Serghiou Stylianos, Patel Chirag J, Tan Yan Yu, Koay Peter, Ioannidis John P A
College of Medicine and Veterinary Medicine, The University of Edinburgh, 47 Little France Crescent, Edinburgh EH16 4TJ, Edinburgh, UK.
Department of Biomedical Informatics, Harvard Medical School, 10 Shattuck Street, 4th Floor, Boston, MA 02115, USA.
J Clin Epidemiol. 2016 Mar;71:58-67. doi: 10.1016/j.jclinepi.2015.09.004. Epub 2015 Sep 28.
Instead of evaluating one risk factor at a time, we illustrate the utility of "field-wide meta-analyses" in considering all available data on all putative risk factors of a disease simultaneously.
We identified studies on putative risk factors of pterygium (surfer's eye) in PubMed, EMBASE, and Web of Science. We mapped which factors were considered, reported, and adjusted for in each study. For each putative risk factor, four meta-analyses were done using univariate only, multivariate only, preferentially univariate, or preferentially multivariate estimates.
A total of 2052 records were screened to identify 60 eligible studies reporting on 65 putative risk factors. Only 4 of 60 studies reported both multivariate and univariate regression analyses. None of the 32 studies using multivariate analysis adjusted for the same set of risk factors. Effect sizes from different types of regression analyses led to significantly different summary effect sizes (P-value < 0.001). Observed heterogeneity was very high for both multivariate (median I(2), 76.1%) and univariate (median I(2), 85.8%) estimates. No single study investigated all 11 risk factors that were statistically significant in at least one of our meta-analyses.
Field-wide meta-analyses can map availability of risk factors and trends in modeling, adjustments and reporting, as well as the impact of differences in model specification.
我们不再一次评估一个风险因素，而是展示“全领域荟萃分析”在同时考虑一种疾病所有假定风险因素的所有可用数据方面的效用。
我们在PubMed、EMBASE和科学网中检索关于翼状胬肉（冲浪者眼）假定风险因素的研究。我们梳理了每项研究中考虑、报告和调整的因素。对于每个假定风险因素，使用仅单变量、仅多变量、优先单变量或优先多变量估计进行了四项荟萃分析。
共筛选了2052条记录，以确定60项符合条件的研究，这些研究报告了65个假定风险因素。60项研究中只有4项报告了多变量和单变量回归分析。32项使用多变量分析的研究中，没有一项对同一组风险因素进行调整。不同类型回归分析的效应量导致显著不同的汇总效应量（P值<0.001）。多变量估计（中位数I(2)，76.1%）和单变量估计（中位数I(2)，85.8%）的观察到的异质性都非常高。没有一项研究调查了在我们至少一项荟萃分析中具有统计学意义的所有11个风险因素。
全领域荟萃分析可以梳理风险因素的可得性以及建模、调整和报告方面的趋势，以及模型规范差异的影响。
