Lane C R, Carville K S, Pierse N, Kelly H A
Epidemiology Unit, Victorian Infectious Disease Reference Laboratory at the Peter Doherty Institute for Infection and Immunity, Melbourne, Australia; National Centre for Epidemiology and Population Health, Australian National University, Canberra, Australia.
National Centre for Epidemiology and Population Health, Australian National University, Canberra, Australia.
Vaccine. 2016 Feb 17;34(8):1070-6. doi: 10.1016/j.vaccine.2016.01.002. Epub 2016 Jan 12.
Influenza vaccine effectiveness (VE) is increasingly estimated using the case-test negative study design. Cases have a symptom complex consistent with influenza and test positive for influenza, while non-cases have the same symptom complex but test negative. We aimed to determine a parsimonious logistic regression model for this study design when applied to patients in the community.
To determine the minimum covariate set required, we used a previously published systematic review to find covariates and restriction criteria commonly included in case-test negative logistic regression models. Covariates were assessed for inclusion using a directed acyclic graph. We used data from the Victorian Influenza Sentinel Practice Network from 2007 to 2013, excluding the pandemic year of 2009, to test the model. VE was estimated as (1-adjusted OR) * 100%. Changes in model fit from addition of specified covariates were examined. Restriction criteria were examined using change in VE estimate. VE was estimated for each year, all years aggregated, and for influenza type and sub-type.
Using publicly available software, the directed acyclic graph indicated that covariates specifying age, time within the influenza season, immunocompromising comorbid conditions and year or study site, where applicable, were required for closure. The inclusion of sex was not required. Inclusions and exclusions were validated when testing the variables (when collected) with our data. Restriction by time between onset and swab was supported by the data. VE for all years aggregated was estimated as 53% (95%CI 38, 64). VE was estimated as 42% (95%CI 19, 59) for H3N2, 75% (95%CI 51, 88) for H1N1pdm09 and 63% (95%CI 38, 79) for influenza B.
Theoretical covariates specified by the directed acyclic graph were validated when tested against surveillance data. A parsimonious model using the case test negative design allows regular estimates of VE and aggregated estimates by year.
越来越多地使用病例 - 检测阴性研究设计来评估流感疫苗效力（VE）。病例具有与流感相符的症状复合体且流感检测呈阳性，而非病例具有相同的症状复合体但检测呈阴性。我们旨在确定将此研究设计应用于社区患者时的简约逻辑回归模型。
为确定所需的最小协变量集，我们利用先前发表的系统评价来查找病例 - 检测阴性逻辑回归模型中通常包含的协变量和限制标准。使用有向无环图评估协变量是否纳入。我们使用了2007年至2013年维多利亚流感哨点实践网络的数据（不包括大流行的2009年）来测试该模型。VE估计为(1 - 调整后的OR) * 100%。检查了添加特定协变量后模型拟合的变化。使用VE估计值的变化来检查限制标准。按年份、所有年份汇总以及流感类型和亚型估计VE。
使用公开可用软件，有向无环图表明，为使模型完整，需要指定年龄、流感季节内的时间、免疫功能低下的合并症以及适用时的年份或研究地点等协变量。不需要纳入性别。在使用我们的数据测试变量（收集时）时，对纳入和排除情况进行了验证。数据支持发病与拭子采集之间的时间限制。所有年份汇总的VE估计为53%（95%CI 38, 64）。H3N2的VE估计为42%（95%CI 19, 59），H1N1pdm09的VE估计为75%（95%CI 51, 88），乙型流感的VE估计为63%（95%CI 38, 79）。
当根据监测数据进行测试时，有向无环图指定的理论协变量得到了验证。使用病例检测阴性设计的简约模型允许定期估计VE并按年份进行汇总估计。
