诊断性试验Meta分析

Revman,Stata,Meta-disc在诊断试验准确性（DTA）

系统评价中的应用

文献数据摘自《ProGRP与NSE对小细胞肺癌诊断价值的meta分析》 文中提取数据 作者 国家 研究方法 盲法 研究金标病例数 Schneider Stieber Molina Nissan 德国 德国 西班牙 以色列 前瞻 回顾 前瞻 前瞻 -- 回顾 回顾 -- -- -- -- -- -- -- -- 是 是 是 -- -- 连续 病理 -- 病理 298 314 802 162 359 245 326 602 100 144 阳性界值（pg/ml） a 29.1 38.3 50 48 49 53 33.8 50 50 46 b 9.6 11.9 25 22 7.5 17 10.6 8,1 16.3 16.3 a 35 41 134 29 74 117 73 80 25 46 TP FP FN TN 对象 准 b 38 39 114 18 49 110 63 79 19 40 a 18 9 79 6 11 2 22 6 6 9 b 35 44 50 12 10 4 43 26 8 16 a 16 46 41 8 40 29 28 47 9 17 b 13 48 61 19 65 36 38 48 15 23 a 229 218 548 119 234 97 203 469 60 72 b 212 183 577 113 235 95 182 449 58 65 连续 病理 连续 病理 连续 病理 -- 病理 Shibayama 日本 Lamy Takada 法国 日本 连续 病理 连续 -- -- -- 病理 病理 Yamaguchi 日本 Sun Yang 中国 中国 注： 表中 10 个原始研究均使用酶联免疫吸附测定法检测阳性界值； TP= 真阳性数； FP= 假阳性数； FN= 假阴性数； TN= 真阴性数a:ProGRP,b: NSE

Revman5.2

新建诊断试验准确性(DTA)系统评价模板

添加所有纳入研究

此处对每篇文献QUADAS2质量特征进行描述，以便探讨异质性来源及作表图

数据分析里面添加所要研究的待评价诊断试验

Patient SelectionIndex Test1424244444465631Reference StandardFlow and Timing0"5U0u100%0%Pu0%Risk of BiasHighUnclearApplicability ConcernsLow Risk of BiasApplicability ConcernsLamy 2000?–?????–??Molina 2009Nissan 2009+?+–+–+?–?–?Schneider 2003Shibayama 2001Stieber 1999+–++––+?–?––––––?+Sun 2005Takada 1996?+?+?+?+?–??––Yamaguchi 1995Yang 2005?+++–High?Unclear+Low Reference Standard–?Reference StandardPatient SelectionPatient SelectionFlow and TimingIndex TestIndex Test–?–???–? 异质性来源

? 在DTA系统评价里面不能直接进行似然比、诊断比值比的森林图以及各指标

漏斗图制作，但可以改变四个表数据模式或直接计算相关指标，添加入干预性系统评价模板中进行制作及查看异质性、发表偏倚（漏斗图）。

Stata12

一 拟合双变量混合效应模型：midas命令

1.计算所有诊断试验统计学指标（敏感度、 特异度、 似然比、 诊断比值比等） 及异质性检验统计量:

SUMMARY DATA AND PERFORMANCE ESTIMATESBivariate Binomial Mixed Model Number of studies = 10 Reference-positive Subjects = 935 Reference-negative Subjects = 2417 Pretest Prob of Disease =0.279 Between-study variance(varlogitSEN) =0.136, 95% CI = [0.041-0.452] Between-study variance(varlogitSPE)= 0.406, 95% CI = [0.133-1.241] Correlation (Mixed Model)= -0.491 ROC Area, AUROC = 0.89 [0.86 - 0.92]Heterogeneity (Chi-square): LRT_Q = 55.419, df =2.00, LRT_p =0.000Inconsistency (I-square): LRT_I2 = 96.39, 95% CI = [93.72-99.06]Parameter Estimate 95% CISensitivity 0.702 [ 0.641, 0.756]Specificity 0.943 [ 0.913, 0.963]Positive Likelihood Ratio 12.348 [ 8.245, 18.494]Negative Likelihood Ratio 0.316 [ 0.262, 0.381]Diagnostic Score 3.665 [ 3.229, 4.102]Diagnostic Odds Ratio 39.071 [ 25.251, 60.456]

2. 绘制敏感度、 特异度森林图:

STUDY(YEAR)SENSITIVITY (95% CI)Yang 2005Sun 2005Yamaguchi 1995Takada 1996Lamy 2000Shibayama 2001Nissan 2009Molina 2009Stieber 1999Schneider 20030.73 [0.60 - 0.83]0.74 [0.56 - 0.87]0.63 [0.54 - 0.71]0.72 [0.62 - 0.81]0.80 [0.73 - 0.86]0.65 [0.55 - 0.74]0.78 [0.62 - 0.90]0.77 [0.70 - 0.83]0.47 [0.36 - 0.58]0.69 [0.54 - 0.81]COMBINEDQ = 39.00, df = 9.00, p = 0.00I2 = 76.92 [62.83 - 91.01]0.4SENSITIVITY0.9 0.70[0.64 - 0.76]

SPECIFICITY (95% CI) STUDY(YEAR)Yang 2005Sun 2005Yamaguchi 1995Takada 1996Lamy 2000Shibayama 2001Nissan 2009Molina 2009Stieber 1999Schneider 20030.89 [0.80 - 0.95]0.91 [0.81 - 0.97]0.99 [0.97 - 1.00]0.90 [0.86 - 0.94]0.98 [0.93 - 1.00]0.96 [0.92 - 0.98]0.95 [0.90 - 0.98]0.87 [0.85 - 0.90]0.96 [0.93 - 0.98]0.93 [0.89 - 0.96]COMBINEDQ = 75.58, df = 9.00, p = 0.00I2 = 88.09 [82.01 - 94.17]0.8SPECIFICITY1.0 0.94[0.91 - 0.96]

3. 绘制 ROC 曲线图：

1.0SROC with Confidence and Predictive EllipsesObserved DataSummary Operating PointSENS = 0.70 [0.64 - 0.76]SPEC = 0.94 [0.91 - 0.96]SROC CurveAUC = 0.89 [0.86 - 0.92]95% Confidence Ellipse95% Prediction EllipseSensitivity0.50.01.00.5Specificity0.0

4. 绘制漏斗图， 识别发表偏倚:

STATISTICAL TESTS FOR SMALL STUDY EFFECTS/PUBLICATION

BIAS

> yb Coef. Std. Err. t P>|t| [95% Conf. Interv> al] > Bias -.7005376 17.62442 -0.04 0.969 -41.34253 39.94> 146 Intercept 3.78234 1.105751 3.42 0.009 1.232475 6.332> 206

1000Log Odds Ratio versus 1/sqrt(Effective Sample Size)(Deeks)StudyRegressionLine10.04101000.061/root(ESS)0.080.100.12

5. 绘制似然比森林图:

STUDY(YEAR)DLR POSITIVE (95% CI)Yang 2005Sun 2005Yamaguchi 1995Takada 1996Lamy 2000Shibayama 2001Nissan 2009Molina 2009Stieber 1999Schneider 20036.57 [3.49 - 12.39]8.09 [3.67 - 17.81]49.87 [22.27 - 111.67]7.39 [4.88 - 11.19]39.67 [10.04 - 156.77]14.46 [7.99 - 26.16]16.33 [7.35 - 36.30]6.08 [4.87 - 7.59]11.89 [6.03 - 23.41]9.42 [5.82 - 15.25]COMBINEDQ = 51.60, df = 9.00, p = 0.00I2 = 72.67 [72.67 - 92.45]3.5DLR POSITIVE156.8 12.35[8.24 - 18.49] STUDY(YEAR)DLR NEGATIVE (95% CI)Yang 2005Sun 2005Yamaguchi 1995Takada 1996Lamy 2000Shibayama 2001Nissan 2009Molina 2009Stieber 1999Schneider 20030.30 [0.20 - 0.46]0.29 [0.17 - 0.51]0.37 [0.30 - 0.47]0.31 [0.22 - 0.42]0.20 [0.15 - 0.28]0.37 [0.29 - 0.47]0.23 [0.12 - 0.42]0.27 [0.20 - 0.35]0.55 [0.45 - 0.67]0.34 [0.23 - 0.51]COMBINEDQ = 42.30, df = 9.00, p = 0.00I2 = 78.72 [66.00 - 91.44]0DLR NEGATIVE1 0.32[0.26 - 0.38]

6. 绘制诊断比值比森林图：

STUDY(YEAR)ODDS RATIO (95% CI)Yang 2005Sun 2005Yamaguchi 1995Takada 1996Lamy 2000Shibayama 2001Nissan 2009Molina 2009Stieber 1999Schneider 200321.65 [8.90 - 52.64]27.78 [8.94 - 86.29]133.05 [55.07 - 321.46]24.06 [12.95 - 44.68]195.67 [45.54 - 840.84]39.35 [19.22 - 80.58]71.90 [23.14 - 223.38]22.67 [14.87 - 34.57]21.59 [9.81 - 47.50]27.83 [12.99 - 59.60]COMBINEDQ =10306.42, df = 9.00, p = 0.00I2 = 99.91 [99.90 - 99.92]9ODDS RATIO841 39.07[25.25 - 60.46] STUDY(YEAR)DIAGNOSTIC SCORE (95% CI)Yang 2005Sun 2005Yamaguchi 1995Takada 1996Lamy 2000Shibayama 2001Nissan 2009Molina 2009Stieber 1999Schneider 20033.07 [1.21 - 3.07]3.32 [1.21 - 3.32]4.89 [2.21 - 4.89]3.18 [1.41 - 3.18]5.28 [2.11 - 5.28]3.67 [1.63 - 3.67]4.28 [1.73 - 4.28]3.12 [1.49 - 3.12]3.07 [1.26 - 3.07]3.33 [1.41 - 3.33]COMBINEDQ = 23.94, df = 9.00, p = 0.00I2 = 62.41 [36.63 - 88.18]1.2DIAGNOSTIC SCORE5.3 3.67[3.23 - 4.10]

7.绘制验前概率、 验后概率图：

验前概率=患病率，验后概率=验前概率*似然比

0.10.20.30.50.712357102030405060708090939597989999.399.599.799.899.9Fagan's Nomogram99.999.899.799.599.39998979593908070605040302010753210.70.50.30.20.1Likelihood Ratio10005002001005020105210.50.20.10.050.020.010.0050.0020.001Post-test Probability (%)Prior Prob (%) = 20LR_Positive = 12Post_Prob_Pos (%) = 76LR_Negative = 0.32Post_Prob_Neg (%) = 7

二 拟合HSROC模型：metandi命令 1.合并统计量命令

. metandi tp fp fn tnRefining starting values: Iteration 0: log likelihood = -73.728348 (not concave)Iteration 1: log likelihood = -69.302533 Iteration 2: log likelihood = -67.980313 Iteration 3: log likelihood = -67.378598 Performing gradient-based optimization: Iteration 0: log likelihood = -67.378598 Iteration 1: log likelihood = -67.370761 Iteration 2: log likelihood = -67.370744 Iteration 3: log likelihood = -67.370744 Meta-analysis of diagnostic accuracyLog likelihood = -67.370744 Number of studies = > 10 > Coef. Std. Err. z P>|z| [95% Conf. Interv> al] > Bivariate E(logitSe) .8564618 .1410043 .5800985 1.132> 825 E(logitSp) 2.808916 .230571 2.357005 3.260> 826Var(logitSe) .1356511 .0833172 .0407018 .4520> 984Var(logitSp) .4063271 .2312948 .1331508 1.23> 996Corr(logits) -.4913658 .3666173 -.9024235 .3879> 676 > HSROC Lambda 3.261869 .2776274 2.71773 3.806> 009 Theta -.5042014 .3386717 -1.167986 .159> 583 beta .5485361 .3866854 1.42 0.156 -.2093534 1.306> 426 s2alpha .2388279 .1917059 .0495257 1.151> 701 s2theta .1750668 .0938807 .0611986 .5008> 023 > Summary pt. Se .7019209 .029502 .6410901 .7563> 599 Sp .9431557 .0123616 .9134894 .9630> 602 DOR 39.07088 8.70083 25.25202 60.45> 194 LR+ 12.34813 2.544551 8.245105 18.49> 296 LR- .3160444 .0300356 .2623332 .3807> 526 1/LR- 3.164112 .3007045 2.626377 3.811> 945 > Covariance between estimates of E(logitSe) & E(logitSp) -.0117264

2.绘制SROC曲线

1Sensitivity01.2.4.6.8.8.6.4Specificity.20Study estimateHSROC curve95% predictionregionSummary point95% confidenceregion

Meta-disc14

表2 Meta-Disc软件的主要功能

主要功能

Describing primary results and exploring heterogeneity ? Tabular results ? Forest

plots(sensitivity,specificity,LRs,dOR) ? ROC plane scatter-plots

? Filtering/subgrouping capacities Exporing Threshold effect

? Spearman correlation coefficient ? ROC plane plots

SROC curve fitting.Area under the curve(AUC) and Q

Meta-regression analysis

? Univariate and multivariate Moses and

Litteenberg model(weight and unweight) Statistical polling of indices ? Fixed effect model ? Random effect model

回归分析，探讨异质性来源

? （加权或未加权）单变量及多变量Moses

Litteenberg模型 合并统计量 ? 固定效应模型 ? 随机效应模型

? 将结果以表格形式列出

? 以森林图形式显示灵敏度、特异度、似然比和诊

断比值比 ? ROC平面散状图 ? 亚组分析 探讨阈值效应 ? Spearman相关系数 ? ROC平面图

拟合SROC曲线、计算AUC和Q指数 说明

描述原始结果和探索异质性

? Cochran-Q,Chi-Square, Inconsistency index ? 判断研究间异质性

数据录入

本文来源：https://www.bwwdw.com/article/8tv2.html