Macalinao Maria Lourdes M, Fornace Kimberly M, Reyes Ralph A, Hall Tom, Bareng Alison Paolo N, Adams John H, Huon Christèle, Chitnis Chetan E, Luchavez Jennifer S, Tetteh Kevin K A, Yui Katsuyuki, Hafalla Julius Clemence R, Espino Fe Esperanza J, Drakeley Chris J
Department of Parasitology and National Reference Laboratory for Malaria and Other Parasites, Research Institute for Tropical Medicine, Department of Health, Muntinlupa City, Philippines.
Faculty of Infectious and Tropical Diseases, Department of Infection Biology, London School of Hygiene and Tropical Medicine, London, United Kingdom.
Lancet Reg Health West Pac. 2023 May 20;37:100792. doi: 10.1016/j.lanwpc.2023.100792. eCollection 2023 Aug.
Assessing the status of malaria transmission in endemic areas becomes increasingly challenging as countries approach elimination. Serology can provide robust estimates of malaria transmission intensities, and multiplex serological assays allow for simultaneous assessment of markers of recent and historical malaria exposure.
Here, we evaluated different statistical and machine learning methods for analyzing multiplex malaria-specific antibody response data to classify recent and historical exposure to and . To assess these methods, we utilized samples from a health-facility based survey (n = 9132) in the Philippines, where we quantified antibody responses against 8 and 6 -specific antigens from 3 sites with varying transmission intensity.
Measurements of antibody responses and seroprevalence were consistent with the 3 sites' known endemicity status. Among the models tested, a machine learning (ML) approach (Random Forest model) using 4 serological markers (PfGLURP R2, Etramp5.Ag1, GEXP18, and PfMSP1) gave better predictions for recent infection in Palawan (AUC: 0.9591, CI 0.9497-0.9684) than individual antigen seropositivity. Although the ML approach did not improve infection predictions, ML classifications confirmed the absence of recent exposure to and in both Occidental Mindoro and Bataan. For predicting historical and transmission, seroprevalence and seroconversion rates based on cumulative exposure markers AMA1 and MSP1 showed reliable trends in the 3 sites.
Our study emphasizes the utility of serological markers in predicting recent and historical exposure in a sub-national elimination setting, and also highlights the potential use of machine learning models using multiplex antibody responses to improve assessment of the malaria transmission status of countries aiming for elimination. This work also provides baseline antibody data for monitoring risk in malaria-endemic areas in the Philippines.
Newton Fund, Philippine Council for Health Research and Development, UK Medical Research Council.
随着各国接近消除疟疾目标，评估疟疾流行地区的传播状况变得越来越具有挑战性。血清学可以提供对疟疾传播强度的可靠估计，而多重血清学检测可以同时评估近期和既往疟疾暴露的标志物。
在此，我们评估了不同的统计和机器学习方法，用于分析多重疟疾特异性抗体反应数据，以分类近期和既往对疟原虫和间日疟原虫的暴露情况。为了评估这些方法，我们利用了菲律宾一项基于医疗机构的调查（n = 9132）中的样本，在该调查中，我们对来自3个传播强度不同地点的针对8种疟原虫和6种间日疟原虫特异性抗原的抗体反应进行了量化。
抗体反应测量值和血清阳性率与3个地点已知的流行状况一致。在所测试的模型中，使用4种血清学标志物（PfGLURP R2、Etramp5.Ag1、GEXP18和PfMSP1）的机器学习（ML）方法（随机森林模型）对巴拉望近期疟原虫感染的预测（AUC：0.9591，CI 0.9497 - 0.9684）优于单个抗原血清阳性率。尽管ML方法并未改善疟原虫感染预测，但ML分类证实西民都洛和巴丹均不存在近期对疟原虫和间日疟原虫的暴露。对于预测既往疟原虫和间日疟原虫传播，基于累积暴露标志物AMA1和MSP1的血清阳性率和血清转换率在3个地点显示出可靠的趋势。
我们的研究强调了血清学标志物在国家以下消除疟疾环境中预测近期和既往暴露的效用，还突出了使用多重抗体反应的机器学习模型在改善对旨在消除疟疾国家的疟疾传播状况评估方面的潜在用途。这项工作还提供了用于监测菲律宾疟疾流行地区风险的基线抗体数据。
牛顿基金、菲律宾卫生研究与发展委员会、英国医学研究理事会。
