Benjamin M. Althouse

Prediction of Dengue Incidence Using Search Query Surveillance

Background The use of internet search data has been demonstrated to be effective at predicting influenza incidence. This approach may be more successful for dengue which has large variation in annual incidence and a more distinctive clinical presentation and mode of transmission. Methods We gathered freely-available dengue incidence data from Singapore (weekly incidence, 2004–2011) and Bangkok (monthly incidence, 2004–2011). Internet search data for the same period were downloaded from Google Insights for Search. Search terms were chosen to reflect three categories of dengue-related search: nomenclature, signs/symptoms, and treatment. We compared three models to predict incidence: a step-down linear regression, generalized boosted regression, and negative binomial regression. Logistic regression and Support Vector Machine (SVM) models were used to predict a binary outcome defined by whether dengue incidence exceeded a chosen threshold. Incidence prediction models were assessed using and Pearson correlation between predicted and observed dengue incidence. Logistic and SVM model performance were assessed by the area under the receiver operating characteristic curve. Models were validated using multiple cross-validation techniques. Results The linear model selected by AIC step-down was found to be superior to other models considered. In Bangkok, the model has an , and a correlation of 0.869 between fitted and observed. In Singapore, the model has an , and a correlation of 0.931. In both Singapore and Bangkok, SVM models outperformed logistic regression in predicting periods of high incidence. The AUC for the SVM models using the 75th percentile cutoff is 0.906 in Singapore and 0.960 in Bangkok. Conclusions Internet search terms predict incidence and periods of large incidence of dengue with high accuracy and may prove useful in areas with underdeveloped surveillance systems. The methods presented here use freely available data and analysis tools and can be readily adapted to other settings.

Seasonality in seeking mental health information on Google.

BACKGROUND Population mental health surveillance is an important challenge limited by resource constraints, long time lags in data collection, and stigma. One promising approach to bridge similar gaps elsewhere has been the use of passively generated digital data. PURPOSE This article assesses the viability of aggregate Internet search queries for real-time monitoring of several mental health problems, specifically in regard to seasonal patterns of seeking out mental health information. METHODS All Google mental health queries were monitored in the U.S. and Australia from 2006 to 2010. Additionally, queries were subdivided among those including the terms ADHD (attention deficit-hyperactivity disorder); anxiety; bipolar; depression; anorexia or bulimia (eating disorders); OCD (obsessive-compulsive disorder); schizophrenia; and suicide. A wavelet phase analysis was used to isolate seasonal components in the trends, and based on this model, the mean search volume in winter was compared with that in summer, as performed in 2012. RESULTS All mental health queries followed seasonal patterns with winter peaks and summer troughs amounting to a 14% (95% CI=11%, 16%) difference in volume for the U.S. and 11% (95% CI=7%, 15%) for Australia. These patterns also were evident for all specific subcategories of illness or problem. For instance, seasonal differences ranged from 7% (95% CI=5%, 10%) for anxiety (followed by OCD, bipolar, depression, suicide, ADHD, schizophrenia) to 37% (95% CI=31%, 44%) for eating disorder queries in the U.S. Several nonclinical motivators for query seasonality (such as media trends or academic interest) were explored and rejected. CONCLUSIONS Information seeking on Google across all major mental illnesses and/or problems followed seasonal patterns similar to those found for seasonal affective disorder. These are the first data published on patterns of seasonality in information seeking encompassing all the major mental illnesses, notable also because they likely would have gone undetected using traditional surveillance.

What Can Digital Disease Detection Learn from (an External Revision to) Google Flu Trends

BACKGROUND Google Flu Trends (GFT) claimed to generate real-time, valid predictions of population influenza-like illness (ILI) using search queries, heralding acclaim and replication across public health. However, recent studies have questioned the validity of GFT. PURPOSE To propose an alternative methodology that better realizes the potential of GFT, with collateral value for digital disease detection broadly. METHODS Our alternative method automatically selects specific queries to monitor and autonomously updates the model each week as new information about CDC-reported ILI becomes available, as developed in 2013. Root mean squared errors (RMSEs) and Pearson correlations comparing predicted ILI (proportion of patient visits indicative of ILI) with subsequently observed ILI were used to judge model performance. RESULTS During the height of the H1N1 pandemic (August 2 to December 22, 2009) and the 2012-2013 season (September 30, 2012, to April 12, 2013), GFTs predictions had RMSEs of 0.023 and 0.022 (i.e., hypothetically, if GFT predicted 0.061 ILI one week, it is expected to err by 0.023) and correlations of r=0.916 and 0.927. Our alternative method had RMSEs of 0.006 and 0.009, and correlations of r=0.961 and 0.919 for the same periods. Critically, during these important periods, the alternative method yielded more accurate ILI predictions every week, and was typically more accurate during other influenza seasons. CONCLUSIONS GFT may be inaccurate, but improved methodologic underpinnings can yield accurate predictions. Applying similar methods elsewhere can improve digital disease detection, with broader transparency, improved accuracy, and real-world public health impacts.

Asymptomatic transmission and the resurgence of Bordetella pertussis

BackgroundThe recent increase in whooping cough incidence (primarily caused by Bordetella pertussis) presents a challenge to both public health practitioners and scientists trying to understand the mechanisms behind its resurgence. Three main hypotheses have been proposed to explain the resurgence: 1) waning of protective immunity from vaccination or natural infection over time, 2) evolution of B. pertussis to escape protective immunity, and 3) low vaccine coverage. Recent studies have suggested a fourth mechanism: asymptomatic transmission from individuals vaccinated with the currently used acellular B. pertussis vaccines.MethodsUsing wavelet analyses of B. pertussis incidence in the United States (US) and United Kingdom (UK) and a phylodynamic analysis of 36 clinical B. pertussis isolates from the US, we find evidence in support of asymptomatic transmission of B. pertussis. Next, we examine the clinical, public health, and epidemiological consequences of asymptomatic B. pertussis transmission using a mathematical model.ResultsWe find that: 1) the timing of changes in age-specific attack rates observed in the US and UK are consistent with asymptomatic transmission; 2) the phylodynamic analysis of the US sequences indicates more genetic diversity in the overall bacterial population than would be suggested by the observed number of infections, a pattern expected with asymptomatic transmission; 3) asymptomatic infections can bias assessments of vaccine efficacy based on observations of B. pertussis-free weeks; 4) asymptomatic transmission can account for the observed increase in B. pertussis incidence; and 5) vaccinating individuals in close contact with infants too young to receive the vaccine (“cocooning” unvaccinated children) may be ineffective.ConclusionsAlthough a clear role for the previously suggested mechanisms still exists, asymptomatic transmission is the most parsimonious explanation for many of the observations surrounding the resurgence of B. pertussis in the US and UK. These results have important implications for B. pertussis vaccination policy and present a complicated scenario for achieving herd immunity and B. pertussis eradication.

Could Behavioral Medicine Lead the Web Data Revolution

Digital footprints left on search engines, social media, and social networking sites can be aggregated and analyzed as health proxies, yielding anonymous and instantaneous insights. On the one hand, nearly all the existing work has focused on acute diseases. This means the value-added from web surveillance is reduced, because the effectiveness of even high profile systems, such as Google Flu Trends, have been found inferior to already strong traditional surveillance.1 On the other hand, the future of web surveillance is promising in an area where traditional surveillance is largely incomplete: behavioral medicine, a multidisciplinary field incorporating medicine, social science, and public health and focusing on health behaviors and mental health. The proportion of illness (or death) attributable to health behaviors or psychological well-being has steadily increased over the last half century, while surveillance of these outcomes has remained largely unchanged. Investigators simply ask people about their health on surveys. However, surveys have well-known limitations, such as respondents’ reluctance to participate, social desirability biases, difficulty in accurately reporting behaviors, long lags between data collection and availability, and provisions (sometimes legal) curtailing the inclusion of politically sensitive topics like gun violence. Most importantly, the expense of surveys means many topics are either not covered or covered restrictively (e.g., clinical depression screeners are included in the Behavioral Risk Factor Surveillance System just every other year). Given the current budget climate, survey capacity will likely worsen before it improves. To overcome these limits, behavioral medicine should now embrace web data. First, behavioral medicine requires observing behavior or the manifestation of mental health problems. Doing so online is easier, more comprehensive, and more effective than with surveys, because many outcomes are passively exhibited there. For example, one study showed how precise health concerns changed during the United States recession of December 2008 through 2011, by systematically selecting Google search queries and using the content of each query to describe the concern and the change in volume to describe concern prevalence. “Stomach ulcer symptoms,” for example, were 228% (95%CI, 35–363) higher than expected during the recession, with queries thematically related to arrhythmia, congestion, pain (including many foci like head, tooth and back) also elevated.2 This approach highlights how web data can reveal largely assumption-free insights, via systematic data generation of hundreds of possible outcomes rather than arbitrary a priori selection of a few outcomes by investigators. Second, web data reflects more than the individual, because social context can also be captured online. Online networks can reveal how mechanistic drivers such as social norms spread and influence population health. For example, social patterns in obesity promotion and suppression have been described by pooling Facebook posts that encourage television watching or going outdoors, which ultimately explained variability in neighborhood obesity rates.3 Moreover, social support concepts are often expressed in web data, like observing specific instances of caregiving and confidence on Twitter. As a result, online behavioral medicine can move away from understanding aggregation based purely on location and towards understanding health in the context of our human interconnectedness. Third, web data are potentially the only source for real-time insights into behavioral medicine, where web data can be available almost immediately compared to a 365-day lag time between annual surveys. By harnessing these data around social events or interventions, programs can be evaluated as they are implemented, hypothetically generating real-time feedback to maximize their effectiveness. Web data in this vein also hold promise for guiding investigator resources. In 2011, when tobacco journals were debating snus (a smokeless tobacco product), and funders were soliciting proposals to understand the snus pandemic, electronic cigarettes already attracted more searches on Google than any other smoking alternative, snus included.4 In this same way, web data can guide traditional surveillance, like vetting the inclusion of questions on surveys using online proxies. Fourth, given all hypotheses are based on some data, web data can be an important source for identifying new hypotheses. Many hypotheses in behavioral medicine can be traced directly to data availability and can appear ad hoc to lay audiences. Many studies have explored birthdate seasonality in mental health problems. Why? Birthdates are routinely found in traditional surveillance, while some mental health problems are too rare to assess incidence or increased severity seasonality. As a result, obvious questions are never explored, until now. Is schizophrenia seasonal? Online interest in schizophrenia and its symptoms – as well as 8 other outcomes - peak in the winter.5 What is the healthiest day? Online interest in quitting smoking across the globe is highest on Monday.6 Behavioral medicine needs to escape the confines of limited data to more fully specify the next frontier of research questions, and going online is one such escape. Fifth, it is beyond present scientific limits for a hypothetical arm to reach out of the screen to inoculate against infection. In behavioral medicine, however, substantial resources have been used to develop online interventions that treat or prevent illness with effectiveness equivalent to their offline counterparts. For example, as early as the mid-1990s, investigators implemented online programs to promote behavioral health. A meta-analysis found these programs relatively increased quitting smoking 44%,7 yet a research agenda for harnessing the surveillance potential of the web has not been articulated. Improving the online surveillance capacity means online interventions can be better disseminated via online screening or linking subjects to existing online treatments (i.e., what advertisements for an online program are most effective?). Sixth, some of the most effective interventions in behavioral medicine involve changes in public policy. Web data can identify alerts for policy changes and pathways for health advocacy. For instance, by archiving online media, places considering policy changes can be identified, and this information can then be passed onto advocacy groups. Case in point, Brazilian President Lula’s laryngeal cancer prompted broad changes in media coverage of tobacco control, and soon after, Brazil became the largest smoke-free nation to date.8 By prospectively analyzing news media content, advocacy resources may be more cost-effectively spent during opportunistic times, including events like Lula’s diagnosis, will be possible. A major criticism is that web data have sampling biases. However, such biases are increasingly eroding at the population level as more people go online. In addition, several studies have demonstrated that valid trends reflecting the entire population, and even subsets of the population, can be extracted from online data. For example, computer science has already developed approaches for identifying the gender, ethnicity or education associated with a Twitter account using the content of a user’s Tweets. Going forward, the research community may mimic these studies and validate methods for obtaining high quality, actionable information in behavioral medicine, then further realizing the comparative value of web data to traditional data. Billions of digital footprints from nearly all parts of the United States and from countries around the world provide a powerful opportunity to expand the evidence-base across medicine. However, for the above reasons and more related reasons yet to be expressed, behavioral medicine potentially has the most to gain from web data and could be essential to the broader web data revolution.

Enhancing disease surveillance with novel data streams: challenges and opportunities

Novel data streams (NDS), such as web search data or social media updates, hold promise for enhancing the capabilities of public health surveillance. In this paper, we outline a conceptual framework for integrating NDS into current public health surveillance. Our approach focuses on two key questions: What are the opportunities for using NDS and what are the minimal tests of validity and utility that must be applied when using NDS? Identifying these opportunities will necessitate the involvement of public health authorities and an appreciation of the diversity of objectives and scales across agencies at different levels (local, state, national, international). We present the case that clearly articulating surveillance objectives and systematically evaluating NDS and comparing the performance of NDS to existing surveillance data and alternative NDS data is critical and has not sufficiently been addressed in many applications of NDS currently in the literature.

A Novel Evaluation of World No Tobacco Day in Latin America

Background World No Tobacco Day (WNTD), commemorated annually on May 31, aims to inform the public about tobacco harms. Because tobacco control surveillance is usually annualized, the effectiveness of WNTD remains unexplored into its 25th year. Objective To explore the potential of digital surveillance (infoveillance) to evaluate the impacts of WNTD on population awareness of and interest in cessation. Methods Health-related news stories and Internet search queries were aggregated to form a continuous and real-time data stream. We monitored daily news coverage of and Internet search queries for cessation in seven Latin American nations from 2006 to 2011. Results Cessation news coverage peaked around WNTD, typically increasing 71% (95% confidence interval [CI] 61–81), ranging from 61% in Mexico to 83% in Venezuela. Queries indicative of cessation interest peaked on WNTD, increasing 40% (95% CI 32–48), ranging from 24% in Colombia to 84% in Venezuela. A doubling in cessation news coverage was associated with approximately a 50% increase in cessation queries. To gain a practical perspective, we compared WNTD-related activity with New Year’s Day and several cigarette excise tax increases in Mexico. Cessation queries around WNTD were typically greater than New Year’s Day and approximated a 2.8% (95% CI –0.8 to 6.3) increase in cigarette excise taxes. Conclusions This novel evaluation suggests WNTD had a significant impact on popular awareness (media trends) and individual interest (query trends) in smoking cessation. Because WNTD is constantly evolving, our work is also a model for real-time surveillance and potential improvement in WNTD and similar initiatives.

Impact of Climate and Mosquito Vector Abundance on Sylvatic Arbovirus Circulation Dynamics in Senegal

Sylvatic arboviruses have been isolated in Senegal over the last 50 years. The ecological drivers of the pattern and frequency of virus infection in these species are largely unknown. We used time series analysis and Bayesian hierarchical count modeling on a long-term arbovirus dataset to test associations between mosquito abundance, weather variables, and the frequency of isolation of dengue, yellow fever, chikungunya, and Zika viruses. We found little correlation between mosquito abundance and viral isolations. Rainfall was a negative predictor of dengue virus (DENV) isolation but a positive predictor of Zika virus isolation. Temperature was a positive predictor of yellow fever virus (YFV) isolations but a negative predictor of DENV isolations. We found slight interference between viruses, with DENV negatively associated with concurrent YFV isolation and YFV negatively associated with concurrent isolation of chikungunya virus. These findings begin to characterize some of the ecological associations of sylvatic arboviruses with each other and climate and mosquito abundance.

A public choice framework for controlling transmissible and evolving diseases

Control measures used to limit the spread of infectious disease often generate externalities. Vaccination for transmissible diseases can reduce the incidence of disease even among the unvaccinated, whereas antimicrobial chemotherapy can lead to the evolution of antimicrobial resistance and thereby limit its own effectiveness over time. We integrate the economic theory of public choice with mathematical models of infectious disease to provide a quantitative framework for making allocation decisions in the presence of these externalities. To illustrate, we present a series of examples: vaccination for tetanus, vaccination for measles, antibiotic treatment of otitis media, and antiviral treatment of pandemic influenza.

Evolution in health and medicine Sackler colloquium: a public choice framework for controlling transmissible and evolving diseases.

Control measures used to limit the spread of infectious disease often generate externalities. Vaccination for transmissible diseases can reduce the incidence of disease even among the unvaccinated, whereas antimicrobial chemotherapy can lead to the evolution of antimicrobial resistance and thereby limit its own effectiveness over time. We integrate the economic theory of public choice with mathematical models of infectious disease to provide a quantitative framework for making allocation decisions in the presence of these externalities. To illustrate, we present a series of examples: vaccination for tetanus, vaccination for measles, antibiotic treatment of otitis media, and antiviral treatment of pandemic influenza.