Ugwu Chigozie Louisa Jane, Zewotir Temesgen
School of Mathematics, Statistics and Computer Science, University of KwaZulu-Natal, Private Bag X54001 Durban 4000, 3630 Westville, Durban, South Africa.
J Epidemiol Glob Health. 2020 Dec;10(4):304-314. doi: 10.2991/jegh.k.200814.001. Epub 2020 Aug 21.
Although malaria burden has declined globally following scale up of intervention, the disease has remained a leading cause of hospitalization and deaths among children aged under-5 years in Nigeria. Malaria is known to be related to climate and environmental conditions. Previous research has usually studied the effects of these factors, neglecting possible correlation between them, high correlation among variables is a source of multicollinearity that induces overfitting in regression modelling. In this paper, a factor analysis was first introduced to circumvent the issue of multicollinearity and a Generalized Additive Model (GAM) was subsequently explored to identify the important risk factors that might influence the prevalence of childhood malaria in Nigeria. The GAM incorporated the complexity of the survey data, while simultaneously modelling the nonlinear and spatial random effects to allow a more precise identification of the major malaria risk factors that influence the geographical distribution of the disease. From our findings, the three latent factor components (constituted by humidity, precipitation, potential evapotranspiration, and wet days/maximum and minimum temperature/proximity to permanent waters, respectively) were significantly associated with malaria prevalence. Our analysis also detected statistically significant and nonlinear effect of altitude: the risk of malaria increased with lower values but declined sharply with higher values. A significant spatial variability in under-5 malaria prevalence across the survey clusters was also observed; malaria burden was higher in the northern part of Nigeria. Investigating the impact of important risk factors and geographical location on childhood malaria is of high relevance for the sustainable development goals (SDGs) 2015-2030 Agenda on malaria eradication, and we believe that the information obtained from this study and the generated risk maps can be useful to effectively target intervention efforts to high-risk areas based on climate and environmental context.
尽管随着干预措施的扩大，全球疟疾负担有所下降，但在尼日利亚，该疾病仍然是5岁以下儿童住院和死亡的主要原因。众所周知，疟疾与气候和环境条件有关。以往的研究通常只考察这些因素的影响，而忽略了它们之间可能存在的相关性，变量之间的高度相关性是多重共线性的一个来源，会导致回归模型中的过度拟合。在本文中，首先引入因子分析以规避多重共线性问题，随后探索了广义相加模型（GAM），以识别可能影响尼日利亚儿童疟疾流行率的重要风险因素。GAM纳入了调查数据的复杂性，同时对非线性和空间随机效应进行建模，以便更精确地识别影响该疾病地理分布的主要疟疾风险因素。根据我们的研究结果，三个潜在因子成分（分别由湿度、降水量、潜在蒸散量以及雨日/最高和最低温度/与永久性水域的距离构成）与疟疾流行率显著相关。我们的分析还检测到海拔具有统计学意义的非线性效应：疟疾风险在海拔较低时增加，但在海拔较高时急剧下降。在整个调查集群中，5岁以下儿童疟疾流行率也存在显著的空间变异性；尼日利亚北部的疟疾负担更高。调查重要风险因素和地理位置对儿童疟疾的影响与2015 - 2030年可持续发展目标（SDGs）中根除疟疾的议程高度相关，我们相信从本研究中获得的信息以及生成的风险地图可有助于根据气候和环境背景有效地将干预措施针对高风险地区。
