Systematic review of best practices for GPS data usage, processing, and linkage in health, exposure science and environmental context research.

Global Positioning System (GPS) technology is increasingly used in health research to capture individual mobility and contextual and environmental exposures. However, the tools, techniques and decisions for using GPS data vary from study to study, making comparisons and reproducibility challenging. OBJECTIVES: The objectives of this systematic review were to (1) identify best practices for GPS data collection and processing; (2) quantify reporting of best practices in published studies; and (3) discuss examples found in reviewed manuscripts that future researchers may employ for reporting GPS data usage, processing and linkage of GPS data in health studies. DESIGN: A systematic review. DATA SOURCES: Electronic databases searched (24 October 2023) were PubMed, Scopus and Web of Science (PROSPERO ID: CRD42022322166). ELIGIBILITY CRITERIA: Included peer-reviewed studies published in English met at least one of the criteria: (1) protocols involving GPS for exposure/context and human health research purposes and containing empirical data; (2) linkage of GPS data to other data intended for research on contextual influences on health; (3) associations between GPS-measured mobility or exposures and health; (4) derived variable methods using GPS data in health research; or (5) comparison of GPS tracking with other methods (eg, travel diary). DATA EXTRACTION AND SYNTHESIS: We examined 157 manuscripts for reporting of best practices including wear time, sampling frequency, data validity, noise/signal loss and data linkage to assess risk of bias. RESULTS: We found that 6% of the studies did not disclose the GPS device model used, only 12.1% reported the per cent of GPS data lost by signal loss, only 15.7% reported the per cent of GPS data considered to be noise and only 68.2% reported the inclusion criteria for their data. CONCLUSIONS: Our recommendations for reporting on GPS usage, processing and linkage may be transferrable to other geospatial devices, with the hope of promoting transparency and reproducibility in this research. PROSPERO REGISTRATION NUMBER: CRD42022322166.

Designing and Evaluating a Hierarchical Framework for Matching Food Outlets across Multi-sourced Geospatial Datasets: a Case Study of San Diego County.

Research on retail food environment (RFE) relies on data availability and accuracy. However, the discrepancies in RFE datasets may lead to imprecision when measuring association with health outcomes. In this research, we present a two-tier hierarchical point of interest (POI) matching framework to compare and triangulate food outlets across multiple geospatial data sources. Two matching parameters were used including the geodesic distance between businesses and the similarity of business names according to Levenshtein distance (LD) and Double Metaphone (DM). Sensitivity analysis was conducted to determine thresholds of matching parameters. Our Tier 1 matching used more restricted parameters to generate high confidence-matched POIs, whereas in Tier 2 we opted for relaxed matching parameters and applied a weighted multi-attribute model on the previously unmatched records. Our case study in San Diego County, California used government, commercial, and crowdsourced data and returned 20.2% matched records from Tier 1 and 18.6% matched from Tier 2. Our manual validation shows a 100% matching rate for Tier 1 and up to 30.6% for Tier 2. Matched and unmatched records from Tier 1 were further analyzed for spatial patterns and categorical differences. Our hierarchical POI matching framework generated highly confident food POIs by conflating datasets and identified some food POIs that are unique to specific data sources. Triangulating RFE data can reduce uncertain and invalid POI listings when representing food environment using multiple data sources. Studies investigating associations between food environment and health outcomes may benefit from improved quality of RFE.

Exploring the impact of environmental exposure changes on metabolic biomarkers: A 6-month GPS-GIS study among women with overweight or obesity.

BACKGROUND: Little is known about the impact of environmental exposure change on metabolic biomarkers associated with cancer risk. Furthermore, this limited epidemiological evidence on metabolic biomarkers focused on residential exposure, without considering the activity space which can be done by modelling dynamic exposures. In this longitudinal study, we aimed to investigate the impact of environmental exposures change on metabolic biomarkers using GPS-GIS based measurements. METHODS: Among two weight loss interventions, the Reach for Health and the MENU studies, which included ∼460 women at risk of breast cancer or breast cancer survivors residing in Southern California, three metabolic biomarkers (insulin resistance, fasting glucose, and C-reactive protein) were assessed. Dynamic GPS-GIS based exposure to green spaces, recreation, walkability, NO2, and PM2.5 were calculated at baseline and 6 months follow-up using time-weighted spatial averaging. Generalized estimating equations models were used to examine the relationship between changes in environmental exposures and biomarker levels over time. RESULTS: Overall, six-month environmental exposure change was not associated with metabolic biomarkers change. Stratified analyses by level of environmental exposures at baseline revealed that reduced NO2 and PM2.5 exposure was associated with reduced fasting glucose concentration among women living in a healthier environment at baseline (ß -0.010, 95%CI -0.025, 0.005; ß -0.019, 95%CI -0.034, -0.003, respectively). Women living in poor environmental conditions at baseline and exposed to greener environments had decreased C-reactive protein concentrations (ß -1.001, 95%CI -1.888, -0.131). CONCLUSIONS: The impact of environmental exposure changes on metabolic biomarkers over time may be modified by baseline exposure conditions.

Aircraft and road traffic noise, insulin resistance, and diabetes: The role of neighborhood socioeconomic status in San Diego County.

Evidence linking traffic noise to insulin resistance and diabetes is limited and unanswered questions remain regarding the potential effect modification by neighborhood socioeconomic status (nSES). We aimed to assess socioeconomic inequalities in noise exposure, whether road and aircraft noise exposures were associated with insulin resistance or diabetes, and whether nSES modified these relationships. Among the Community of Mine Study in San Diego County, road and aircraft noise exposure at enrollment was calculated based on the static (participant's administrative boundary, and circular buffer around participant homes), and dynamic (mobility data by global positioning system, GPS) spatio-temporal aggregation methods. Associations of noise with insulin resistance (HOMA-IR) or type 2 diabetes (T2DM) were quantified using generalized estimating equation models adjusted for sex, age, ethnicity, individual income, and air pollution (nitrogen dioxide) exposure. Additive interaction between noise and nSES was assessed. Among 573 participants (mean age 58.7 y), participants living in low nSES were exposed to higher levels of aircraft and road noise using noise level at the census tract, circular buffer, or Kernel Density Estimation (KDE) of GPS data. Participants exposed to road noise greater or equal to the median (53 dB(A)) at the census tract and living in low nSES had an increased level of insulin resistance (ß = 0.15, 95%CI: -0.04, 0.34) and higher odds of T2DM (Odds Ratio = 2.34, 95%CI: 1.12, 4.90). A positive additive interaction was found as participants living in low nSES had higher odds of T2DM. The impact of noise exposure on insulin resistance and T2DM differs substantially by nSES. Public health benefits of reducing exposure to road or aircraft noise would be larger in individuals living in low nSES.

Neighborhood disadvantage is associated with KRAS-mutated non-small cell lung cancer risk.

PURPOSE: It remains unclear why individuals living in disadvantaged neighborhoods have shorter non-small cell lung cancer (NSCLC) survival. It is possible that living in these deprived areas is linked with increased risk of developing aggressive NSCLC biology. Here, we explored the association of somatic KRAS mutations, which are associated with shorter survival in NSCLC patients, and 11 definitions of neighborhood disadvantage spanning socioeconomic and structural environmental elements. METHODS: We analyzed data from 429 NSCLC patients treated at a Comprehensive Cancer Center from 2015 to 2018. Data were abstracted from medical records and each patient's home address was used to assign publicly available indices of neighborhood disadvantage. Prevalence Ratios (PRs) for the presence of somatic KRAS mutations were estimated using modified Poisson regression models adjusted for age, sex, smoking status, race/ethnicity, educational attainment, cancer stage, and histology. RESULTS: In the NSCLC cohort, 29% had KRAS mutation-positive tumors. We found that five deprivation indices of socioeconomic disadvantage were associated with KRAS mutation. A one decile increase in several of these socioeconomic disadvantage indices was associated with a 1.06 to 1.14 increased risk of KRAS mutation. Measures of built structural environment were not associated with KRAS mutation status. CONCLUSION: Socioeconomic disadvantage at the neighborhood level is associated with higher risk of KRAS mutation while disadvantage related to built environmental structural measures was inversely associated. Our results indicate not only that neighborhood disadvantage may contribute to aggressive NSCLC biology, but the pathways linking biology to disadvantage are likely operating through socioeconomic-related stress.

Accounting for space, time, and behavior using GPS derived dynamic measures of environmental exposure.

Time-weighted spatial averaging approaches (TWSA) are an increasingly utilized method for calculating exposure using global positioning system (GPS) mobility data for health-related research. They can provide a time-weighted measure of exposure, or dose, to various environments or health hazards. However, little work has been done to compare existing methodologies, nor to assess how sensitive these methods are to mobility data inputs (e.g., walking vs driving), the type of environmental data being assessed as the exposure (e.g., continuous surfaces vs points of interest), and underlying point-pattern clustering of participants (e.g., if a person is highly mobile vs predominantly stationary). Here we contrast three TWSA approaches that have been previously used or recently introduced in the literature: Kernel Density Estimation (KDE), Density Ranking (DR), and Point Overlay (PO). We feed GPS and accelerometer data from 602 participants through each method to derive time-weighted activity spaces, comparing four mobility behaviors: all movement, stationary time, walking time, and in-vehicle time. We then calculate exposure values derived from the various TWSA activity spaces with four environmental layer data types (point, line, area, surface). Similarities and differences across TWSA derived exposures for the sample and between individuals are explored, and we discuss interpretation of TWSA outputs providing recommendations for researchers seeking to apply these methods to health-related studies.

How do environmental characteristics jointly contribute to cardiometabolic health? A quantile g-computation mixture analysis.

Accumulating evidence links cardiometabolic health with social and environmental neighborhood exposures, which may contribute to health inequities. We examined whether environmental characteristics were individually or jointly associated with insulin resistance, hypertension, obesity, type 2 diabetes, and metabolic syndrome in San Diego County, CA. As part of the Community of Mine Study, cardiometabolic outcomes of insulin resistance, hypertension, BMI, diabetes, and metabolic syndrome were collected in 570 participants. Seven census tract level characteristics of participants' residential environment were assessed and grouped as follows: economic, education, health care access, neighborhood conditions, social environment, transportation, and clean environment. Generalized estimating equation models were performed, to take into account the clustered nature of the data and to estimate ß or relative risk (RR) and 95 % confidence intervals (CIs) between each of the seven environmental characteristics and cardiometabolic outcomes. Quantile g-computation was used to examine the association between the joint effect of a simultaneous increase in all environmental characteristics and cardiometabolic outcomes. Among 570 participants (mean age 58.8 ± 11 years), environmental economic, educational and health characteristics were individually associated with insulin resistance, diabetes, obesity, and metabolic syndrome. In the mixture analyses, a joint quartile increase in all environmental characteristics (i.e., improvement) was associated with decreasing insulin resistance (ß, 95 %CI: -0.09, -0.18-0.01)), risk of diabetes (RR, 95 %CI: 0.59, 0.36-0.98) and obesity (RR, 95 %CI: 0.81, 0.64-1.02). Environmental characteristics synergistically contribute to cardiometabolic health and independent analysis of these determinants may not fully capture the potential health impact of social and environmental determinants of health.

The Role of Neighborhood Air Pollution Exposure on Somatic Non-Small Cell Lung Cancer Mutations in the Los Angeles Basin (2013-2018).

Limited previous work has identified a relationship between exposure to ambient air pollution and aggressive somatic lung tumor mutations. More work is needed to confirm this relationship, especially using spatially resolved air pollution. We aimed to quantify the association between different air pollution metrics and aggressive tumor biology. Among patients treated at City of Hope Comprehensive Cancer Center in Duarte, CA (2013-2018), three non-small cell lung cancer somatic tumor mutations, TP53, KRAS, and KRAS G12C/V, were documented. PM2.5 exposure was assessed using state-of-the art ensemble models five and ten years before lung cancer diagnosis. We also explored the role of NO2 using inverse-distance-weighting approaches. We fitted logistic regression models to estimate odds ratio (OR) and their 95% confidence intervals (CIs). Among 435 participants (median age: 67, female: 51%), an IQR increase in NO2 exposure (3.5 µg/m3) five years before cancer diagnosis was associated with an increased risk in TP53 mutation (OR, 95% CI: 1.30, 0.99-1.71). We found an association between highly-exposed participants to PM2.5 (>12 µg/m3) five and ten years before cancer diagnosis and TP53 mutation (OR, 95% CI: 1.61, 0.95-2.73; 1.57, 0.93-2.64, respectively). Future studies are needed to confirm this association and better understand how air pollution impacts somatic profiles and the molecular mechanisms through which they operate.

Spatial and temporal pattern of cannabis use disorder in California 2010-2019.

As cannabis use is being legalized in an increasing number of states, it is important to understand the changing dynamic of the risk in cannabis use disorder (CUD). Shape-based time-series clustering was used to identify ZIP Code Tabulation Areas (ZCTAs) with similar changing pattern in CUD over time. We conducted a cross-sectional logistic regression analysis to investigate the most recent ZCTA socio-demographic characteristics in relation to the changing CUD rates. The emergency discharge rates generally increased during 2010-2016. Increase during 2017-2019 was found in Sacramento and Santa Barbara County. Approximately 13% of ZCTAs showed an increasing trend of hospitalization discharge during 2017-2019. Males and non-Hispanic Black had larger increase than other groups during 2017-2019. The recent growing trend was found associated with greater racial diversity and rural ZCTAs. The findings from this study hold promise for local public health officials to adjust the cannabis intervention strategies in target districts and improve overall health outcomes.

Enhancing Nutrition Care Through Real-Time, Sensor-Based Capture of Eating Occasions: A Scoping Review.

As food intake patterns become less structured, different methods of dietary assessment may be required to capture frequently omitted snacks, smaller meals, and the time of day when they are consumed. Incorporating sensors that passively and objectively detect eating behavior may assist in capturing these eating occasions into dietary assessment methods. The aim of this study was to identify and collate sensor-based technologies that are feasible for dietitians to use to assist with performing dietary assessments in real-world practice settings. A scoping review was conducted using the PRISMA extension for scoping reviews (PRISMA-ScR) framework. Studies were included if they were published between January 2016 and December 2021 and evaluated the performance of sensor-based devices for identifying and recording the time of food intake. Devices from included studies were further evaluated against a set of feasibility criteria to determine whether they could potentially be used to assist dietitians in conducting dietary assessments. The feasibility criteria were, in brief, consisting of an accuracy ≥80%; tested in settings where subjects were free to choose their own foods and activities; social acceptability and comfort; a long battery life; and a relatively rapid detection of an eating episode. Fifty-four studies describing 53 unique devices and 4 device combinations worn on the wrist (n = 18), head (n = 16), neck (n = 9), and other locations (n = 14) were included. Whilst none of the devices strictly met all feasibility criteria currently, continuous refinement and testing of device software and hardware are likely given the rapidly changing nature of this emerging field. The main reasons devices failed to meet the feasibility criteria were: an insufficient or lack of reporting on battery life (91%), the use of a limited number of foods and behaviors to evaluate device performance (63%), and the device being socially unacceptable or uncomfortable to wear for long durations (46%). Until sensor-based dietary assessment tools have been designed into more inconspicuous prototypes and are able to detect most food and beverage consumption throughout the day, their use will not be feasible for dietitians in practice settings.

Air pollution and metabolic disorders: Dynamic versus static measures of exposure among Hispanics/Latinos and non-Hispanics.

INTRODUCTION: Exposure to air pollution disproportionately affects racial/ethnic minorities that could contribute to health inequalities including metabolic disorders. However, most existing studies used a static assessment of air pollution exposure (mostly using the residential address) and do not account for activity space when modelling exposure to air pollution. The aim of this study is to understand how exposure to air pollution impacts metabolic disorders biomarkers, how this effect differs according to ethnicity, and for the first time compare these findings with two methods of exposure assessment: dynamic and static measures. METHODS: Among the Community of Mine study, a cross-sectional study conducted in San Diego County, insulin resistance, diabetes, hypertension, obesity, dyslipidemia, and metabolic syndrome (MetS) were assessed. Exposure to air pollution (PM2.5, NO2, traffic) was calculated using static measures around the home, and dynamic measures of mobility derived from Global Positioning Systems (GPS) traces using kernel density estimators to account for exposure variability across space and time. Associations of air pollution with metabolic disorders were quantified using generalized estimating equation models to account for the clustered nature of the data. RESULTS: Among 552 participants (mean age 58.7 years, 42% Hispanic/Latino), Hispanics/Latinos had a higher exposure to PM2.5 compared to non-Hispanics using static measures. In contrast, Hispanics/Latinos had less exposure to PM2.5 using dynamic measures. For all participants, higher dynamic exposure to PM2.5 and NO2 was associated with increased insulin resistance and cholesterol levels, and increased risk of obesity, dyslipidemia and MetS (RR 1.17, 95% CI: 1.07-1.28; RR 1.21, 95% CI: 1.12-1.30, respectively). The association between dynamic PM2.5 exposure and MetS differed by Hispanic/Latino ethnicity. CONCLUSION: These results highlight the importance of considering people's daily mobility in assessing the impact of air pollution on health.

Perspective: A Framework for Addressing Dynamic Food Consumption Processes.

The study of food consumption, diet, and related concepts is motivated by diverse goals, including understanding why food consumption impacts our health, and why we eat the foods we do. These varied motivations can make it challenging to define and measure consumption, as it can be specified across nearly infinite dimensions-from micronutrients to carbon footprint to food preparation. This challenge is amplified by the dynamic nature of food consumption processes, with the underlying phenomena of interest often based on the nature of repeated interactions with food occurring over time. This complexity underscores a need to not only improve how we measure food consumption but is also a call to support theoreticians in better specifying what, how, and why food consumption occurs as part of processes, as a prerequisite step to rigorous measurement. The purpose of this Perspective article is to offer a framework, the consumption process framework, as a tool that researchers in a theoretician role can use to support these more robust definitions of consumption processes. In doing so, the framework invites theoreticians to be a bridge between practitioners who wish to measure various aspects of food consumption and methodologists who can develop measurement protocols and technologies that can support measurement when consumption processes are clearly defined. In the paper we justify the need for such a framework, introduce the consumption process framework, illustrate the framework via a use case, and discuss existing technologies that enable the use of this framework and, by extension, more rigorous study of consumption. This consumption process framework demonstrates how theoreticians could fundamentally shift how food consumption is defined and measured towards more rigorous study of what, how, and why food is eaten as part of dynamic processes and a deeper understanding of linkages between behavior, food, and health.

Device-Measured and Self-Reported Active Travel Associations with Cardiovascular Disease Risk Factors in an Ethnically Diverse Sample of Adults.

Active travel (AT) provides an opportunity to alleviate the physical inactivity and climate crises contributing to the global chronic disease burden, including cardiovascular diseases (CVD). Though AT shows promising links to reduced CVD risk, prior studies relied on self-reported AT assessment. In the present study, device-measured and self-reported AT were compared across population subgroups and relationships with CVD risk biomarkers were evaluated for both measures. The study recruited an ethnically diverse sample (N = 602, mean age 59 years, 42% Hispanic/Latino ethnicity) from neighborhoods that varied by walkability and food access. AT was assessed using concurrently collected accelerometer and GPS data and self-report data from a validated survey. Relationships with body mass index (BMI), triglycerides, high-density lipoprotein (HDL) cholesterol, blood pressure (BP), and moderate-to-vigorous physical activity (MVPA) were modeled using multivariable linear regression. Devices captured more AT than did self-report. We found differences in AT measures by population subgroups, including race, ethnicity, education, income, vehicle access, and walkability. Men had more accelerometer-measured MVPA, though women self-reported more daily minutes. Both device and survey AT measures were positively associated with total accelerometer-measured MVPA, though the relationship was stronger with device-measured AT. Device-measured AT was associated with lower BMI. No other CVD risk biomarker was associated with either AT measure. No effect modification by Hispanic/Latino ethnicity was detected. Further studies with device-based measures are warranted to better understand the relationship between AT and cardiovascular health.

Increased Social Interactions Reduce the Association Between Constricted Life-Space and Lower Daily Happiness in Older Adults With and Without HIV: A GPS and Ecological Momentary Assessment Study.

OBJECTIVE: Older persons with human immunodeficiency virus (HIV) (PWH) are particularly susceptible to life-space restrictions. The aims of this study included: 1) using global positioning system (GPS) derived indicators as an assessment of time spent at home among older adults with and without HIV; 2) using ecological momentary assessment (EMA) to examine real-time relationships between life-space, mood (happiness, sadness, anxious), fatigue, and pain; and 3) determining if number of daily social interactions moderated the effect of life-space on mood. METHODS: Eighty-eight older adults (PWH n = 54, HIV-negative n = 34) completed smartphone-based EMA surveys assessing mood, fatigue, pain, and social interactions four times per day for two weeks. Participants' smartphones were GPS enabled throughout the study. Mixed-effects regression models analyzed concurrent and lagged associations among life-space and behavioral indicators of health. RESULTS: PWH spent more of their time at home (79% versus 70%, z = -2.08; p = 0.04) and reported lower mean happiness (3.2 versus 3.7; z = 2.63; p = 0.007) compared to HIV-negative participants. Controlling for covariates, more daily social interactions were associated with higher ratings of real-time happiness (b = 0.12; t = 5.61; df = 1087.9; p< 0.001). Similar findings were seen in lagged analyses: prior day social interactions (b = 0.15; t = 7.3; df = 1024.9; p < 0.0001) and HIV status (b = -0.48; t = -2.56; df = 1026.8; p = 0.01) attenuated the effect of prior day time spent at home on happiness. CONCLUSION: Accounting for engagement in social interactions reduced the significant effect of time spent at home and lower happiness. Interventions targeting social isolation within the context of constricted life-space may be beneficial for increasing positive mood in older adults, and especially relevant to older PWH.

Environmental Exposures during Puberty: Window of Breast Cancer Risk and Epigenetic Damage.

During puberty, a woman's breasts are vulnerable to environmental damage ("window of vulnerability"). Early exposure to environmental carcinogens, endocrine disruptors, and unhealthy foods (refined sugar, processed fats, food additives) are hypothesized to promote molecular damage that increases breast cancer risk. However, prospective human studies are difficult to perform and effective interventions to prevent these early exposures are lacking. It is difficult to prevent environmental exposures during puberty. Specifically, young women are repeatedly exposed to media messaging that promotes unhealthy foods. Young women living in disadvantaged neighborhoods experience additional challenges including a lack of access to healthy food and exposure to contaminated air, water, and soil. The purpose of this review is to gather information on potential exposures during puberty. In future directions, this information will be used to help elementary/middle-school girls to identify and quantitate environmental exposures and develop cost-effective strategies to reduce exposures.

An Online Geographic Data Visualization Tool to Relate Preterm Births to Environmental Factors.

Preterm birth (<37 weeks gestation) continues to be a significant cause of disease and death in the United States. Its complex causes are associated with several genetic, biological, environmental, and sociodemographic factors. Organizing and visualizing various data that may be related to preterm birth is an essential step for pattern exploration and hypothesis generation and presents an opportunity to increase public and stakeholder involvement. In this article, we describe a collaborative effort to create an online geographic data visualization tool using open software to explore preterm birth in Fresno County, where rates are the highest in California. The tool incorporates information on births, environmental exposures, sociodemographic characteristics, the built environment, and access to care. We describe data sets used to build the tool, the data-hosting platform, and the process used to engage stakeholders in its creation. We highlight an important example of how collaboration can increase the utility of geographic data visualization to improve public health and address health equity in birth outcomes.

Characteristics Of Likely Precision Medicine Initiative Participants Drawn From A Large Blood Donor Population.

A goal of the Precision Medicine Initiative All of Us Research Program (AoURP) is recruitment of participants who reflect the diversity of the US. Recruitment from among blood bank donors, which may better reflect the demographic makeup of local communities, is one proposed strategy. We evaluated this strategy by analyzing the results of a survey of San Diego Blood Bank donors conducted in November 2015. Whites were more likely than nonwhites to respond to the survey (7.1 percent versus 3.9 percent). However, race was not a significant predictor of interest in participating in precision medicine research. Using census data linked to donors' ZIP codes, we also found that people who indicated interest in research participation were more likely to come from regions with higher educational attainment. Although blood banks represent a viable recruitment strategy for AoURP, our findings indicate that bias toward inclusion of whites and more highly educated people persists.

The reliability of tweets as a supplementary method of seasonal influenza surveillance.

BACKGROUND: Existing influenza surveillance in the United States is focused on the collection of data from sentinel physicians and hospitals; however, the compilation and distribution of reports are usually delayed by up to 2 weeks. With the popularity of social media growing, the Internet is a source for syndromic surveillance due to the availability of large amounts of data. In this study, tweets, or posts of 140 characters or less, from the website Twitter were collected and analyzed for their potential as surveillance for seasonal influenza. OBJECTIVE: There were three aims: (1) to improve the correlation of tweets to sentinel-provided influenza-like illness (ILI) rates by city through filtering and a machine-learning classifier, (2) to observe correlations of tweets for emergency department ILI rates by city, and (3) to explore correlations for tweets to laboratory-confirmed influenza cases in San Diego. METHODS: Tweets containing the keyword "flu" were collected within a 17-mile radius from 11 US cities selected for population and availability of ILI data. At the end of the collection period, 159,802 tweets were used for correlation analyses with sentinel-provided ILI and emergency department ILI rates as reported by the corresponding city or county health department. Two separate methods were used to observe correlations between tweets and ILI rates: filtering the tweets by type (non-retweets, retweets, tweets with a URL, tweets without a URL), and the use of a machine-learning classifier that determined whether a tweet was "valid", or from a user who was likely ill with the flu. RESULTS: Correlations varied by city but general trends were observed. Non-retweets and tweets without a URL had higher and more significant (P<.05) correlations than retweets and tweets with a URL. Correlations of tweets to emergency department ILI rates were higher than the correlations observed for sentinel-provided ILI for most of the cities. The machine-learning classifier yielded the highest correlations for many of the cities when using the sentinel-provided or emergency department ILI as well as the number of laboratory-confirmed influenza cases in San Diego. High correlation values (r=.93) with significance at P<.001 were observed for laboratory-confirmed influenza cases for most categories and tweets determined to be valid by the classifier. CONCLUSIONS: Compared to tweet analyses in the previous influenza season, this study demonstrated increased accuracy in using Twitter as a supplementary surveillance tool for influenza as better filtering and classification methods yielded higher correlations for the 2013-2014 influenza season than those found for tweets in the previous influenza season, where emergency department ILI rates were better correlated to tweets than sentinel-provided ILI rates. Further investigations in the field would require expansion with regard to the location that the tweets are collected from, as well as the availability of more ILI data.

The complex relationship of realspace events and messages in cyberspace: case study of influenza and pertussis using tweets.

BACKGROUND: Surveillance plays a vital role in disease detection, but traditional methods of collecting patient data, reporting to health officials, and compiling reports are costly and time consuming. In recent years, syndromic surveillance tools have expanded and researchers are able to exploit the vast amount of data available in real time on the Internet at minimal cost. Many data sources for infoveillance exist, but this study focuses on status updates (tweets) from the Twitter microblogging website. OBJECTIVE: The aim of this study was to explore the interaction between cyberspace message activity, measured by keyword-specific tweets, and real world occurrences of influenza and pertussis. Tweets were aggregated by week and compared to weekly influenza-like illness (ILI) and weekly pertussis incidence. The potential effect of tweet type was analyzed by categorizing tweets into 4 categories: nonretweets, retweets, tweets with a URL Web address, and tweets without a URL Web address. METHODS: Tweets were collected within a 17-mile radius of 11 US cities chosen on the basis of population size and the availability of disease data. Influenza analysis involved all 11 cities. Pertussis analysis was based on the 2 cities nearest to the Washington State pertussis outbreak (Seattle, WA and Portland, OR). Tweet collection resulted in 161,821 flu, 6174 influenza, 160 pertussis, and 1167 whooping cough tweets. The correlation coefficients between tweets or subgroups of tweets and disease occurrence were calculated and trends were presented graphically. RESULTS: Correlations between weekly aggregated tweets and disease occurrence varied greatly, but were relatively strong in some areas. In general, correlation coefficients were stronger in the flu analysis compared to the pertussis analysis. Within each analysis, flu tweets were more strongly correlated with ILI rates than influenza tweets, and whooping cough tweets correlated more strongly with pertussis incidence than pertussis tweets. Nonretweets correlated more with disease occurrence than retweets, and tweets without a URL Web address correlated better with actual incidence than those with a URL Web address primarily for the flu tweets. CONCLUSIONS: This study demonstrates that not only does keyword choice play an important role in how well tweets correlate with disease occurrence, but that the subgroup of tweets used for analysis is also important. This exploratory work shows potential in the use of tweets for infoveillance, but continued efforts are needed to further refine research methods in this field.