基于数据挖掘的预警方法在双黄连注射剂不良反应监测中的应用研究

Research on signal detection methods based on data mining in adverse drug reaction of Shuanghuanglian injection

目的:采用信号监测方法,结合江苏省药物自发报告系统中成药不良反应数据,探讨双黄连注射剂的不良反应表现特点,并对3种方法进行评价。方法:基于江苏省不良反应自发报告数据,采用3种方法分别进行预警,通过Kappa检验考察方法间预警结果,及各种方法与说明书提及不良反应的一致性。通过预警统计量可信区间在时间上的趋势,探索信号变化趋势。结果:在任何情况下,本研究涉及的2种传统频数法(PRR法及ROR法)一致性程度非常高。在目标不良反应报告数较多时,BCPNN法与传统频数法预警结果一致程度较高。在报告数较小时(小于3例),PRR,ROR估计误差太大,导致假阳性高于BCPNN法。较自频数法,基于BCPNN法的可信区间趋势图能够更有效地展示信号趋势。结论:在基于自发报告系统的中成药不良反应信号预警研究中,如果当前研究的药物-不良反应所涉及的报告数目较多,则采用传统频数法;当探索预警信号的时间趋势时,选用BCPNN法为佳。

Objective： This paper is aimed to explore the adverse reaction condition of Shuanghuangli an injection with three common used signal detecting methods based on SRS database of Jiangsu province, and to evaluate the performance of three methods. Method ： Three methods would be used to detect the signals based on the SRS database of Jiangsu province. Consistency of the results of these three methods with that proved in descriptions was evaluated by Kappa test. The trend graph of the confidence intervals of several time points was used to demonstrate the trend of the signal. Result： The PRR method was consistent with ROR method in high degree in any situation. The results of BCPNN method was close to PRR and ROR method only when the related report count was larger. PRR and ROR methods had higher false positive rate than BCPNN method. Conclusion： PRR or ROR method is proposed for signal detecting when the report count is large. BCPNN method is proposed for trend demonstration of signal with graph.