Next-generation algorithms using multiple biomarkers for precise estimation of HIV infection duration and population level incidence

总结 鉴于艾滋病毒发病率估计在艾滋病毒监测和评价中的核心作用， 预防计划，必须有可靠，廉价和可用的方法， 以近乎实时的方式快速估计发病率。本提案旨在通过以下方式推进这一目标： 利用我们最近的发展，使用一系列生物标志物数据和艾滋病毒， 监测数据，以计算自 最近确诊的感染者。我们认为，这种估计不仅包含更多的 关于感染时间的准确信息比标准的二元分类（最近/长期）更准确。 术语），但它也给出了艾滋病毒发病率估计的现实置信区间，因为它 适当地考虑生物标志物测量和患者变化。我们进一步 一个准确的艾滋病发病率估计必须考虑到三个组成部分：1） 多检测算法（MAA）调整的感染时间（而不是诊断日期或 简单的近期/非近期二元分类）; 2）生活在 艾滋病毒感染者（PLHIV）在某个时间（可能是后来诊断）;和3）艾滋病毒感染者的数量 进入研究人群的阳性个体。具体而言，我们将：（1）开发 用于测量和建模个体生物标志物的实验室方案和算法， HIV感染时间的概率估计; 2）将联合收割机生物标志物组合成广义MAA 改进对艾滋病毒感染时间和艾滋病毒发病率的估计;以及3）将新的MAA应用于 确定感染的持续时间并估计不同人群中的艾滋病毒发病率。 使用大数据，总共> 13，000名患者确定了多达7种不同的生物标志物， 瑞典、美国、南非和津巴布韦，我们将建立纵向和单一时间点 培训数据以及验证数据，以开发经过验证的、公开可用的方法， 估计：1）个体感染艾滋病毒后的完整时间分布; 2）人口- 基于研究样本或现有监测系统的艾滋病毒发病率水平，以及 其随时间的变化，使用新的生物标志物、算法和模型。

SUMMARY Given the central role of HIV incidence estimation in both surveillance and evaluation of HIV prevention programs, it is essential to have reliable, inexpensive, and usable methods for quickly estimating incidence in near-real time. This proposal aims to advance that goal by capitalizing on our recent developments in using a spectrum of biomarker data and HIV surveillance data to compute the complete posterior probability distribution of the time since infection for recently diagnosed persons. We posit that such estimates contain not only more accurate information about time of infection than standard binary classification (recent/long- term), but that it also gives realistic confidence bounds on HIV incidence estimates as it appropriately takes biomarker measurements and patient variation into account. We further posit that an accurate HIV incidence estimate must take three components into account: 1) The multi-assay algorithm (MAA)-adjusted time of infection (rather than the date of diagnosis or simple recent/non-recent binary classification); 2) The number of undiagnosed persons living with HIV (PLHIV) at a certain time (which may be diagnosed later); and 3) The number of HIV positive individuals that move into the study population. Specifically, we will 1) Develop laboratory protocols and algorithms for measuring and modeling individual biomarkers for probabilistic estimation of time of HIV infection; 2) Combine biomarkers into a generalized MAA for improved estimation of time of HIV infection and HIV incidence; and 3) Apply new MAA to determine the duration of infection and estimate HIV incidence in different human populations. Using large data, in total >13,000 patients with up to 7 different biomarkers determined, from Sweden, USA, South Africa and Zimbabwe, we will establish longitudinal and single-time point training data, as well as validation data, to develop validated, publicly available, methods to estimate: 1) the full distribution of time since infection for individual PLHIV; and 2) population- level HIV incidence based on either study-based samples or existing surveillance systems, and changes thereof over time, using new biomarkers, algorithms and models.