Geoffrey Fairchild

Global Disease Monitoring and Forecasting with Wikipedia

Infectious disease is a leading threat to public health, economic stability, and other key social structures. Efforts to mitigate these impacts depend on accurate and timely monitoring to measure the risk and progress of disease. Traditional, biologically-focused monitoring techniques are accurate but costly and slow; in response, new techniques based on social internet data, such as social media and search queries, are emerging. These efforts are promising, but important challenges in the areas of scientific peer review, breadth of diseases and countries, and forecasting hamper their operational usefulness. We examine a freely available, open data source for this use: access logs from the online encyclopedia Wikipedia. Using linear models, language as a proxy for location, and a systematic yet simple article selection procedure, we tested 14 location-disease combinations and demonstrate that these data feasibly support an approach that overcomes these challenges. Specifically, our proof-of-concept yields models with up to 0.92, forecasting value up to the 28 days tested, and several pairs of models similar enough to suggest that transferring models from one location to another without re-training is feasible. Based on these preliminary results, we close with a research agenda designed to overcome these challenges and produce a disease monitoring and forecasting system that is significantly more effective, robust, and globally comprehensive than the current state of the art.

Forecasting the 2013-2014 influenza season using Wikipedia.

Infectious diseases are one of the leading causes of morbidity and mortality around the world; thus, forecasting their impact is crucial for planning an effective response strategy. According to the Centers for Disease Control and Prevention (CDC), seasonal influenza affects 5% to 20% of the U.S. population and causes major economic impacts resulting from hospitalization and absenteeism. Understanding influenza dynamics and forecasting its impact is fundamental for developing prevention and mitigation strategies. We combine modern data assimilation methods with Wikipedia access logs and CDC influenza-like illness (ILI) reports to create a weekly forecast for seasonal influenza. The methods are applied to the 2013-2014 influenza season but are sufficiently general to forecast any disease outbreak, given incidence or case count data. We adjust the initialization and parametrization of a disease model and show that this allows us to determine systematic model bias. In addition, we provide a way to determine where the model diverges from observation and evaluate forecast accuracy. Wikipedia article access logs are shown to be highly correlated with historical ILI records and allow for accurate prediction of ILI data several weeks before it becomes available. The results show that prior to the peak of the flu season, our forecasting method produced 50% and 95% credible intervals for the 2013-2014 ILI observations that contained the actual observations for most weeks in the forecast. However, since our model does not account for re-infection or multiple strains of influenza, the tail of the epidemic is not predicted well after the peak of flu season has passed.

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.

Primary Stroke Centers Should Be Located Using Maximal Coverage Models for Optimal Access

Background and Purpose— The current self-initiated approach by which hospitals acquire Primary Stroke Center (PSC) certification provides insufficient coverage for large areas of the United States. An alternative, directed, algorithmic approach to determine near optimal locations of PSCs would be justified if it significantly improves coverage. Methods— Using geographic location–allocation modeling techniques, we developed a universal web-based calculator for selecting near optimal PSC locations designed to maximize the population coverage in any state. We analyzed the current PSC network population coverage in Iowa and compared it with the coverage that would exist if a maximal coverage model had instead been used to place those centers. We then estimated the expected gains in population coverage if additional PSCs follow the current self-initiated model and compared it against the more efficient coverage expected by use of a maximal coverage model to select additional locations. Results— The existing 12 self-initiated PSCs in Iowa cover 37% of the population, assuming a time–distance radius of 30 minutes. The current population coverage would have been 47.5% if those 12 PSCs had been located using a maximal coverage model. With the current self-initiated approach, 54 additional PSCs on average will be needed to improve coverage to 75% of the population. Conversely, only 31 additional PSCs would be needed to achieve the same degree of population coverage if a maximal coverage model is used. Conclusion— Given the substantial gain in population access to adequate acute stroke care, it appears justified to direct the location of additional PSCs or recombinant tissue-type plasminogen activator-capable hospitals through a maximal coverage model algorithmic approach.

Epidemic forecasting is messier than weather forecasting: The role of human behavior and internet data streams in epidemic forecast

Mathematical models, such as those that forecast the spread of epidemics or predict the weather, must overcome the challenges of integrating incomplete and inaccurate data in computer simulations, estimating the probability of multiple possible scenarios, incorporating changes in human behavior and/or the pathogen, and environmental factors. In the past 3 decades, the weather forecasting community has made significant advances in data collection, assimilating heterogeneous data steams into models and communicating the uncertainty of their predictions to the general public. Epidemic modelers are struggling with these same issues in forecasting the spread of emerging diseases, such as Zika virus infection and Ebola virus disease. While weather models rely on physical systems, data from satellites, and weather stations, epidemic models rely on human interactions, multiple data sources such as clinical surveillance and Internet data, and environmental or biological factors that can change the pathogen dynamics. We describe some of similarities and differences between these 2 fields and how the epidemic modeling community is rising to the challenges posed by forecasting to help anticipate and guide the mitigation of epidemics. We conclude that some of the fundamental differences between these 2 fields, such as human behavior, make disease forecasting more challenging than weather forecasting.

How many suffice? A computational framework for sizing sentinel surveillance networks

BackgroundData from surveillance networks help epidemiologists and public health officials detect emerging diseases, conduct outbreak investigations, manage epidemics, and better understand the mechanics of a particular disease. Surveillance networks are used to determine outbreak intensity (i.e., disease burden) and outbreak timing (i.e., the start, peak, and end of the epidemic), as well as outbreak location. Networks can be tuned to preferentially perform these tasks. Given that resources are limited, careful site selection can save costs while minimizing performance loss.MethodsWe study three different site placement algorithms: two algorithms based on the maximal coverage model and one based on the K-median model. The maximal coverage model chooses sites that maximize the total number of people within a specified distance of a site. The K-median model minimizes the sum of the distances from each individual to the individual’s nearest site. Using a ground truth dataset consisting of two million de-identified Medicaid billing records representing eight complete influenza seasons and an evaluation function based on the Huff spatial interaction model, we empirically compare networks against the existing Iowa Department of Public Health influenza-like illness network by simulating the spread of influenza across the state of Iowa.ResultsWe show that it is possible to design a network that achieves outbreak intensity performance identical to the status quo network using two fewer sites. We also show that if outbreak timing detection is of primary interest, it is actually possible to create a network that matches the existing network’s performance using 59% fewer sites.ConclusionsBy simulating the spread of influenza across the state of Iowa, we show that our methods are capable of designing networks that perform better than the status quo in terms of both outbreak intensity and timing. Additionally, our results suggest that network size may only play a minimal role in outbreak timing detection. Finally, we show that it may be possible to reduce the size of a surveillance system without affecting the quality of surveillance information produced.

Measuring Global Disease with Wikipedia: Success, Failure, and a Research Agenda

Effective disease monitoring provides a foundation for effective public health systems. This has historically been accomplished with patient contact and bureaucratic aggregation, which tends to be slow and expensive. Recent internet-based approaches promise to be real-time and cheap, with few parameters. However, the question of when and how these approaches work remains open. We addressed this question using Wikipedia access logs and category links. Our experiments, replicable and extensible using our open source code and data, test the effect of semantic article filtering, amount of training data, forecast horizon, and model staleness by comparing across 6 diseases and 4 countries using thousands of individual models. We found that our minimal-configuration, language-agnostic article selection process based on semantic relatedness is effective for improving predictions, and that our approach is relatively insensitive to the amount and age of training data. We also found, in contrast to prior work, very little forecasting value, and we argue that this is consistent with theoretical considerations about the nature of forecasting. These mixed results lead us to propose that the currently observational field of internet-based disease surveillance must pivot to include theoretical models of information flow as well as controlled experiments based on simulations of disease.

Optimizing human activity patterns using global sensitivity analysis

Implementing realistic activity patterns for a population is crucial for modeling, for example, disease spread, supply and demand, and disaster response. Using the dynamic activity simulation engine, DASim, we generate schedules for a population that capture regular (e.g., working, eating, and sleeping) and irregular activities (e.g., shopping or going to the doctor). We use the sample entropy (SampEn) statistic to quantify a schedule’s regularity for a population. We show how to tune an activity’s regularity by adjusting SampEn, thereby making it possible to realistically design activities when creating a schedule. The tuning process sets up a computationally intractable high-dimensional optimization problem. To reduce the computational demand, we use Bayesian Gaussian process regression to compute global sensitivity indices and identify the parameters that have the greatest effect on the variance of SampEn. We use the harmony search (HS) global optimization algorithm to locate global optima. Our results show that HS combined with global sensitivity analysis can efficiently tune the SampEn statistic with few search iterations. We demonstrate how global sensitivity analysis can guide statistical emulation and global optimization algorithms to efficiently tune activities and generate realistic activity patterns. Though our tuning methods are applied to dynamic activity schedule generation, they are general and represent a significant step in the direction of automated tuning and optimization of high-dimensional computer simulations.

The Biosurveillance Analytics Resource Directory (BARD): Facilitating the Use of Epidemiological Models for Infectious Disease Surveillance

Epidemiological modeling for infectious disease is important for disease management and its routine implementation needs to be facilitated through better description of models in an operational context. A standardized model characterization process that allows selection or making manual comparisons of available models and their results is currently lacking. A key need is a universal framework to facilitate model description and understanding of its features. Los Alamos National Laboratory (LANL) has developed a comprehensive framework that can be used to characterize an infectious disease model in an operational context. The framework was developed through a consensus among a panel of subject matter experts. In this paper, we describe the framework, its application to model characterization, and the development of the Biosurveillance Analytics Resource Directory (BARD; http://brd.bsvgateway.org/brd/), to facilitate the rapid selection of operational models for specific infectious/communicable diseases. We offer this framework and associated database to stakeholders of the infectious disease modeling field as a tool for standardizing model description and facilitating the use of epidemiological models.

Cross-Disciplinary Consultancy to Bridge Public Health Technical Needs and Analytic Developers: Asyndromic Surveillance Use Case

Introduction: We document a funded effort to bridge the gap between constrained scientific challenges of public health surveillance and methodologies from academia and industry. Component tasks are the collection of epidemiologists’ use case problems, multidisciplinary consultancies to refine them, and dissemination of problem requirements and shareable datasets. We describe an initial use case and consultancy as a concrete example and challenge to developers. Materials and Methods: Supported by the Defense Threat Reduction Agency Biosurveillance Ecosystem project, the International Society for Disease Surveillance formed an advisory group to select tractable use case problems and convene inter-disciplinary consultancies to translate analytic needs into well-defined problems and to promote development of applicable solution methods. The initial consultancy’s focus was a problem originated by the North Carolina Department of Health and its NC DETECT surveillance system: Derive a method for detection of patient record clusters worthy of follow-up based on free-text chief complaints and without syndromic classification. Results: Direct communication between public health problem owners and analytic developers was informative to both groups and constructive for the solution development process. The consultancy achieved refinement of the asyndromic detection challenge and of solution requirements. Participants summarized and evaluated solution approaches and discussed dissemination and collaboration strategies. Practice Implications: A solution meeting the specification of the use case described above could improve human monitoring efficiency with expedited warning of events requiring follow-up, including otherwise overlooked events with no syndromic indicators. This approach can remove obstacles to collaboration with efficient, minimal data-sharing and without costly overhead.