Faculty of Health Sciences, Department of Biostatistics, University of the Free State, Bloemfontein, South Africa.
South African Department of Science and Technology-National Research Foundation Centre of Excellence in Epidemiological Modelling and Analysis, Stellenbosch University, Stellenbosch, South Africa.
PLoS One. 2022 Jul 28;17(7):e0271763. doi: 10.1371/journal.pone.0271763. eCollection 2022.
Testing for 'recent HIV infection' is common in surveillance, where only population-level estimates (of incidence) are reported. Typically, 'recent infection' is a category, obtained by applying a threshold on an underlying continuous biomarker from some laboratory assay(s). Interpreting the biomarker values obtained for individual subjects, as estimates of the date of infection, has obvious potential applications in the context of studies of early infection, and has also for some years attracted significant interest as an extra component of post-test counselling and treatment initiation. The applicable analyses have typically run aground on the complexity of the full biomarker growth model, which is in principle a non-linear mixed-effects model of unknown structure, the fitting of which seems infeasible from realistically obtainable data.
It is known that to estimate Mean Duration of Recent Infection (MDRI) at a given value of the recent/non-recent -infection discrimination threshold, one may compress the full biomarker growth model into a relation capturing the probability of a recent test result as a function of time t since infection, given a value of assay threshold h which defines the recent/non-recent discrimination. We demonstrate that the derivative (gradient), with respect to h. of the probability of recent infection, seen as a function of both t and h, is identical to the formal likelihood relevant to Bayesian inference of the time since seroconversion, for a subject yielding an assay result h, at or close to the date of their first positive HIV test. This observation bypasses the need for fitting a complex detailed biomarker growth model. Using publicly available data from the CEPHIA collaboration, we calibrated this likelihood function for the Sedia Lag assay, and performed Bayesian inference on hypothetical infection data.
We demonstrate the generation of posteriors for infection date, for patients with various delays between their last negative and first positive HIV test, and a range of LAg assay results (ODn) hypothetically obtained on the date of the first positive result.
Depending on the last-negative / first-positive interval, there is a range of ODn values that yields posteriors significantly different from the uniform prior one would be left with based merely on interval censoring. Hence, a LAg ODn obtained on the date of, or soon after, diagnosis contains potentially significant information about infection dating. It seems worth analysing other assays with meaningful dynamic range, especially tests already routinely used in primary HIV diagnosis (for example chemiluminescent assays and reader/cartridge lateral flow tests which admit objective variable line intensity readings) which have a sufficient dynamic range that corresponds to a clinically meaningful range of times-since-infection that are worth distinguishing from each other.
在监测中，检测“近期 HIV 感染”很常见，仅报告人群水平的估计值（发病率）。通常，“近期感染”是一个类别，通过在某些实验室检测（多个）的基础连续生物标志物上应用阈值来获得。将个体受试者的生物标志物值解释为感染日期的估计值，在早期感染研究中具有明显的潜在应用价值，并且多年来作为检测后咨询和治疗启动的额外组成部分也引起了极大的关注。适用于这些分析的方法通常因完整生物标志物增长模型的复杂性而搁浅，从原则上讲，该模型是未知结构的非线性混合效应模型，从实际可获得的数据来看，拟合该模型似乎是不可行的。
众所周知，要估计特定近期/非近期感染区分阈值下的近期感染平均持续时间（MDRI），可以将完整的生物标志物增长模型压缩为一种关系，该关系捕获了给定检测阈值 h 的情况下，给定 h 值定义了近期/非近期的区分，测试结果为近期的概率作为时间 t 的函数，t 是自感染以来的时间。我们证明，作为 t 和 h 的函数的近期感染概率的导数（梯度），与针对特定 h 值的受试者的血清转换后时间的贝叶斯推断的正式似然函数相同，对于在接近首次 HIV 检测日期时获得 h 的检测结果的受试者。该观察结果绕过了拟合复杂详细生物标志物增长模型的需要。使用 CEPHIA 合作组织提供的公开数据，我们针对 Sedia Lag 检测校准了该似然函数，并对假设的感染数据进行了贝叶斯推断。
我们证明了对于各种最后一次阴性和首次阳性 HIV 检测之间的延迟以及假设在首次阳性结果日期获得的各种 LAg 检测结果（ODn）的患者，都可以生成感染日期的后验。
根据最后一次阴性/首次阳性的间隔，存在一个 ODn 值范围，这些值产生的后验与仅基于区间 censoring 而得出的均匀先验明显不同。因此，在诊断日期或诊断后不久获得的 LAg ODn 包含有关感染日期的潜在重要信息。似乎值得分析其他具有有意义的动态范围的检测方法，特别是已经在初次 HIV 诊断中常规使用的检测方法（例如化学发光检测法和接受客观可变线强度读数的阅读器/盒式横向流动检测法），这些检测方法具有足够的动态范围，可以对应于彼此之间有临床意义的感染时间范围。
