Improved Methods for Assessing Immune Correlates in HIV-1 Vaccine Trials

说明（由申请人提供）：艾滋病毒疫苗研究是防治全球艾滋病毒/艾滋病流行病的重要组成部分。发现预测疫苗诱导保护的免疫反应，或“免疫相关物”，对艾滋病毒疫苗研究具有重大影响，因为它揭示了这种保护的机制，并有助于改进疫苗设计。该方案由两套统计方法组成，以提高识别免疫相关物的统计能力。目的1开发方法，将有助于研究人员选择最佳的免疫反应阳性标准。免疫反应的阳性标准是一套将受试者的反应分为阴性或阳性的规则。一个好的阳性标准可以用来帮助定义免疫反应变量，以评估它们与感染状态的相关性。该方案采用决策理论的方法来选择判定标准，以平衡假阳性率和假阴性率。目标2开发用于测量免疫反应的重要技术的分析方法：多路头阵列（MBA）。处理MBA数据的关键步骤是使用包含已知量待测分析物的标准样品的数据拟合浓度-响应曲线。较差的曲线拟合导致较大的测量误差，这将削弱感染状态与观察到的免疫反应之间的统计关联。该提案探讨了曲线拟合中的两个问题。首先，比较了对数变换荧光强度的加权最小二乘和非加权最小二乘两种处理观测荧光强度均方差关系的方法。其次，比较了浓度-响应曲线形状的三种模型：四参数logistic模型、五参数logistic模型和半参数模型。

DESCRIPTION (provided by applicant): HIV vaccine research is an essential part of the fight against the global HIV/AIDS pandemic. Finding immune responses that predict vaccine-induced protection, or "immune correlates", has a great impact on the HIV vaccine research, as it sheds light on the mechanism of the protection and helps improve vaccine design. This proposal is composed of two sets of statistical methods for improving the statistical power of identifying immune correlates. Aim 1 develops methods that will help investigators choose the best immune response positivity criteria. A positivity criterion for an immune response is a set of rules that classifies a subject's response as either negative or positive. A good positivity criteron can be used to help define immune response variables to assess their correlation with infection status. The proposal takes a decision theoretic approach to choosing positivity criteria that balance the false positive rates and the false negative rates. Aim 2 develops analytical methods for an important technology used to measure immune responses: the multiplex bead array (MBA). A key step in processing MBA data is to fit a concentration-response curve using data from standard samples which contain known amounts of analytes to be measured. Poor curve fits lead to large measurement error, which will dampen the statistical association between the infection status and the observed immune responses. The proposal examines two issues in curve fitting. First, two approaches for handling the mean-variance relationship of the observed fluorescence intensity are compared: weighted least squares and un-weighted least squares on log-transformed fluorescence intensity. Second, three models for the shape of the concentration-response curves are compared: four-parameter logistic model, five-parameter logistic model, and a semi-parametric model.