Accurate Influenza Monitoring and Forecasting Using Novel Internet Data Streams: A Case Study in the Boston Metropolis

Abstract

BackgroundInfluenza outbreaks pose major challenges to public health around the world, leading to thousands of deaths a year in the United States alone. Accurate systems that track influenza activity at the city level are necessary to provide actionable information that can be used for clinical, hospital, and community outbreak preparation.ObjectiveAlthough Internet-based real-time data sources such as Google searches and tweets have been successfully used to produce influenza activity estimates ahead of traditional health care–based systems at national and state levels, influenza tracking and forecasting at finer spatial resolutions, such as the city level, remain an open question. Our study aimed to present a precise, near real-time methodology capable of producing influenza estimates ahead of those collected and published by the Boston Public Health Commission (BPHC) for the Boston metropolitan area. This approach has great potential to be extended to other cities with access to similar data sources.MethodsWe first tested the ability of Google searches, Twitter posts, electronic health records, and a crowd-sourced influenza reporting system to detect influenza activity in the Boston metropolis separately. We then adapted a multivariate dynamic regression method named ARGO (autoregression with general online information), designed for tracking influenza at the national level, and showed that it effectively uses the above data sources to monitor and forecast influenza at the city level 1 week ahead of the current date. Finally, we presented an ensemble-based approach capable of combining information from models based on multiple data sources to more robustly nowcast as well as forecast influenza activity in the Boston metropolitan area. The performances of our models were evaluated in an out-of-sample fashion over 4 influenza seasons within 2012-2016, as well as a holdout validation period from 2016 to 2017.ResultsOur ensemble-based methods incorporating information from diverse models based on multiple data sources, including ARGO, produced the most robust and accurate results. The observed Pearson correlations between our out-of-sample flu activity estimates and those historically reported by the BPHC were 0.98 in nowcasting influenza and 0.94 in forecasting influenza 1 week ahead of the current date.ConclusionsWe show that information from Internet-based data sources, when combined using an informed, robust methodology, can be effectively used as early indicators of influenza activity at fine geographic resolutions.

Improved state-level influenza activity nowcasting in the United States leveraging Internet-based data sources and network approaches via ARGONet

“…Previous work has shown the versatility of ARGO, one of the component models in our ensemble, 152 over a variety of disease estimation scenarios [5,[20][21][22]. At the state level, where it had not been applied 153 before, it clearly outperformed existing benchmarks over the study period, namely Google Flu Trends 154 and an autoregression.…”

“…Established influenza reporting systems currently exist over large geographic scales in 18 the United States, coordinated by the Centers for Disease Control and Prevention (CDC). These systems 19 provide weekly reports of influenza statistics, aggregated over the national, regional (10 groups as defined by 20 the Health and Human Services), and starting in fall 2017, state level. Of particular interest, U.S. Outpatient 21 Influenza-like Illness Surveillance Network (ILINet) records the percentage of patients reporting to outpatient 22 clinics with symptoms of influenza-like illness (ILI), which is defined by fever over 100°F in addition to sore 23 throat or cough, over the total number of patient visits [4].…”

“…State-level ARGO models outperform existing benchmarks 63 We first adapted the ARGO (AutoRegression with General Online information) methodology to state-level 64 flu detection. ARGO has previously demonstrated the ability to infer flu activity with high precision over 65 a variety of geographical areas and scales [5,19,20]. The adapted model dynamically fits a regularized 66 multivariable regression on state-level Google search engine frequencies, electronic health record reports from 67 athenahealth, and historical CDC %ILI estimates (see Methods section).…”

“…• Filtering features by correlation: For each week, non-autoregressive features ranked outside the top 10 250 by correlation were removed to reduce noise from poor predictors. Based on previous research, this 251 complementary feature selection process benefits the performance of lasso, which can be unstable in 252 variable selection [20,21].…”

Estimating influenza incidence using search query deceptiveness and generalized ridge regression

“…For example, Brooks et al's top-performing entry [4] to the CDC's flu forecasting challenge [9] was improved by nowcasting. Lu et al tested autoregressive nowcasts using several internet data sources and found that they improved 1-week-ahead forecasts [23]. Kandula et al measured the benefit of nowcasting to their flu forecasting model at 8-35% [21].…”