Flu Trends Research Articles

Optimizing Provider Recruitment for Influenza Surveillance Networks

The increasingly complex and rapid transmission dynamics of many infectious diseases necessitates the use of new, more advanced methods for surveillance, early detection, and decision-making. Here, we demonstrate that a new method for optimizing surveillance networks can improve the quality of epidemiological information produced by typical provider-based networks. Using past surveillance and Internet search data, it determines the precise locations where providers should be enrolled. When applied to redesigning the provider-based, influenza-like-illness surveillance network (ILINet) for the state of Texas, the method identifies networks that are expected to significantly outperform the existing network with far fewer providers. This optimized network avoids informational redundancies and is thereby more effective than networks designed by conventional methods and a recently published algorithm based on maximizing population coverage. We show further that Google Flu Trends data, when incorporated into a network as a virtual provider, can enhance but not replace traditional surveillance methods.

Google Flu Trends: Correlation With Emergency Department Influenza Rates and Crowding Metrics

Google Flu Trends (GFT) is a novel Internet-based influenza surveillance system that uses search engine query data to estimate influenza activity and is available in near real time. This study assesses the temporal correlation of city GFT data to cases of influenza and standard crowding indices from an inner-city emergency department (ED). This study was performed during a 21-month period (from January 2009 through October 2010) at an urban academic hospital with physically and administratively separate adult and pediatric EDs. We collected weekly data from GFT for Baltimore, Maryland; ED Centers for Disease Control and Prevention-reported standardized influenzalike illness (ILI) data; laboratory-confirmed influenza data; and ED crowding indices (patient volume, number of patients who left without being seen, waiting room time, and length of stay for admitted and discharged patients). Pediatric and adult data were analyzed separately using cross-correlation with GFT. GFT correlated with both number of positive influenza test results (adult ED, r = 0.876; pediatric ED, r = 0.718) and number of ED patients presenting with ILI (adult ED, r = 0.885; pediatric ED, r = 0.652). Pediatric but not adult crowding measures, such as total ED volume (r = 0.649) and leaving without being seen (r = 0.641), also had good correlation with GFT. Adult crowding measures for low-acuity patients, such as waiting room time (r = 0.421) and length of stay for discharged patients (r = 0.548), had moderate correlation with GFT. City-level GFT shows strong correlation with influenza cases and ED ILI visits, validating its use as an ED surveillance tool. GFT correlated with several pediatric ED crowding measures and those for low-acuity adult patients.

Google flu trends: correlation with emergency department influenza rates and crowding metrics

Background. Google Flu Trends (GFT) is a novel Internet-based influenza surveillance system that uses search engine query data to estimate influenza activity and is available in near real time. This study assesses the temporal correlation of city GFT data to cases of influenza and standard crowding indices from an inner-city emergency department (ED). Methods. This study was performed during a 21-month period (from January 2009 through October 2010) at an urban academic hospital with physically and administratively separate adult and pediatric EDs. We collected weekly data from GFT for Baltimore, Maryland; ED Centers for Disease Control and Prevention‐reported standardized influenzalike illness (ILI) data; laboratory-confirmed influenza data; and ED crowding indices (patient volume, number of patients who left without being seen, waiting room time, and length of stay for admitted and discharged patients). Pediatric and adult data were analyzed separately using cross-correlation with GFT. Results. GFT correlated with both number of positive influenza test results (adult ED, r 5 0.876; pediatric ED, r 5 0.718) and number of ED patients presenting with ILI (adult ED, r 5 0.885; pediatric ED, r 5 0.652). Pediatric but not adult crowding measures, such as total ED volume (r 5 0.649) and leaving without being seen (r 5 0.641), also had good correlation with GFT. Adult crowding measures for low-acuity patients, such as waiting room time (r 5 0.421) and length of stay for discharged patients (r 5 0.548), had moderate correlation with GFT. Conclusions. City-level GFT shows strong correlation with influenza cases and ED ILI visits, validating its use as an ED surveillance tool. GFT correlated with several pediatric ED crowding measures and those for low-acuity adult patients.

Correlation of "Google Flu Trends" with Sentinel Surveillance Data for Influenza in 2009 in Japan

Google Flu Trends (http://www.google.org/flutrends/) (GFT) aggregates Google search data to estimate flu activity in 28 countries. This study explored the correlation of GFT with Japanese national sentinel surveillance data in 2009. We obtained GFT and national sentinel surveillance data for influenza in all 47 Japanese prefectures from 29 June- 31 Dec 2009. Pearson correlation coefficients were calculated between GFT and sentinel surveillance data in each prefecture. Multiple regression analysis was also used to determine associations between them. Correlation coefficients were greater than 0.9 between GFT and the sentinel surveillance data in Tokyo, Kanagawa, Aichi, Osaka, and Hyogo, which have relatively large populations. Peaks for GFT were 2 weeks earlier than sentinel surveillance in these prefectures. Multiple regressions showed that only Tokyo (r = 0.97; P < 0.01) was significantly associated with GFT. People's interest in influenza affects GFT data, which could reflect influenza spread in mega-cities such as Tokyo. Further development of technology to identify prefecture-wide epidemic patterns could offer more accurate information on influenza spread in Japan.

Assessing Google Flu Trends Performance in the United States during the 2009 Influenza Virus A (H1N1) Pandemic

BackgroundGoogle Flu Trends (GFT) uses anonymized, aggregated internet search activity to provide near-real time estimates of influenza activity. GFT estimates have shown a strong correlation with official influenza surveillance data. The 2009 influenza virus A (H1N1) pandemic [pH1N1] provided the first opportunity to evaluate GFT during a non-seasonal influenza outbreak. In September 2009, an updated United States GFT model was developed using data from the beginning of pH1N1.Methodology/Principal FindingsWe evaluated the accuracy of each U.S. GFT model by comparing weekly estimates of ILI (influenza-like illness) activity with the U.S. Outpatient Influenza-like Illness Surveillance Network (ILINet). For each GFT model we calculated the correlation and RMSE (root mean square error) between model estimates and ILINet for four time periods: pre-H1N1, Summer H1N1, Winter H1N1, and H1N1 overall (Mar 2009–Dec 2009). We also compared the number of queries, query volume, and types of queries (e.g., influenza symptoms, influenza complications) in each model. Both models' estimates were highly correlated with ILINet pre-H1N1 and over the entire surveillance period, although the original model underestimated the magnitude of ILI activity during pH1N1. The updated model was more correlated with ILINet than the original model during Summer H1N1 (r = 0.95 and 0.29, respectively). The updated model included more search query terms than the original model, with more queries directly related to influenza infection, whereas the original model contained more queries related to influenza complications.ConclusionsInternet search behavior changed during pH1N1, particularly in the categories “influenza complications” and “term for influenza.” The complications associated with pH1N1, the fact that pH1N1 began in the summer rather than winter, and changes in health-seeking behavior each may have played a part. Both GFT models performed well prior to and during pH1N1, although the updated model performed better during pH1N1, especially during the summer months.

Evaluating Syndromic surveillance systems at institutions of higher education (IHEs): A retrospective analysis of the 2009 H1N1 influenza pandemic at two universities

BackgroundSyndromic surveillance has been widely adopted as a real-time monitoring tool for timely response to disease outbreaks. During the second wave of the pH1N1 pandemic in Fall 2009, two major universities in Washington, DC collected data that were potentially indicative of influenza-like illness (ILI) cases in students and staff. In this study, our objectives were three-fold. The primary goal of this study was to characterize the impact of pH1N1 on the campuses as clearly as possible given the data available and their likely biases. In addition, we sought to evaluate the strengths and weaknesses of the data series themselves, in order to inform these two universities and other institutions of higher education (IHEs) about real-time surveillance systems that are likely to provide the most utility in future outbreaks (at least to the extent that it is possible to generalize from this analysis).MethodsWe collected a wide variety of data that covered both student ILI cases reported to medical and non-medical staff, employee absenteeism, and hygiene supply distribution records (from University A only). Communication data were retrieved from university broadcasts, university preparedness websites, and H1N1-related on campus media reports. Regional data based on the Centers for Disease Control and Prevention Outpatient Influenza-like Illness Surveillance Network (CDC ILINet) surveillance network, American College Health Association (ACHA) pandemic influenza surveillance data, and local Google Flu Trends were used as external data sets. We employed a "triangulation" approach for data analysis in which multiple contemporary data sources are compared to identify time patterns that are likely to reflect biases as well as those that are more likely to be indicative of actual infection rates.ResultsMedical personnel observed an early peak at both universities immediately after school began in early September and a second peak in early November; only the second peak corresponded to patterns in the community at large. Self-reported illness to university deans' offices was also relatively increased during mid-term exam weeks. The overall volume of pH1N1-related communication messages similarly peaked twice, corresponding to the two peaks of student ILI cases.ConclusionsDuring the 2009 H1N1 pandemic, both University A and B experienced a peak number of ILI cases at the beginning of the Fall term. This pattern, seen in surveillance systems at these universities and to a lesser extent in data from other IHEs, most likely resulted from students bringing the virus back to campus from their home states coupled with a sudden increase in population density in dormitories and lecture halls. Through comparison of data from different syndromic surveillance data streams, paying attention to the likely biases in each over time, we have determined, at least in the case of the pH1N1 pandemic, that student health center data more accurately depicted disease transmission on campus at both universities during the Fall 2009 pandemic than other available data sources.

“Google Flu Trends” and Emergency Department Triage Data Predicted the 2009 Pandemic H1N1 Waves in Manitoba

We assessed the performance of syndromic indicators based on Google Flu Trends (GFT) and emergency department (ED) data for the early detection and monitoring of the 2009 H1N1 pandemic waves in Manitoba. Time-series curves for the weekly counts of laboratory-confirmed H1N1 cases in Manitoba during the 2009 pandemic were plotted against the three syndromic indicators: 1) GFT data, based on flu-related Internet search queries, 2) weekly count of all ED visits triaged as influenza-like illness (ED ILI volume), and 3) percentage of all ED visits that were triaged as an ILI (ED ILI percent). A linear regression model was fitted separately for each indicator and correlations with weekly virologic data were calculated for different lag periods for each pandemic wave. All three indicators peaked 1-2 weeks earlier than the epidemic curve of laboratory-confirmed cases. For GFT data, the best-fitting model had about a 2-week lag period in relation to the epidemic curve. Similarly, the best-fitting models for both ED indicators were observed for a time lag of 1-2 weeks. All three indicators performed better as predictors of the virologic time trends during the second wave as compared to the first. There was strong congruence between the time series of the GFT and both the ED ILI volume and the ED ILI percent indicators. During an influenza season characterized by high levels of disease activity, GFT and ED indicators provided a good indication of weekly counts of laboratory-confirmed influenza cases in Manitoba 1-2 weeks in advance.

Web query-based surveillance in Sweden during the influenza A(H1N1)2009 pandemic, April 2009 to February 2010

At the Swedish Institute for Communicable Disease Control, statistical models based on queries submitted to a Swedish medical website are used as a complement to the regular influenza surveillance. The models have previously been shown to perform well for seasonal influenza. The purpose of the present study was to evaluate the performance of the statistical models in the context of the influenza A(H1N1)2009 pandemic, a period when many factors, for example the media, could have influenced people's search behaviour on the Internet and consequently the performance of the models. Our evaluation indicates consistent good reliability for the statistical models also during the pandemic. When compared to Google Flu Trends for Sweden, they were at least equivalent in terms of estimating the influenza activity, and even seemed to be more precise in estimating the peak incidence of the influenza pandemic.

Monitoring Influenza Activity in the United States: A Comparison of Traditional Surveillance Systems with Google Flu Trends

BackgroundGoogle Flu Trends was developed to estimate US influenza-like illness (ILI) rates from internet searches; however ILI does not necessarily correlate with actual influenza virus infections.Methods and FindingsInfluenza activity data from 2003–04 through 2007–08 were obtained from three US surveillance systems: Google Flu Trends, CDC Outpatient ILI Surveillance Network (CDC ILI Surveillance), and US Influenza Virologic Surveillance System (CDC Virus Surveillance). Pearson's correlation coefficients with 95% confidence intervals (95% CI) were calculated to compare surveillance data. An analysis was performed to investigate outlier observations and determine the extent to which they affected the correlations between surveillance data. Pearson's correlation coefficient describing Google Flu Trends and CDC Virus Surveillance over the study period was 0.72 (95% CI: 0.64, 0.79). The correlation between CDC ILI Surveillance and CDC Virus Surveillance over the same period was 0.85 (95% CI: 0.81, 0.89). Most of the outlier observations in both comparisons were from the 2003–04 influenza season. Exclusion of the outlier observations did not substantially improve the correlation between Google Flu Trends and CDC Virus Surveillance (0.82; 95% CI: 0.76, 0.87) or CDC ILI Surveillance and CDC Virus Surveillance (0.86; 95%CI: 0.82, 0.90).ConclusionsThis analysis demonstrates that while Google Flu Trends is highly correlated with rates of ILI, it has a lower correlation with surveillance for laboratory-confirmed influenza. Most of the outlier observations occurred during the 2003–04 influenza season that was characterized by early and intense influenza activity, which potentially altered health care seeking behavior, physician testing practices, and internet search behavior.

Internet Search Limitations and Pandemic Influenza, Singapore

To the Editor: In the past few years, several publications have reported that Internet search queries may usefully supplement other, traditional surveillance programs for infectious diseases (1–3). The philanthropic arm of Google offers Flu Trends, a site that provides up-to-date estimates of influenza activity in 20 countries of the Pacific Rim and Europe (4) by using data mining techniques to find good predictors of historic influenza indicators (1). This service has yet to be extended to other countries and other diseases because access to official surveillance data is required, among other reasons. However, another Google service, Insights for Search, enables users to find and download time-series data of relative counts of arbitrary searches for a large number of countries (5). Pelat et al. have shown that a few, well-chosen searches on Google Insights provide data that closely correlate with French surveillance data for seasonal influenza, chickenpox, and gastroenteritis (3). Although Internet searches appear to be a promising tool for public health surveillance, our experience from using Google Insights in the context of pandemic (H1N1) 2009 in Singapore suggests it has important limitations. In Singapore, the recent pandemic caused an outbreak that peaked at the start of August 2009; the first confirmed importation was at the end of May and first confirmed unlinked case was at the end of June. However, the number Google searches for “influenza,” “H1N1,” “swine flu,” and similar terms (in English and Chinese), as well as symptoms associated with the disease, peaked much earlier than did the number of cases (Figure). The number of searches surged after newsworthy events but was low during the epidemic itself and had declined to about 20% of maximum search volume by the time of the actual peak, as shown by traditional surveillance. Furthermore, no discernible local maxima were observed that corresponded to the peak in case data. In contrast, alternative traditional measures of influenza incidence—prevalence of the novel strain among viral samples and general practice surveillance (6,7)—provide a consistent description of the outbreak. Figure Number of Google searches conducted for “influenza” (black lines) and “H1N1” (gray lines) compared with number of acute respiratory infections (ARI, gray bars) reported in government clinics, Singapore, 2009. During the ... This finding echoes a major point raised by Carneiro and Mylonakis (2), namely, that without adjusting for spikes driven by disease publicity rather than the disease itself, Internet searches may lose much of their value in supplementing traditional surveillance measures. Our experience is that using Google Insights to survey a disease may not work well for diseases with considerable media exposure, in particular, emerging diseases such as pandemic (H1N1) 2009 or severe acute respiratory syndrome. Such outbreaks may require the more sophisticated approach used by Flu Trends, should it be extended to other diseases and more corners of the globe. We agree with Pelat et al. (3) that Google Insights may work well for less-publicized infectious diseases. The dividing line between well-publicized and unpublicized diseases may, however, remain ambiguous. Thus, to ensure that web search data reflect disease incidence requires validation against traditional surveillance, although in that situation, the availability of corroborating traditional methods of surveillance limits the value of web-query data.

Predicting consumer behavior with Web search

Recent work has demonstrated that Web search volume can "predict the present," meaning that it can be used to accurately track outcomes such as unemployment levels, auto and home sales, and disease prevalence in near real time. Here we show that what consumers are searching for online can also predict their collective future behavior days or even weeks in advance. Specifically we use search query volume to forecast the opening weekend box-office revenue for feature films, first-month sales of video games, and the rank of songs on the Billboard Hot 100 chart, finding in all cases that search counts are highly predictive of future outcomes. We also find that search counts generally boost the performance of baseline models fit on other publicly available data, where the boost varies from modest to dramatic, depending on the application in question. Finally, we reexamine previous work on tracking flu trends and show that, perhaps surprisingly, the utility of search data relative to a simple autoregressive model is modest. We conclude that in the absence of other data sources, or where small improvements in predictive performance are material, search queries provide a useful guide to the near future.

Monitoring influenza activity in Europe with Google Flu Trends: comparison with the findings of sentinel physician networks – results for 2009-10

The number of Internet searches has recently been used by Google to estimate the influenza incidence in the United States. We examined the correlation between the Google Flu Trends tool and sentinel networks estimates in several European countries during the 2009 influenza A(H1N1) pandemic and found a good correlation between estimates and peak incidence timing, with the highest peaks in countries where Internet is most frequently used for health-related searching. Although somehow limited, Google could be a valuable tool for syndromic surveillance.

Temporal data classification using linear classifiers

Data classification is usually based on measurements recorded at the same time. This paper considers temporal data classification where the input is a temporal database that describes measurements over a period of time in history while the predicted class is expected to occur in the future. We describe a new temporal classification method that improves the accuracy of standard classification methods. The benefits of the method are tested on weather forecasting using the meteorological database from the Texas Commission on Environmental Quality and on influenza using the Google Flu Trends database.

Use of Google in Study of Noninfectious Medical Conditions

To the Editors: Google's search engine (Google Inc, Mountain View, CA; http://www.google.com) has emerged as a ubiquitous tool of Internet users. In 2009, Google commanded a 65% share of the roughly 130 billion online searches performed in the United States.1 The website Google Flu Trends monitors search data of flu-related terms to localize outbreaks 7–10 days earlier than traditional systems.2,3 Other infectious disease diagnoses, such as avian influenza and West Nile virus, have demonstrated high correlation with Google search volume.2,3 In August 2008, Google introduced Insights for Search (http://www.google.com/insights/search), which provides normalized, relative search volume for selected terms over specific time ranges and geographic regions. We hypothesized that chronic noninfectious diseases with known variations in seasonal incidence (such as diabetes mellitus, blood pressure, myocardial infarction, and nephrolithiasis) would show seasonal variations in number of searches. On February 10, 2010, we queried Google Insights for Search with the following terms: “diabetes,” “high blood pressure + hypertension,” “low blood pressure + hypotension,” “heart attack + myocardial infarction,” and “kidney stones.” A new query was generated for each set of terms, and restricted to searches done in the United States from February 2005 to January 2010. This produced 261 weeks of normalized search volume, which we divided into monthly intervals. Values greater than 3 standard deviations from the mean were excluded because they were likely driven by a nonmedical search (eg, “heart attack” after Michael Jackson's June 2009 death). Google Insights for Search provides the probability that a random user searched for a particular term from a specific location and time. Output is normalized on a scale of 0–100 and does not provide absolute search volume. We used ANOVA to assess differences within the variables. Of 1305 disease-weeks of data, 1300 were included for analysis. Each search term showed seasonal variation (Fig.). “Diabetes” had a sinusoidal pattern with a peak in March and a trough in August. “High blood pressure + hypertension” peaked during colder months, while “low blood pressure + hypotension” peaked in the summer. “Heart attack + myocardial infarction” peaked in the winter months and declined in the summer. “Kidney stones” peaked in the summer months. Each search term's relative volume decreased in December, likely corresponding to increased online shopping during the holiday season.4FIGURE.: Use of Google Insights for Search to monitor medical conditions with known seasonal variation in the United States from February 2005 to January 2010. A, Diabetes. B, Heart attack. C, High blood pressure (solid lines) and low blood pressure (dashed lines). D, Kidney stones. The y-axis represents relative search volume and the x-axis represents the cumulative monthly totals.We searched a readily available, public-use Web site that processes millions of individual health-related searches to derive an ecologic surrogate of disease burden. The technology has been applied to infectious diseases,2 but to our knowledge has not been used to track other conditions. Each disease we studied showed a plausible seasonal variation. Diabetes incidence and hemoglobin A1c levels are known to peak in March, potentially relating to the inactivity of winter.5 The mechanism for blood pressure elevation in winter is uncertain, but has been hypothesized to involve cold-weather sympathetic-nervous-system activation, catecholamine release, and increased peripheral vascular resistance.6 This may also contribute to the increased cardiovascular morbidity seen in winter months.7 Nephrolithiasis is thought to be more common in summer due to elevated temperatures, dehydration, and increased sun exposure.8 Google Insights for Search provides robust, immediate, and free temporal and geographic data from billions of search queries. It is most effective in areas with widespread Internet use. Online search volume can provide useful information for epidemiologic study and medical research. Benjamin N. Breyer Michael L. Eisenberg Department of Urology School of Medicine, University of California San Francisco, CA [email protected]

Sistemas de usuario en la gestión online de contenidos biomédicos

Public demand for information about health matters has led several organizations to develop custom systems for self-management of medical information. These systems facilitate the management of this information by the public, health care professionals, public institutions, companies and groups of patients. Thanks to new technologies, patients are taking an active role in searching the internet with queries about health-related issues, which gives rise to a debate about the accreditations of web pages that disseminate health information. In this context, Google, with applications such as Google Health, Flu Trends and Android, and Microsoft, with its HealthVault utility, have become the maximum exponents of medical information management software.

Assessing North American influenza dynamics with a statistical SIRS model

We present a general statistical modeling framework to characterize continental-level influenza dynamics in the United States for the purposes of examining state-level epidemiological sources and sinks. The methods we describe depend directly on state-level influenza data that are prepared on a weekly basis by Google Flu Trends. The Google Flu Trends team has provided a powerful new approach to collecting and reporting epidemiological data and, when used in conjunction with sophisticated statistical models, can allow for the identification and quantification of the flow of influenza across the continental United States. Our proposed methods, when conditioned on such a comprehensive search query product, can provide unprecedented scientific learning about large-scale pathways and barriers to disease transmission which can ultimately be helpful for policy, remediation, and response efforts.

Tracking Flu Epidemics Using Google Flu Trends and Particle Learning

In the second half of 2009 the world experienced an intense influenza activity. The new 2009 H1N1 virus, formerly known as the swine flu, has in only five months found its way from Mexico to a majority of the countries on the planet. The fears of a large second-wave pandemic and its potential impact on health and economic outcomes have underlined the importance of accurate and fast disease surveillance mechanisms capable of suggesting timely public health interventions.In this paper we introduce a state-space tracking approach, based on particle learning (PL) for classic compartmental epidemics models (such as, for example, the susceptible-exposed-infected-recovered (SEIR)). The proposed approach is particularly well-suited to on-line learning and surveillance of infectious diseases as it is capable of assessing the odds of an epidemic at each time point, while simultaneously accounting for uncertainty in disease parameters and producing real-time predictive distributions. As compared to the now widely used MCMC \cite{ONeill/Roberts:1999,Elderdetal:2006,Leman2009} and perfect sampling \cite{Fearnhead:2004} based methods, the PL method, which is based on a clever use of an essential state vector, is easier to implement, computationally faster, as well as more readily generalizable to problems with complex and time-varying dynamics. In particular, we show how the PL approach, in combination with Bayes Factors, can be used as an on-line diagnostic and surveillance tool for tracking influenza using the Google Flu Trends data. We take a closer look at the spread of flu in the US during 2003-2009, and in New Zealand during 2006-2009, with a special emphasis on the recent epidemic season. We also outline several future directions, including the development of a PL methodology for on-line spatial surveillance as a tool for guiding public health interventions and emergency resource allocation.

Google Trends: A Web‐Based Tool for Real‐Time Surveillance of Disease Outbreaks

Google Flu Trends can detect regional outbreaks of influenza 7-10 days before conventional Centers for Disease Control and Prevention surveillance systems. We describe the Google Trends tool, explain how the data are processed, present examples, and discuss its strengths and limitations. Google Trends shows great promise as a timely, robust, and sensitive surveillance system. It is best used for surveillance of epidemics and diseases with high prevalences and is currently better suited to track disease activity in developed countries, because to be most effective, it requires large populations of Web search users. Spikes in search volume are currently hard to interpret but have the benefit of increasing vigilance. Google should work with public health care practitioners to develop specialized tools, using Google Flu Trends as a blueprint, to track infectious diseases. Suitable Web search query proxies for diseases need to be established for specialized tools or syndromic surveillance. This unique and innovative technology takes us one step closer to true real-time outbreak surveillance.

Interpreting “Google Flu Trends” data for pandemic H1N1 influenza: The New Zealand experience

For the period of the spread of pandemic H1N1 influenza in New Zealand during 2009, we compared results from Google Flu Trends with data from existing surveillance systems. The patterns from Google Flu Trends were closely aligned with (peaking a week before and a week after) two independent national surveillance systems for influenza-like illness (ILI) cases. It was much less congruent with (delayed by three weeks) data from ILI-related calls to a national free-phone Healthline and with media coverage of pandemic influenza. Some patterns were unique to Google Flu Trends and may not have reflected the actual ILI burden in the community. Overall, Google Flu Trends appears to provide a useful free surveillance system but it should probably be seen as supplementary rather than as an alternative.