Incidence estimation from sentinel surveillance data; a simulation study and application to data from the Belgian laboratory sentinel surveillance
Incidence estimation from sentinel surveillance data; a simulation study and application to data from the Belgian laboratory sentinel surveillance
复制标题
DOI：
10.1186/s12889-019-7279-y
复制
发表时间：
2019-07-23
影响因子：
4.5
通讯作者：
Hens, Niel
中科院分区：
文献类型：
--
作者：
Braeye, Toon;Quoilin, Sophie;Hens, Niel
背景反向概率加权(IPW)方法可以用来估计通过哨点监测收集的样本中的病例总数。这些方法的核心是逆权重，可以用几种方法得出，在这种情况下，逆权重代表实验室(LAB)哨点监测发现实验室确认病例的概率。权重是根据所有实验室的参与实验室比例得出的。我们根据实验室对人口的吸引力和密度来调整这些权重。计量经济学赫夫模型估计的哨兵实验室的市场份额也被考虑在内。此外，我们调查了当哨兵实验室没有报告病例时，不将其识别为哨兵实验室的影响。我们估计了与不同权重相关的偏差，即模拟病例数量与哨兵样本估计总数之间的差异。作为激励数据示例，我们将扩展的Huff模型应用于2010至2015年间比利时实验室哨兵监测下的四种病原体，并讨论了模型的拟合。我们估计与轮状病毒、流感病毒、小肠结肠炎耶尔森菌和弯曲杆菌有关的实验室确诊病例总数。扩展的Huff模型考虑了实验室的概念、报销数量和部门数量、实验室密度、区域边界、距离和实验室之间的竞争。在数据实例中，确定了几个显著的系数，但Huff模型与比利时哨点监测数据的拟合使得市场份额的许多变化没有得到解释。结论Huff模型可以估计哨点监测的空间和人口覆盖率，并通过IPW方法估计病例总数。Huff-Models重力功能使我们能够在根据完整数据集进行估计的同时区分区域内部。我们的数据例子表明，为了更准确地估计，关于实验室参与监测和实践的额外数据是必要的。
BackgroundInverse probability weighting (IPW) methods can be used to estimate the total number of cases from the sample collected through sentinel surveillance. Central to these methods are the inverse weights which can be derived in several ways and, in this case, represent the probability that laboratory (lab) sentinel surveillance detects a lab-confirmed case.MethodsWe compare different weights in a simulation study. Weights are obtained from the proportion of participating labs over all labs. We adjust these weights for attractiveness and density of labs over population. The market share of sentinel labs, as estimated by the econometric Huff-model, is also considered. Additionally, we investigate the effect of not recognizing sentinel labs as sentinel labs when they report no cases. We estimate the bias associated with the different weights as the difference between the simulated number of cases and the estimate of this total from the sentinel sample.As motivating data examples, we apply an extended Huff-model to four pathogens under laboratory sentinel surveillance in Belgium between 2010 and 2015 and discuss the model fit. We estimate the total number of lab-confirmed cases associated with Rotavirus, influenza virus, Y. enterocolitica and Campylobacter spp.. The extended Huff-model takes the lab-concept, the number of reimbursements and the number of departments, lab-density, regional borders, distance and competition between labs in account.ResultsEstimates obtained with the Huff-model were most accurate in the more complex simulation scenarios as compared to other weights. In the data examples, several significant coefficients are identified, but the fit of the Huff-model to the Belgian sentinel surveillance data leaves much variability in market shares unexplained.ConclusionThe Huff-model allows for estimation of the spatial and population coverage of sentinel surveillance and through IPW-methods also for the estimation of the total number of cases. The Huff-models gravity function allows us to differentiate inside an area while estimating from the full dataset. Our data examples show that additional data on the participation to surveillance and practices of labs is necessary for a more accurate estimation.