Methods for Epidemiology Studies

一项研究评估了当混杂因素(如倾向得分或疾病风险得分)的汇总分数(而不是混杂因素本身)用于分析观测数据时，暴露影响估计的渐近偏差条件协变量。这项研究评估了队列数据、病例对照和配对病例对照研究的回归模型，这些研究根据总结分数进行了调整(和匹配)，并得出了渐近偏差。一项研究评估了线性混合模型(LMM)的综合拟合优度检验，方法是计算协变量空间划分单元内模型计算的观测值和期望值之间的差值的二次方形式。证明了在一些较温和的条件下，检验统计量具有渐近的卡方分布，并导出了局部替代下检验统计量的幂的解析表达式。开发了一种新的方法来确定具有共同危险因素的疾病亚型。这一方法被应用于国际淋巴瘤流行病学联合会，以显示B细胞和T细胞淋巴瘤风险特征的显著差异。还发展了一种新的方法来估计可归因于中介危险因素的遗传原因的疾病遗传性的比例，并利用该方法表明约24%的肺癌和7%的膀胱癌的遗传性可归因于吸烟的遗传决定因素。一项研究开发了一种混合模型，用于从筛查数据中估计风险，将基线存在的疾病风险与发病风险分开。标准的Kaplan-Meier估计在这种情况下是有偏见的。另一项研究开发了一种新的风险分层框架，称为平均风险分层(MRS)，即诊断测试为患者揭示的平均额外疾病数量。使用MRS，研究表明，对于很少呈阳性的检测，较大的风险差异并不意味着很好的风险分层，较大的约登指数(或AUC)并不意味着较好的风险分层，如果疾病太罕见，诊断测试的预期好处是仅通过MRS测试的函数。一份报告表明，使用肿瘤材料来检测病例中感染的分子研究的关联度量可能高估或低估感染与后续癌症风险之间的关系。为了提高Logistic回归模型在病例对照研究中的效率，提出了一种统计方法，即通过一系列无偏估计方程，利用关于特定协变量疾病患病率的辅助信息。提出了一种利用来自大数据来源的外部汇总水平信息进行约束最大似然分析的模型校正方法。进行了大量统计遗传学和基因组学研究。开发了稳健的统计程序来识别对所有疾病亚型具有统一效应或跨不同亚型具有异质性效应的遗传风险因素。提出了一种遗传关联检验，该检验可以解释在不同模型下由于基因-环境相互作用而产生的遗传效应的异质性。一项研究扩展了测试基因-环境相互作用的各种方法，以解释推测的基因数据。一项研究开发了一种方法，利用来自全球气候分析的总结水平的结果来估计效果-大小分布。将该方法应用于几种疾病的大型GWA的结果表明，涉及数千到数千个易感SNP的复杂性状的高度多基因结构。几项研究正在进行中，以探讨如何改进多基因风险预测模型的性能，该模型结合了来自大型GWAS的总结水平的结果和关于效应大小分布的各种类型的先验信息。建立了一种基于似然检验和I型错误评估的有效方法，用于癌症驱动基因检测中的互斥分析。这些方法被开发并应用于癌症基因组图谱(TCGA)项目的数据分析，从而识别出一些新的驱动基因。

A study assessed the asymptotic bias of estimates of exposure effects conditional on covariates when summary scores of confounders (e.g. the propensity score or disease risk score) , instead of the confounders themselves, are used to analyze observational data. The study evaluated regression models for cohort data, case-control and matched case-control studies that are adjusted for (and matched on) summary scores and derive the asymptotic bias. A study evaluated omnibus goodness of fit test for linear mixed models (LMMs) by computing a quadratic form of the differences between the observed and expected values computed from the model within cells of a partition of the covariate space. It showed that under some mild conditions, the test statistic has an asymptotic chi-square distribution and derived analytic expressions for the power of the test statistic under a local alternative. A new method was developed for determining subtypes of disease that share common risk factors. This methodology was applied in International Lymphoma Epidemiology Consortium to show strong differences in the risk profiles for B-Cell and T-Cell lymphomas. A new method was also developed for estimating the proportion of disease heritability that can be attributed genetic causes of mediating risk factors, and used this methodology to show that approximately 24% of lung cancer and 7% of bladder cancer heritability can be attributed to the genetic determinants of smoking. A study developed a mixture model for estimating risk from screening data, separating risk of disease present at baseline from risk of onset of incident disease. Standard Kaplan-Meier estimates are biased in this situation. Another study developed a new framework for risk stratification called mean risk stratification (MRS), which is the average amount of extra disease that a diagnostic test reveals for a patient. Using MRS, it was shown that a big risk difference does not imply good risk stratification for tests that are rarely positive, that a large Youden's index (or AUC) does not imply good risk stratification if disease is too rare, and that the expected benefit of a diagnostic test is a function of the test solely through MRS. A report showed that measures of association for molecular studies that use material from tumors to detect infection in cases can over- or underestimate the relationship between infections and subsequent cancer risk. A statistical procedure has been proposed to improves the efficiency of the logistic regression model for a case-control study by utilizing auxiliary information on covariate-specific disease prevalence via a series of unbiased estimating equations. A method was developed for constrained maximum likelihood analysis for model calibration using external summary-level information from big-data sources.A number of studies statistical genetic and genomics were conducted A robust statistical procedure has been developed to identify genetic risk factors that have either a uniform effect for all disease subtypes or heterogeneous effects across different subtypes. A test for genetic association was proposed that can account for heterogeneity in genetic effects due to gene-environment interactions under alternative models. A study developed extensions of various methods for testing gene-environment interactions to account for imputed genotype data. A study developed method for estimation of effect-size distribution using summary-level results from GWAS. Application of the method to results from large GWAS of several diseases indicate highly polygenic architecture of complex traits involving thousands to tends of thousands of susceptibility SNPs. Several studies are ongoing to investigate how to improve performance of polygenic risk prediction models incorporating summary-level results from large GWAS and various types of prior information on effect-size-distribution. A likelihood-based test and a valid method for type-I error evaluation were developed for mutual exclusivity analysis in detection of cancer driver gene. The methods were developed and applied for analysis of data from the The Cancer Genome Atlas (TCGA) project leading to identification a number of novel driver genes.