基于人时的相对危险度分布的Monte-Carlo模拟估计

目的应用Monte-Carlo模拟进行基于人时的相对危险度的分布估计。方法结合实例进行相对危险度的模型构建、拉丁超立方抽样和概率分布的拟合及RR可信区间的几种计算方法比较。结果模拟的RR频率分布经拟合符合Pearson5、Lognorm、Gamma和InvGauss4种分布,以Pearson5分布拟合最佳。模拟的RR值95%可信区间结果与统计量函数计算值、Wald法和Score法大致相当,但其上限值和下限值均略小。结论应用Monte-Carlo模拟结合拉丁超立方抽样技术,实现了基于人时的相对危险度的分布估计,该方法可应用于更为复杂的参数分布估计。

Monte-Carlo simulation estimation of rate ratio based on person year

Objective To estimate the distribution of person-year based rate ratio by using Monte-Carlo simulation. Methods According to the example given, rate ratio (RR) model was constructed and simulated with Latin Hypercube Sampling. The sampled frequency distribution was fitted and 95% confidence interval of RR was calculated with several methods. Results The simulated RR frequency distribution was fitted good among Pearson 5, Lognorm, Gamma and InvGauss distribution with Pearson 5 best. 95% confidence interval of RR was similar among the different methods, but both interval of simulated RR were more narrower. Conclusions The method of combining Monte-Carlo simulation with Latin Hypercube Sampling could be used to estimate the distribution of the rate ratio based on person-year. It was expected to extend to much more complicated parameter estimation.