Nyaga Victoria Nyawira, Arbyn Marc
Unit of Cancer Epidemiology - Belgian Cancer Centre, Sciensano, Juliette Wytsmanstraat 14, 1050, Brussels, Belgium.
Arch Public Health. 2022 Mar 29;80(1):95. doi: 10.1186/s13690-021-00747-5.
Although statistical procedures for pooling of several epidemiological metrics are generally available in statistical packages, those for meta-analysis of diagnostic test accuracy studies including options for multivariate regression are lacking. Fitting regression models and the processing of the estimates often entails lengthy and tedious calculations. Therefore, packaging appropriate statistical procedures in a robust and user-friendly program is of great interest to the scientific community.
metadta is a statistical program for pooling of diagnostic accuracy test data in Stata. It implements both the bivariate random-effects and the fixed-effects model, allows for meta-regression, and presents the results in tables, a forest plot and/or summary receiver operating characteristic (SROC) plot. For a model without covariates, it quantifies the unexplained heterogeneity due to between-study variation using an I statistic that accounts for the mean-variance relationship and the correlation between sensitivity and specificity. To demonstrate metadta, we applied the program on two published meta-analyses on: 1) the sensitivity and specificity of cytology and other markers including telomerase for primary diagnosis of bladder cancer, and 2) the accuracy of human papillomavirus (HPV) testing on self-collected versus clinician-collected samples to detect cervical precancer.
Without requiring a continuity correction, the pooled sensitivity and specificity generated by metadta of telomerase for the diagnosis of primary bladder cancer was 0.77 [95% CI, 0.70, 0.82] and 0.91 [95% CI, 0.75, 0.97] respectively. Metadta also allowed to assess the relative accuracy of HPV testing on self- versus clinician-taken specimens using data from comparative studies conducted in different clinical settings. The analysis showed that HPV testing with target-amplification assays on self-samples was as sensitive as on clinician-samples in detecting cervical pre-cancer irrespective of the clinical setting.
The metadta program implements state of art statistical procedures in an attempt to close the gap between methodological statisticians and systematic reviewers. We expect the program to popularize the use of appropriate statistical methods for diagnostic meta-analysis further.
尽管在统计软件包中通常可以获得用于汇总多个流行病学指标的统计程序，但缺乏用于诊断试验准确性研究的荟萃分析的程序，包括多变量回归选项。拟合回归模型和估计值的处理通常需要冗长而繁琐的计算。因此，将适当的统计程序打包成一个强大且用户友好的程序对科学界具有极大的吸引力。
metadta是一个用于在Stata中汇总诊断准确性测试数据的统计程序。它实现了双变量随机效应模型和固定效应模型，允许进行荟萃回归，并以表格、森林图和/或汇总受试者工作特征（SROC）图的形式呈现结果。对于无协变量的模型，它使用一个考虑了均数-方差关系以及敏感性和特异性之间相关性的I统计量来量化由于研究间变异导致的无法解释的异质性。为了演示metadta，我们将该程序应用于两项已发表的荟萃分析：1）细胞学和其他标志物（包括端粒酶）对膀胱癌初步诊断的敏感性和特异性，以及2）人乳头瘤病毒（HPV）检测在自我采集样本与临床医生采集样本上检测宫颈上皮内瘤变的准确性。
在无需连续性校正的情况下，metadta得出的端粒酶用于原发性膀胱癌诊断的合并敏感性和特异性分别为0.77 [95%可信区间，0.70, 0.82] 和0.91 [95%可信区间，0.75, 0.97]。Metadta还允许使用在不同临床环境中进行的比较研究的数据来评估HPV检测在自我采集样本与临床医生采集样本上的相对准确性。分析表明，无论临床环境如何，使用靶向扩增检测法对自我采集样本进行HPV检测在检测宫颈上皮内瘤变方面与对临床医生采集样本一样敏感。
metadta程序实施了先进的统计程序，试图缩小方法统计学家和系统评价者之间的差距。我们期望该程序能进一步推广适用于诊断性荟萃分析的统计方法的使用。
