on Bayesian method may be more suitable for warfarin dose prediction of patients in unconditional warfarin related genotypes detection area.