我们引入新的估计量的优势比稀有事件使用经验贝叶斯方法在两个独立的二项分布。我们比较优势比的建议预算有两个估计,修正极大似然估计量(MMLE)和修改值无偏估计量(MMUE),使用估计的相对误差(之前)作为比较的标准。发现新的估计量与其他方法相比更有效率。
的比率衡量的是两个独立的组分类响应之间的联系有两个可能的结果,成功和失败。两个独立的团体可以两个治疗组或治疗和控制。的优势比广泛用于医学和社会科学研究的许多领域。在流行病学最常用来表达一些临床试验的结果,如在病例对照研究。
每组的受试者人数下跌在每个类别可以概括在一个双向列联表。总数1组和2组的受试者<我nline-formula>
在本文中,我们专注于“小概率事件”,偶尔观察到零个或小项有趣的事件发生在一个给定的时期或一个给定的样本,如自然灾害或一些疾病。正如上面提到的,罕见的事件造成难以估计的优势比0的发生或小分子或分母观察数量或在两者中,导致大的标准误差,因此较少的精确置信区间。因此只有粗略估计的优势比。研究涉及关联分类变量列联表一直研究,使用经典和贝叶斯方法。好(
正如前面提到的,估计协会双向列联表的测量可以进行基于古典和贝叶斯方法。确切的分布使用经典的方法,然而,相当困难的数学温顺。在贝叶斯方法,之前的信念是纳入推导的后验密度,hyperparameters,描述前密度,研究人员往往是未知的,需要评估无论当前数据。然而,争议仍然存在。另外,hyperparameters的估算是进行合理的使用当前数据与经验贝叶斯方法的概念来估计未知的hyperparameters,与贝叶斯方法。因此,我们专注于利用经验贝叶斯估计方法的优势比双向列联表,关注小成功的比例。我们计划的评估往往比传统的估计量,MMLE, MMUE没有干涉原始数据。
本文的其余部分被组织在以下序列。在下一节中,我们将讨论中值的无偏估计量。第三部分描述了使用EB的优势比估计的方法。第四部分说明了模拟结果和EB的效率相比MMLE和MUE。第五部分显示我们的方法应用到真实的数据。我们在最后一部分得出结论。
Parzen et al。
计算的值<我nline-formula>
让<我nline-formula>
现在假设<我nline-formula>
同样的,当<我nline-formula>
然后,MMUE优势比估计的定义是
在本节中,我们提出了一种新的方法优势比使用经验贝叶斯方法估计在两个独立的二项分布。让<我nline-formula>
然后,hyperparameters每组可以使用最大似然估计方法。后边际似然函数的分布函数然后写成
的后验分布函数<我nline-formula>
模拟研究进行了使用R程序(3.2.0版)(
仿真结果与优势比样本大小的估计<我nline-formula>
估计的值的比值比<我nline-formula>
|
|
|
|
|
|
|---|---|---|---|---|
| (0.01,0.01) | 1.0000 | 1.3665 | 1.1514 | 1.2043 |
| (0.01,0.03) | 0.3266 | 0.3931 | 1.0029 | 1.0377 |
| (0.01,0.05) | 0.1919 | 0.2248 | 0.8723 | 0.8935 |
| (0.01,0.10) | 0.0909 | 0.1040 | 0.6219 | 0.6184 |
| (0.01,0.15) | 0.0572 | 0.0650 | 0.4481 | 0.4307 |
| (0.03,0.01) | 3.0619 | 3.8746 | 1.6008 | 1.7848 |
| (0.03,0.03) | 1.0000 | 1.1119 | 1.3933 | 1.5383 |
| (0.03,0.05) | 0.5876 | 0.6363 | 1.2128 | 1.3244 |
| (0.03,0.10) | 0.2784 | 0.2942 | 0.8640 | 0.9156 |
| (0.03,0.15) | 0.1753 | 0.1838 | 0.6227 | 0.6378 |
| (0.05,0.01) | 5.2105 | 6.5657 | 2.0724 | 2.4094 |
| (0.05,0.03) | 1.7018 | 1.8851 | 1.8036 | 2.0763 |
| (0.05,0.05) | 1.0000 | 1.0787 | 1.5693 | 1.7868 |
| (0.05,0.10) | 0.4737 | 0.4989 | 1.1181 | 1.2356 |
| (0.05,0.15) | 0.2982 | 0.3116 | 0.8059 | 0.8609 |
| (0.10,0.01) | 11.0000 | 13.7434 | 3.3472 | 4.1489 |
| (0.10,0.03) | 3.5926 | 3.9471 | 2.9135 | 3.5759 |
| (0.10,0.05) | 2.1111 | 2.2585 | 2.5352 | 3.0777 |
| (0.10,0.10) | 1.0000 | 1.0445 | 1.8068 | 2.1288 |
| (0.10,0.15) | 0.6296 | 0.6523 | 1.3027 | 1.4839 |
| (0.15,0.01) | 17.4706 | 21.7299 | 4.7827 | 6.1533 |
| (0.15,0.03) | 5.7059 | 6.2349 | 4.1625 | 5.3026 |
| (0.15,0.05) | 3.3529 | 3.5678 | 3.6225 | 4.5648 |
| (0.15,0.10) | 1.5882 | 1.6498 | 2.5812 | 3.1568 |
| (0.15,0.15) | 1.0000 | 1.0303 | 1.8602 | 2.1990 |
估计的值的比值比<我nline-formula>
|
|
|
|
|
|
|---|---|---|---|---|
| (0.01,0.01) | 1.0000 | 1.3656 | 2.9352 | 2.7925 |
| (0.01,0.03) | 0.3266 | 0.4023 | 2.0063 | 1.8490 |
| (0.01,0.05) | 0.1919 | 0.2279 | 1.4146 | 1.2576 |
| (0.01,0.10) | 0.0909 | 0.4043 | 0.6802 | 0.5486 |
| (0.01,0.15) | 0.0572 | 0.0652 | 0.3981 | 0.2971 |
| (0.03,0.01) | 3.0619 | 3.8640 | 4.0816 | 4.1395 |
| (0.03,0.03) | 1.0000 | 1.1385 | 2.7872 | 2.7376 |
| (0.03,0.05) | 0.5876 | 0.6452 | 1.9663 | 1.8632 |
| (0.03,0.10) | 0.2784 | 1.1466 | 0.9457 | 0.8131 |
| (0.03,0.15) | 0.1753 | 0.1845 | 0.5536 | 0.4406 |
| (0.05,0.01) | 5.2105 | 6.5508 | 5.2833 | 5.5870 |
| (0.05,0.03) | 1.7018 | 1.9307 | 3.6077 | 3.6950 |
| (0.05,0.05) | 1.0000 | 1.0940 | 2.5446 | 2.5144 |
| (0.05,0.10) | 0.4737 | 1.9430 | 1.2236 | 1.0969 |
| (0.05,0.15) | 0.2982 | 0.3128 | 0.7163 | 0.5944 |
| (0.10,0.01) | 11.0000 | 13.7159 | 8.5346 | 9.6223 |
| (0.10,0.03) | 3.5926 | 4.0427 | 5.8289 | 6.3651 |
| (0.10,0.05) | 2.1111 | 2.2907 | 4.1132 | 4.3339 |
| (0.10,0.10) | 1.0000 | 1.0327 | 1.6556 | 1.5221 |
| (0.10,0.15) | 0.6296 | 0.6549 | 1.1578 | 1.0247 |
| (0.15,0.01) | 17.4706 | 21.6662 | 12.1932 | 14.2687 |
| (0.15,0.03) | 5.7059 | 6.3850 | 8.3278 | 9.4395 |
| (0.15,0.05) | 3.3529 | 3.6181 | 5.8732 | 6.4225 |
| (0.15,0.10) | 1.5882 | 6.4257 | 2.8237 | 2.8019 |
| (0.15,0.15) | 1.0000 | 1.0345 | 1.6529 | 1.5181 |
估计的值的比值比<我nline-formula>
|
|
|
|
|
|
|---|---|---|---|---|
| (0.01,0.01) | 1.0000 | 1.5096 | 4.2803 | 3.9580 |
| (0.01,0.03) | 0.3266 | 0.4210 | 2.3790 | 2.0799 |
| (0.01,0.05) | 0.1919 | 0.6975 | 1.4513 | 1.1915 |
| (0.01,0.10) | 0.0909 | 0.1057 | 0.6167 | 0.4475 |
| (0.01,0.15) | 0.0572 | 0.0656 | 0.3656 | 0.2515 |
| (0.03,0.01) | 3.0619 | 4.2760 | 5.9515 | 5.8667 |
| (0.03,0.03) | 1.0000 | 1.1926 | 3.3063 | 3.0810 |
| (0.03,0.05) | 0.5876 | 1.9756 | 2.0173 | 1.7653 |
| (0.03,0.10) | 0.2784 | 0.2992 | 0.8578 | 0.6636 |
| (0.03,0.15) | 0.1753 | 0.1856 | 0.5083 | 0.3728 |
| (0.05,0.01) | 5.2105 | 7.2486 | 7.7012 | 7.9149 |
| (0.05,0.03) | 1.7018 | 2.0224 | 4.2796 | 4.1586 |
| (0.05,0.05) | 1.0000 | 3.3494 | 2.6114 | 2.3833 |
| (0.05,0.10) | 0.4737 | 0.5073 | 1.1100 | 0.8954 |
| (0.05,0.15) | 0.2982 | 0.3147 | 0.6579 | 0.5031 |
| (0.10,0.01) | 11.0000 | 15.1780 | 12.4415 | 13.6337 |
| (0.10,0.03) | 3.5926 | 4.2347 | 6.9165 | 7.1666 |
| (0.10,0.05) | 2.1111 | 7.0132 | 4.2213 | 4.1087 |
| (0.10,0.10) | 1.0000 | 1.0621 | 1.7938 | 1.5433 |
| (0.10,0.15) | 0.6296 | 0.6589 | 1.0631 | 0.8669 |
| (0.15,0.01) | 17.4706 | 23.9776 | 17.7763 | 20.2195 |
| (0.15,0.03) | 5.7059 | 6.6884 | 9.8777 | 10.6225 |
| (0.15,0.05) | 3.3529 | 11.0773 | 6.0256 | 6.0860 |
| (0.15,0.10) | 1.5882 | 1.6775 | 2.5614 | 2.2869 |
| (0.15,0.15) | 1.0000 | 1.0408 | 1.5181 | 1.2848 |
估计的相对误差百分比的比值比估计<我nline-formula>
|
|
|
|
|
|---|---|---|---|
| (0.01,0.01) | 36.6535 | 15.1385 | 20.4316 |
| (0.01,0.03) | 20.3505 | 206.7910 | 217.7388 |
| (0.01,0.05) | 17.1281 | 354.5204 | 365.6116 |
| (0.01,0.10) | 14.3630 | 584.0771 | 580.2372 |
| (0.01,0.15) | 13.4914 | 682.8616 | 652.4367 |
| (0.03,0.01) | 26.5436 | 47.7187 | 41.7082 |
| (0.03,0.03) | 11.1895 | 39.3332 | 53.8303 |
| (0.03,0.05) | 8.2767 | 106.3838 | 125.3783 |
| (0.03,0.10) | 5.6857 | 210.3971 | 228.9552 |
| (0.03,0.15) | 4.8556 | 255.2977 | 263.9451 |
| (0.05,0.01) | 26.0092 | 60.2273 | 53.7592 |
| (0.05,0.03) | 10.7763 | 5.9852 | 22.0098 |
| (0.05,0.05) | 7.8661 | 56.9330 | 78.6812 |
| (0.05,0.10) | 5.3167 | 136.0498 | 160.8427 |
| (0.05,0.15) | 4.4721 | 170.2257 | 188.6564 |
| (0.10,0.01) | 24.9404 | 69.5705 | 62.2831 |
| (0.10,0.03) | 9.8685 | 18.9032 | 0.4659 |
| (0.10,0.05) | 6.9818 | 20.0893 | 45.7877 |
| (0.10,0.10) | 4.4467 | 80.6791 | 112.8820 |
| (0.10,0.15) | 3.6016 | 106.9073 | 135.6854 |
| (0.15,0.01) | 24.3801 | 72.6244 | 64.7792 |
| (0.15,0.03) | 9.2720 | 27.0495 | 7.0675 |
| (0.15,0.05) | 6.4085 | 8.0392 | 36.1422 |
| (0.15,0.10) | 3.8773 | 62.5224 | 98.7588 |
| (0.15,0.15) | 3.0338 | 86.0169 | 119.9015 |
估计的相对误差百分比的比值比估计<我nline-formula>
|
|
|
|
|
|---|---|---|---|
| (0.01,0.01) | 36.5558 | 193.5166 | 179.2467 |
| (0.01,0.03) | 23.1721 | 514.2925 | 466.1248 |
| (0.01,0.05) | 18.7270 | 637.0581 | 555.2608 |
| (0.01,0.10) | 344.7753 | 648.1943 | 503.4379 |
| (0.01,0.15) | 13.8933 | 595.4561 | 419.0998 |
| (0.03,0.01) | 26.1981 | 33.3063 | 35.1969 |
| (0.03,0.03) | 13.8541 | 178.7227 | 173.7624 |
| (0.03,0.05) | 9.7978 | 234.6089 | 217.0772 |
| (0.03,0.10) | 311.9138 | 239.7364 | 192.0982 |
| (0.03,0.15) | 5.2578 | 215.8678 | 151.3923 |
| (0.05,0.01) | 25.7219 | 1.3963 | 7.2250 |
| (0.05,0.03) | 13.4545 | 112.0004 | 117.1259 |
| (0.05,0.05) | 9.3998 | 154.4641 | 151.4381 |
| (0.05,0.10) | 310.1875 | 158.3124 | 131.5756 |
| (0.05,0.15) | 4.8844 | 140.1677 | 99.2831 |
| (0.10,0.01) | 24.6897 | 22.4126 | 12.5246 |
| (0.10,0.03) | 12.5282 | 62.2482 | 77.1730 |
| (0.10,0.05) | 8.5055 | 94.8364 | 105.2921 |
| (0.10,0.10) | 3.2693 | 65.5584 | 52.2104 |
| (0.10,0.15) | 4.0198 | 83.8863 | 62.7515 |
| (0.15,0.01) | 24.0150 | 30.2071 | 18.3274 |
| (0.15,0.03) | 11.9013 | 45.9518 | 65.4338 |
| (0.15,0.05) | 7.9089 | 75.1670 | 91.5471 |
| (0.15,0.10) | 304.5808 | 77.7868 | 76.4163 |
| (0.15,0.15) | 3.4464 | 65.2920 | 51.8093 |
估计的相对误差百分比的比值比估计<我nline-formula>
|
|
|
|
|
|---|---|---|---|
| (0.01,0.01) | 50.9628 | 328.0343 | 295.7966 |
| (0.01,0.03) | 28.9076 | 628.4030 | 536.8268 |
| (0.01,0.05) | 263.4419 | 656.1777 | 520.8249 |
| (0.01,0.10) | 16.2619 | 578.3828 | 392.2232 |
| (0.01,0.15) | 14.5797 | 538.7736 | 339.4512 |
| (0.03,0.01) | 39.6535 | 94.3767 | 91.6069 |
| (0.03,0.03) | 19.2635 | 230.6324 | 208.1003 |
| (0.03,0.05) | 236.1914 | 243.2889 | 200.4045 |
| (0.03,0.10) | 7.4747 | 208.1561 | 138.4178 |
| (0.03,0.15) | 5.8990 | 190.0357 | 112.7091 |
| (0.05,0.01) | 39.1153 | 47.8002 | 51.9027 |
| (0.05,0.03) | 18.8401 | 151.4810 | 144.3693 |
| (0.05,0.05) | 234.9386 | 161.1356 | 138.3292 |
| (0.05,0.10) | 7.0955 | 134.3384 | 89.0341 |
| (0.05,0.15) | 5.5239 | 120.5896 | 68.6865 |
| (0.10,0.01) | 37.9815 | 13.1041 | 23.9423 |
| (0.10,0.03) | 17.8718 | 92.5224 | 99.4826 |
| (0.10,0.05) | 232.2048 | 99.9578 | 94.6234 |
| (0.10,0.10) | 6.2116 | 79.3838 | 54.3286 |
| (0.10,0.15) | 4.6546 | 68.8420 | 37.6836 |
| (0.15,0.01) | 37.2457 | 1.7496 | 15.7345 |
| (0.15,0.03) | 17.2201 | 73.1152 | 86.1673 |
| (0.15,0.05) | 230.3755 | 79.7100 | 81.5120 |
| (0.15,0.10) | 5.6227 | 61.2754 | 43.9902 |
| (0.15,0.15) | 4.0763 | 51.8150 | 28.4826 |
之前的百分比优势比使用EB估计,MMLE, MMUE当<我nline-formula>
我们的第一个例子是来自良好的研究(
真正的优势比,他们利用EB估计MMLE, MMUE,之前的相应百分比。
| 方法 | |||||
|---|---|---|---|---|---|
| 真正的 | 海尔哥哥 | MMLE | MMUE | ||
| 1例 |
|
2.4444 |
2.4309 | 2.4131 | 2.3430 |
| 之前 | - - - - - - | 0.5523 | 1.2805 | 4.1483 | |
|
|
|||||
| 第二个例子 |
|
0.1169 |
0.1230 | 0.1642 | 0.1350 |
| 之前 | - - - - - - | 5.2097 | 40.4643 | 15.5305 | |
第二个例子是取自Perondi等的研究。
基于模拟的优势比估计的研究罕见的事件有两个独立的二项数据,结果表明,该方法很好地执行。EB估计量的比值比也比其他两个更有效的估计,MMLE MMUE。另外,我们计划的估计量的另一种方法是优势比估计MMLE方法没有令人不安的原始数据。
作者宣称没有利益冲突。
作者感谢研究生学院,蒙国王科技大学北曼谷的金融支持。