Effect of non-pharmaceutical interventions to contain COVID-19 in China.

On 11 March 2020, the World Health Organization (WHO) declared coronavirus disease 2019 (COVID-19) a pandemic1. The strategies based on non-pharmaceutical interventions that were used to contain the outbreak in China appear to be effective2, but quantitative research is still needed to assess the efficacy of non-pharmaceutical interventions and their timings3. Here, using epidemiological data on COVID-19 and anonymized data on human movement4,5, we develop a modelling framework that uses daily travel networks to simulate different outbreak and intervention scenarios across China. We estimate that there were a total of 114,325 cases of COVID-19 (interquartile range 76,776-164,576) in mainland China as of 29 February 2020. Without non-pharmaceutical interventions, we predict that the number of cases would have been 67-fold higher (interquartile range 44-94-fold) by 29 February 2020, and we find that the effectiveness of different interventions varied. We estimate that early detection and isolation of cases prevented more infections than did travel restrictions and contact reductions, but that a combination of non-pharmaceutical interventions achieved the strongest and most rapid effect. According to our model, the lifting of travel restrictions from 17 February 2020 does not lead to an increase in cases across China if social distancing interventions can be maintained, even at a limited level of an on average 25% reduction in contact between individuals that continues until late April. These findings improve our understanding of the effects of non-pharmaceutical interventions on COVID-19, and will inform response efforts across the world.

Mathematical Assessment of the Role of Human Behavior Changes on SARS-CoV-2 Transmission Dynamics in the United States.

The COVID-19 pandemic has not only presented a major global public health and socio-economic crisis, but has also significantly impacted human behavior towards adherence (or lack thereof) to public health intervention and mitigation measures implemented in communities worldwide. This study is based on the use of mathematical modeling approaches to assess the extent to which SARS-CoV-2 transmission dynamics is impacted by population-level changes of human behavior due to factors such as (a) the severity of transmission (such as disease-induced mortality and level of symptomatic transmission), (b) fatigue due to the implementation of mitigation interventions measures (e.g., lockdowns) over a long (extended) period of time, (c) social peer-pressure, among others. A novel behavior-epidemiology model, which takes the form of a deterministic system of nonlinear differential equations, is developed and fitted using observed cumulative SARS-CoV-2 mortality data during the first wave in the United States. The model fits the observed data, as well as makes a more accurate prediction of the observed daily SARS-CoV-2 mortality during the first wave (March 2020-June 2020), in comparison to the equivalent model which does not explicitly account for changes in human behavior. This study suggests that, as more newly-infected individuals become asymptomatically-infectious, the overall level of positive behavior change can be expected to significantly decrease (while new cases may rise, particularly if asymptomatic individuals have higher contact rate, in comparison to symptomatic individuals).

Pediatric and Young Adult Household Transmission of the Initial Waves of SARS-CoV-2 in the United States: Administrative Claims Study.

BACKGROUND: The correlates responsible for the temporal changes of intrahousehold SARS-CoV-2 transmission in the United States have been understudied mainly due to a lack of available surveillance data. Specifically, early analyses of SARS-CoV-2 household secondary attack rates (SARs) were small in sample size and conducted cross-sectionally at single time points. From these limited data, it has been difficult to assess the role that different risk factors have had on intrahousehold disease transmission in different stages of the ongoing COVID-19 pandemic, particularly in children and youth. OBJECTIVE: This study aimed to estimate the transmission dynamic and infectivity of SARS-CoV-2 among pediatric and young adult index cases (age 0 to 25 years) in the United States through the initial waves of the pandemic. METHODS: Using administrative claims, we analyzed 19 million SARS-CoV-2 test records between January 2020 and February 2021. We identified 36,241 households with pediatric index cases and calculated household SARs utilizing complete case information. Using a retrospective cohort design, we estimated the household SARS-CoV-2 transmission between 4 index age groups (0 to 4 years, 5 to 11 years, 12 to 17 years, and 18 to 25 years) while adjusting for sex, family size, quarter of first SARS-CoV-2 positive record, and residential regions of the index cases. RESULTS: After filtering all household records for greater than one member in a household and missing information, only 36,241 (0.85%) of 4,270,130 households with a pediatric case remained in the analysis. Index cases aged between 0 and 17 years were a minority of the total index cases (n=11,484, 11%). The overall SAR of SARS-CoV-2 was 23.04% (95% CI 21.88-24.19). As a comparison, the SAR for all ages (0 to 65+ years) was 32.4% (95% CI 32.1-32.8), higher than the SAR for the population between 0 and 25 years of age. The highest SAR of 38.3% was observed in April 2020 (95% CI 31.6-45), while the lowest SAR of 15.6% was observed in September 2020 (95% CI 13.9-17.3). It consistently decreased from 32% to 21.1% as the age of index groups increased. In a multiple logistic regression analysis, we found that the youngest pediatric age group (0 to 4 years) had 1.69 times (95% CI 1.42-2.00) the odds of SARS-CoV-2 transmission to any family members when compared with the oldest group (18 to 25 years). Family size was significantly associated with household viral transmission (odds ratio 2.66, 95% CI 2.58-2.74). CONCLUSIONS: Using retrospective claims data, the pediatric index transmission of SARS-CoV-2 during the initial waves of the COVID-19 pandemic in the United States was associated with location and family characteristics. Pediatric SAR (0 to 25 years) was less than the SAR for all age other groups. Less than 1% (n=36,241) of all household data were retained in the retrospective study for complete case analysis, perhaps biasing our findings. We have provided measures of baseline household pediatric transmission for tracking and comparing the infectivity of later SARS-CoV-2 variants.

Leveraging Serosurveillance and Postmortem Surveillance to Quantify the Impact of Coronavirus Disease 2019 in Africa.

BACKGROUND: The coronavirus disease 2019 (COVID-19) pandemic has had a devastating impact on global health, the magnitude of which appears to differ intercontinentally: For example, reports suggest that 271 900 per million people have been infected in Europe versus 8800 per million people in Africa. While Africa is the second-largest continent by population, its reported COVID-19 cases comprise <3% of global cases. Although social and environmental explanations have been proposed to clarify this discrepancy, systematic underascertainment of infections may be equally responsible. METHODS: We sought to quantify magnitudes of underascertainment in COVID-19's cumulative incidence in Africa. Using serosurveillance and postmortem surveillance, we constructed multiplicative factors estimating ratios of true infections to reported cases in Africa since March 2020. RESULTS: Multiplicative factors derived from serology data (subset of 12 nations) suggested a range of COVID-19 reporting rates, from 1 in 2 infections reported in Cape Verde (July 2020) to 1 in 3795 infections reported in Malawi (June 2020). A similar set of multiplicative factors for all nations derived from postmortem data points toward the same conclusion: Reported COVID-19 cases are unrepresentative of true infections, suggesting that a key reason for low case burden in many African nations is significant underdetection and underreporting. CONCLUSIONS: While estimating the exact burden of COVID-19 is challenging, the multiplicative factors we present furnish incidence estimates reflecting likely-to-worst-case ranges of infection. Our results stress the need for expansive surveillance to allocate resources in areas experiencing discrepancies between reported cases, projected infections, and deaths.

Near real-time surveillance of the SARS-CoV-2 epidemic with incomplete data.

When responding to infectious disease outbreaks, rapid and accurate estimation of the epidemic trajectory is critical. However, two common data collection problems affect the reliability of the epidemiological data in real time: missing information on the time of first symptoms, and retrospective revision of historical information, including right censoring. Here, we propose an approach to construct epidemic curves in near real time that addresses these two challenges by 1) imputation of dates of symptom onset for reported cases using a dynamically-estimated "backward" reporting delay conditional distribution, and 2) adjustment for right censoring using the NobBS software package to nowcast cases by date of symptom onset. This process allows us to obtain an approximation of the time-varying reproduction number (Rt) in real time. We apply this approach to characterize the early SARS-CoV-2 outbreak in two Spanish regions between March and April 2020. We evaluate how these real-time estimates compare with more complete epidemiological data that became available later. We explore the impact of the different assumptions on the estimates, and compare our estimates with those obtained from commonly used surveillance approaches. Our framework can help improve accuracy, quantify uncertainty, and evaluate frequently unstated assumptions when recovering the epidemic curves from limited data obtained from public health systems in other locations.

A nowcasting framework for correcting for reporting delays in malaria surveillance.

Time lags in reporting to national surveillance systems represent a major barrier for the control of infectious diseases, preventing timely decision making and resource allocation. This issue is particularly acute for infectious diseases like malaria, which often impact rural and remote communities the hardest. In Guyana, a country located in South America, poor connectivity among remote malaria-endemic regions hampers surveillance efforts, making reporting delays a key challenge for elimination. Here, we analyze 13 years of malaria surveillance data, identifying key correlates of time lags between clinical cases occurring and being added to the central data system. We develop nowcasting methods that use historical patterns of reporting delays to estimate occurred-but-not-reported monthly malaria cases. To assess their performance, we implemented them retrospectively, using only information that would have been available at the time of estimation, and found that they substantially enhanced the estimates of malaria cases. Specifically, we found that the best performing models achieved up to two-fold improvements in accuracy (or error reduction) over known cases in selected regions. Our approach provides a simple, generalizable tool to improve malaria surveillance in endemic countries and is currently being implemented to help guide existing resource allocation and elimination efforts.

Estimating the cumulative incidence of COVID-19 in the United States using influenza surveillance, virologic testing, and mortality data: Four complementary approaches.

Effectively designing and evaluating public health responses to the ongoing COVID-19 pandemic requires accurate estimation of the prevalence of COVID-19 across the United States (US). Equipment shortages and varying testing capabilities have however hindered the usefulness of the official reported positive COVID-19 case counts. We introduce four complementary approaches to estimate the cumulative incidence of symptomatic COVID-19 in each state in the US as well as Puerto Rico and the District of Columbia, using a combination of excess influenza-like illness reports, COVID-19 test statistics, COVID-19 mortality reports, and a spatially structured epidemic model. Instead of relying on the estimate from a single data source or method that may be biased, we provide multiple estimates, each relying on different assumptions and data sources. Across our four approaches emerges the consistent conclusion that on April 4, 2020, the estimated case count was 5 to 50 times higher than the official positive test counts across the different states. Nationally, our estimates of COVID-19 symptomatic cases as of April 4 have a likely range of 2.3 to 4.8 million, with possibly as many as 7.6 million cases, up to 25 times greater than the cumulative confirmed cases of about 311,000. Extending our methods to May 16, 2020, we estimate that cumulative symptomatic incidence ranges from 4.9 to 10.1 million, as opposed to 1.5 million positive test counts. The proposed combination of approaches may prove useful in assessing the burden of COVID-19 during resurgences in the US and other countries with comparable surveillance systems.

Real-time estimation of disease activity in emerging outbreaks using internet search information.

Understanding the behavior of emerging disease outbreaks in, or ahead of, real-time could help healthcare officials better design interventions to mitigate impacts on affected populations. Most healthcare-based disease surveillance systems, however, have significant inherent reporting delays due to data collection, aggregation, and distribution processes. Recent work has shown that machine learning methods leveraging a combination of traditionally collected epidemiological information and novel Internet-based data sources, such as disease-related Internet search activity, can produce meaningful "nowcasts" of disease incidence ahead of healthcare-based estimates, with most successful case studies focusing on endemic and seasonal diseases such as influenza and dengue. Here, we apply similar computational methods to emerging outbreaks in geographic regions where no historical presence of the disease of interest has been observed. By combining limited available historical epidemiological data available with disease-related Internet search activity, we retrospectively estimate disease activity in five recent outbreaks weeks ahead of traditional surveillance methods. We find that the proposed computational methods frequently provide useful real-time incidence estimates that can help fill temporal data gaps resulting from surveillance reporting delays. However, the proposed methods are limited by issues of sample bias and skew in search query volumes, perhaps as a result of media coverage.

Gender-specificity of resilience in major depressive disorder.

INTRODUCTION: The major stressors associated with the COVID-19 pandemic provide an opportunity to understand the extent to which protective factors against depression may exhibit gender-specificity. METHOD: This study examined responses from multiple waves of a 50 states non-probability internet survey conducted between May 2020 and January 2021. Participants completed the PHQ-9 as a measure of depression, as well as items characterizing social supports. We used logistic regression models with population reweighting to examine association between absence of even mild depressive symptoms and sociodemographic features and social supports, with interaction terms and stratification used to investigate sex-specificity. RESULTS: Among 73,917 survey respondents, 31,199 (42.2%) reported absence of mild or greater depression-11,011/23,682 males (46.5%) and 20,188/50,235 (40.2%) females. In a regression model, features associated with greater likelihood of depression-resistance included at least weekly attendance of religious services (odds ratio [OR]: 1.10, 95% confidence interval [CI]: 1.04-1.16) and greater trust in others (OR: 1.04 for a 2-unit increase, 95% CI: 1.02-1.06), along with level of social support measured as number of social ties available who could provide care (OR: 1.05, 95% CI: 1.02-1.07), talk to them (OR: 1.10, 95% CI: 1.07-1.12), and help with employment (OR: 1.06, 95% CI: 1.04-1.08). The first two features showed significant interaction with gender (p < .0001), with markedly greater protective effects among women. CONCLUSION: Aspects of social support are associated with diminished risk of major depressive symptoms, with greater effects of religious service attendance and trust in others observed among women than men.

Avoidable Serum Potassium Testing in the Cardiac ICU: Development and Testing of a Machine-Learning Model.

OBJECTIVES: To create a machine-learning model identifying potentially avoidable blood draws for serum potassium among pediatric patients following cardiac surgery. DESIGN: Retrospective cohort study. SETTING: Tertiary-care center. PATIENTS: All patients admitted to the cardiac ICU at Boston Children's Hospital between January 2010 and December 2018 with a length of stay greater than or equal to 4 days and greater than or equal to two recorded serum potassium measurements. INTERVENTIONS: None. MEASUREMENTS AND MAIN RESULTS: We collected variables related to potassium homeostasis, including serum chemistry, hourly potassium intake, diuretics, and urine output. Using established machine-learning techniques, including random forest classifiers, and hyperparameter tuning, we created models predicting whether a patient's potassium would be normal or abnormal based on the most recent potassium level, medications administered, urine output, and markers of renal function. We developed multiple models based on different age-categories and temporal proximity of the most recent potassium measurement. We assessed the predictive performance of the models using an independent test set. Of the 7,269 admissions (6,196 patients) included, serum potassium was measured on average of 1 (interquartile range, 0-1) time per day. Approximately 96% of patients received at least one dose of IV diuretic and 83% received a form of potassium supplementation. Our models predicted a normal potassium value with a median positive predictive value of 0.900. A median percentage of 2.1% measurements (mean 2.5%; interquartile range, 1.3-3.7%) was incorrectly predicted as normal when they were abnormal. A median percentage of 0.0% (interquartile range, 0.0-0.4%) critically low or high measurements was incorrectly predicted as normal. A median of 27.2% (interquartile range, 7.8-32.4%) of samples was correctly predicted to be normal and could have been potentially avoided. CONCLUSIONS: Machine-learning methods can be used to predict avoidable blood tests accurately for serum potassium in critically ill pediatric patients. A median of 27.2% of samples could have been saved, with decreased costs and risk of infection or anemia.

Enhancing Situational Awareness to Prevent Infectious Disease Outbreaks from Becoming Catastrophic.

Catastrophic epidemics, if they occur, will very likely start from localized and far smaller (non-catastrophic) outbreaks that grow into much greater threats. One key bulwark against this outcome is the ability of governments and the health sector more generally to make informed decisions about control measures based on accurate understanding of the current and future extent of the outbreak. Situation reporting is the activity of periodically summarizing the state of the outbreak in a (usually) public way. We delineate key classes of decisions whose quality depends on high-quality situation reporting, key quantities for which estimates are needed to inform these decisions, and the traditional and novel sources of data that can aid in estimating these quantities. We emphasize the important role of situation reports as providing public, shared planning assumptions that allow decision makers to harmonize the response while making explicit the uncertainties that underlie the scenarios outlined for planning. In this era of multiple data sources and complex factors informing the interpretation of these data sources, we describe four principles for situation reporting: (1) Situation reporting should be thematic, concentrating on essential areas of evidence needed for decisions. (2) Situation reports should adduce evidence from multiple sources to address each area of evidence, along with expert assessments of key parameters. (3) Situation reports should acknowledge uncertainty and attempt to estimate its magnitude for each assessment. (4) Situation reports should contain carefully curated visualizations along with text and tables.

COVID-19: US federal accountability for entry, spread, and inequities-lessons for the future.

The United States (US) has been among those nations most severely affected by the first-and subsequent-phases of the pandemic of COVID-19, the disease caused by SARS-CoV-2. With only 4% of the worldwide population, the US has seen about 22% of COVID-19 deaths. Despite formidable advantages in resources and expertise, presently the per capita mortality rate is over 585/million, respectively 2.4 and 5 times higher compared to Canada and Germany. As we enter Fall 2020, the US is enduring ongoing outbreaks across large regions of the country. Moreover, within the US, an early and persistent feature of the pandemic has been the disproportionate impact on populations already made vulnerable by racism and dangerous jobs, inadequate wages, and unaffordable housing, and this is true for both the headline public health threat and the additional disastrous economic impacts. In this article we assess the impact of missteps by the Federal Government in three specific areas: the introduction of the virus to the US and the establishment of community transmission; the lack of national COVID-19 workplace standards and enforcement, and lack of personal protective equipment (PPE) for workplaces as represented by complaints to the Occupational Safety and Health Administration (OSHA) which we find are correlated with deaths 16 days later (ρ = 0.83); and the total excess deaths in 2020 to date already total more than 230,000, while COVID-19 mortality rates exhibit severe-and rising-inequities in race/ethnicity, including among working age adults.

Real-Time Forecasting of the COVID-19 Outbreak in Chinese Provinces: Machine Learning Approach Using Novel Digital Data and Estimates From Mechanistic Models.

BACKGROUND: The inherent difficulty of identifying and monitoring emerging outbreaks caused by novel pathogens can lead to their rapid spread; and if left unchecked, they may become major public health threats to the planet. The ongoing coronavirus disease (COVID-19) outbreak, which has infected over 2,300,000 individuals and caused over 150,000 deaths, is an example of one of these catastrophic events. OBJECTIVE: We present a timely and novel methodology that combines disease estimates from mechanistic models and digital traces, via interpretable machine learning methodologies, to reliably forecast COVID-19 activity in Chinese provinces in real time. METHODS: Our method uses the following as inputs: (a) official health reports, (b) COVID-19-related internet search activity, (c) news media activity, and (d) daily forecasts of COVID-19 activity from a metapopulation mechanistic model. Our machine learning methodology uses a clustering technique that enables the exploitation of geospatial synchronicities of COVID-19 activity across Chinese provinces and a data augmentation technique to deal with the small number of historical disease observations characteristic of emerging outbreaks. RESULTS: Our model is able to produce stable and accurate forecasts 2 days ahead of the current time and outperforms a collection of baseline models in 27 out of 32 Chinese provinces. CONCLUSIONS: Our methodology could be easily extended to other geographies currently affected by COVID-19 to aid decision makers with monitoring and possibly prevention.

Correction: Real-Time Forecasting of the COVID-19 Outbreak in Chinese Provinces: Machine Learning Approach Using Novel Digital Data and Estimates From Mechanistic Models.

[This corrects the article DOI: 10.2196/20285.].

Rates of increase of antibiotic resistance and ambient temperature in Europe: a cross-national analysis of 28 countries between 2000 and 2016.

BackgroundThe rapid increase of bacterial antibiotic resistance could soon render our most effective method to address infections obsolete. Factors influencing pathogen resistance prevalence in human populations remain poorly described, though temperature is known to contribute to mechanisms of spread.AimTo quantify the role of temperature, spatially and temporally, as a mechanistic modulator of transmission of antibiotic resistant microbes.MethodsAn ecologic analysis was performed on country-level antibiotic resistance prevalence in three common bacterial pathogens across 28 European countries, collectively representing over 4 million tested isolates. Associations of minimum temperature and other predictors with change in antibiotic resistance rates over 17 years (2000-2016) were evaluated with multivariable models. The effects of predictors on the antibiotic resistance rate change across geographies were quantified.ResultsDuring 2000-2016, for Escherichia coli and Klebsiella pneumoniae, European countries with 10°C warmer ambient minimum temperatures compared to others, experienced more rapid resistance increases across all antibiotic classes. Increases ranged between 0.33%/year (95% CI: 0.2 to 0.5) and 1.2%/year (95% CI: 0.4 to 1.9), even after accounting for recognised resistance drivers including antibiotic consumption and population density. For Staphylococcus aureus a decreasing relationship of -0.4%/year (95% CI: -0.7 to 0.0) was found for meticillin resistance, reflecting widespread declines in meticillin-resistant S. aureus across Europe over the study period.ConclusionWe found evidence of a long-term effect of ambient minimum temperature on antibiotic resistance rate increases in Europe. Ambient temperature might considerably influence antibiotic resistance growth rates, and explain geographic differences observed in cross-sectional studies. Rising temperatures globally may hasten resistance spread, complicating mitigation efforts.

Advances in using Internet searches to track dengue.

Dengue is a mosquito-borne disease that threatens over half of the world's population. Despite being endemic to more than 100 countries, government-led efforts and tools for timely identification and tracking of new infections are still lacking in many affected areas. Multiple methodologies that leverage the use of Internet-based data sources have been proposed as a way to complement dengue surveillance efforts. Among these, dengue-related Google search trends have been shown to correlate with dengue activity. We extend a methodological framework, initially proposed and validated for flu surveillance, to produce near real-time estimates of dengue cases in five countries/states: Mexico, Brazil, Thailand, Singapore and Taiwan. Our result shows that our modeling framework can be used to improve the tracking of dengue activity in multiple locations around the world.

Comparison of crowd-sourced, electronic health records based, and traditional health-care based influenza-tracking systems at multiple spatial resolutions in the United States of America.

BACKGROUND: Influenza causes an estimated 3000 to 50,000 deaths per year in the United States of America (US). Timely and representative data can help local, state, and national public health officials monitor and respond to outbreaks of seasonal influenza. Data from cloud-based electronic health records (EHR) and crowd-sourced influenza surveillance systems have the potential to provide complementary, near real-time estimates of influenza activity. The objectives of this paper are to compare two novel influenza-tracking systems with three traditional healthcare-based influenza surveillance systems at four spatial resolutions: national, regional, state, and city, and to determine the minimum number of participants in these systems required to produce influenza activity estimates that resemble the historical trends recorded by traditional surveillance systems. METHODS: We compared influenza activity estimates from five influenza surveillance systems: 1) patient visits for influenza-like illness (ILI) from the US Outpatient ILI Surveillance Network (ILINet), 2) virologic data from World Health Organization (WHO) Collaborating and National Respiratory and Enteric Virus Surveillance System (NREVSS) Laboratories, 3) Emergency Department (ED) syndromic surveillance from Boston, Massachusetts, 4) patient visits for ILI from EHR, and 5) reports of ILI from the crowd-sourced system, Flu Near You (FNY), by calculating correlations between these systems across four influenza seasons, 2012-16, at four different spatial resolutions in the US. For the crowd-sourced system, we also used a bootstrapping statistical approach to estimate the minimum number of reports necessary to produce a meaningful signal at a given spatial resolution. RESULTS: In general, as the spatial resolution increased, correlation values between all influenza surveillance systems decreased. Influenza-like Illness rates in geographic areas with more than 250 crowd-sourced participants or with more than 20,000 visit counts for EHR tracked government-lead estimates of influenza activity. CONCLUSIONS: With a sufficient number of reports, data from novel influenza surveillance systems can complement traditional healthcare-based systems at multiple spatial resolutions.

Accurate estimation of influenza epidemics using Google search data via ARGO.

Accurate real-time tracking of influenza outbreaks helps public health officials make timely and meaningful decisions that could save lives. We propose an influenza tracking model, ARGO (AutoRegression with GOogle search data), that uses publicly available online search data. In addition to having a rigorous statistical foundation, ARGO outperforms all previously available Google-search-based tracking models, including the latest version of Google Flu Trends, even though it uses only low-quality search data as input from publicly available Google Trends and Google Correlate websites. ARGO not only incorporates the seasonality in influenza epidemics but also captures changes in people's online search behavior over time. ARGO is also flexible, self-correcting, robust, and scalable, making it a potentially powerful tool that can be used for real-time tracking of other social events at multiple temporal and spatial resolutions.

Using electronic health records and Internet search information for accurate influenza forecasting.

BACKGROUND: Accurate influenza activity forecasting helps public health officials prepare and allocate resources for unusual influenza activity. Traditional flu surveillance systems, such as the Centers for Disease Control and Prevention's (CDC) influenza-like illnesses reports, lag behind real-time by one to 2 weeks, whereas information contained in cloud-based electronic health records (EHR) and in Internet users' search activity is typically available in near real-time. We present a method that combines the information from these two data sources with historical flu activity to produce national flu forecasts for the United States up to 4 weeks ahead of the publication of CDC's flu reports. METHODS: We extend a method originally designed to track flu using Google searches, named ARGO, to combine information from EHR and Internet searches with historical flu activities. Our regularized multivariate regression model dynamically selects the most appropriate variables for flu prediction every week. The model is assessed for the flu seasons within the time period 2013-2016 using multiple metrics including root mean squared error (RMSE). RESULTS: Our method reduces the RMSE of the publicly available alternative (Healthmap flutrends) method by 33, 20, 17 and 21%, for the four time horizons: real-time, one, two, and 3 weeks ahead, respectively. Such accuracy improvements are statistically significant at the 5% level. Our real-time estimates correctly identified the peak timing and magnitude of the studied flu seasons. CONCLUSIONS: Our method significantly reduces the prediction error when compared to historical publicly available Internet-based prediction systems, demonstrating that: (1) the method to combine data sources is as important as data quality; (2) effectively extracting information from a cloud-based EHR and Internet search activity leads to accurate forecast of flu.

Combining Search, Social Media, and Traditional Data Sources to Improve Influenza Surveillance.

We present a machine learning-based methodology capable of providing real-time ("nowcast") and forecast estimates of influenza activity in the US by leveraging data from multiple data sources including: Google searches, Twitter microblogs, nearly real-time hospital visit records, and data from a participatory surveillance system. Our main contribution consists of combining multiple influenza-like illnesses (ILI) activity estimates, generated independently with each data source, into a single prediction of ILI utilizing machine learning ensemble approaches. Our methodology exploits the information in each data source and produces accurate weekly ILI predictions for up to four weeks ahead of the release of CDC's ILI reports. We evaluate the predictive ability of our ensemble approach during the 2013-2014 (retrospective) and 2014-2015 (live) flu seasons for each of the four weekly time horizons. Our ensemble approach demonstrates several advantages: (1) our ensemble method's predictions outperform every prediction using each data source independently, (2) our methodology can produce predictions one week ahead of GFT's real-time estimates with comparable accuracy, and (3) our two and three week forecast estimates have comparable accuracy to real-time predictions using an autoregressive model. Moreover, our results show that considerable insight is gained from incorporating disparate data streams, in the form of social media and crowd sourced data, into influenza predictions in all time horizons.