Using Google Flu Trends data in forecasting influenza-like-illness related ED visits in Omaha, Nebraska
Using Google Flu Trends data in forecasting influenza-like-illness related ED visits in Omaha, Nebraska
复制标题
DOI：
10.1016/j.ajem.2014.05.052
复制
发表时间：
2014-09-01
影响因子：
3.6
通讯作者：
Muelleman, Robert L.
中科院分区：
文献类型：
--
作者：
Araz, Ozgur M.;Bentley, Dan;Muelleman, Robert L.
引言：流感季节期间急诊（艾德）就诊增加。为了建立一个准确的艾德就诊预测模型，必须确定统计学上显著的相关性。预测流感样疾病（ILI）相关的艾德访问可以显着帮助制定强大的资源管理策略，在EDs.Methods：我们首先进行相关性分析，以了解ILI相关的艾德访问的几个预测因素之间的时间相关性。我们使用了内布拉斯加州最大的县道格拉斯县、该州最大的城市奥马哈和奥马哈一家大医院的数据。数据集包括医院的总流感检测结果和阳性流感检测结果（即抗原快速（Ag）和呼吸道合胞病毒感染（RSV）检测）;基于互联网的流感监测系统数据，即内布拉斯加州和奥马哈的Google流感趋势;道格拉斯县因ILI导致的艾德就诊总数;道格拉斯县和内布拉斯加州的ILI监测网络数据作为预测因子，医院ILI相关艾德就诊数据作为因变量。我们使用季节性自回归积分移动平均和Holt Winters方法以及3个线性回归模型来预测ILI相关的艾德就诊，并通过比较均方根误差（RMSE）来评估模型性能。由于2008年至2012年与ILI相关的艾德就诊率呈强正相关，我们验证了Google流感趋势数据在艾德流感监测工具中的预测作用。在我们测试的5个预测模型中，当Google Flu Trends数据作为预测因子时，线性回归模型的表现明显更好。包括Google流感趋势数据作为预测变量的回归模型具有较低的RMSE，并且在我们2013年前5周的预测实验中，当所有其他变量也包括在模型中时，达到最低值（RMSE = 57.61）。谷歌流感趋势数据在统计上提高了预测道格拉斯县ILI相关艾德就诊的性能，这一结果可以推广到其他社区。在流感季节以及大流行爆发期间及时准确地估计艾德量，可以帮助医院相应地规划其艾德资源，并通过优化供应和人员配置来降低成本，还可以通过减少艾德等待时间和过度拥挤来提高服务质量。(C)2014爱思唯尔公司版权所有
Introduction: Emergency department (ED) visits increase during the influenza seasons. It is essential to identify statistically significant correlates in order to develop an accurate forecasting model for ED visits. Forecasting influenza-like-illness (ILI)-related ED visits can significantly help in developing robust resource management strategies at the EDs.Methods: We first performed correlation analyses to understand temporal correlations between several predictors of ILI-related ED visits. We used the data available for Douglas County, the biggest county in Nebraska, for Omaha, the biggest city in the state, and for a major hospital in Omaha. The data set included total and positive influenza test results from the hospital (ie, Antigen rapid (Ag) and Respiratory Syncytial Virus Infection (RSV) tests); an Internet-based influenza surveillance system data, that is, Google Flu Trends, for both Nebraska and Omaha; total ED visits in Douglas County attributable to ILI; and ILI surveillance network data for Douglas County and Nebraska as the predictors and data for the hospital's ILI-related ED visits as the dependent variable. We used Seasonal Autoregressive Integrated Moving Average and Holt Winters methods with3 linear regression models to forecast ILI-related ED visits at the hospital and evaluated model performances by comparing the root means square errors (RMSEs).Results: Because of strong positive correlations with ILI-related ED visits between 2008 and 2012, we validated the use of Google Flu Trends data as a predictor in an ED influenza surveillance tool. Of the 5 forecasting models we have tested, linear regression models performed significantly better when Google Flu Trends data were included as a predictor. Regression models including Google Flu Trends data as a predictor variable have lower RMSE, and the lowest is achieved when all other variables are also included in the model in our forecasting experiments for the first 5 weeks of 2013 (with RMSE = 57.61).Conclusions: Google Flu Trends data statistically improve the performance of predicting ILI-related ED visits in Douglas County, and this result can be generalized to other communities. Timely and accurate estimates of ED volume during the influenza season, as well as during pandemic outbreaks, can help hospitals plan their ED resources accordingly and lower their costs by optimizing supplies and staffing and can improve service quality by decreasing ED wait times and overcrowding. (C) 2014 Elsevier Inc. All rights reserved