Search

ImportancePrior research has established that Hispanic and non-Hispanic Black residents in the US experienced substantially higher COVID-19 mortality rates in 2020 than non-Hispanic White residents owing to structural racism. In 2021, these disparities decreased.ObjectiveTo assess to what extent national decreases in racial and ethnic disparities in COVID-19 mortality between the initial pandemic wave and subsequent Omicron wave reflect reductions in mortality vs other factors, such as the pandemic’s changing geography.Design, Setting, and ParticipantsThis cross-sectional study was conducted using data from the US Centers for Disease Control and Prevention for COVID-19 deaths from March 1, 2020, through February 28, 2022, among adults aged 25 years and older residing in the US. Deaths were examined by race and ethnicity across metropolitan and nonmetropolitan areas, and the national decrease in racial and ethnic disparities between initial and Omicron waves was decomposed. Data were analyzed from June 2021 through March 2023.ExposuresMetropolitan vs nonmetropolitan areas and race and ethnicity.Main Outcomes and MeasuresAge-standardized death rates.ResultsThere were death certificates for 977 018 US adults aged 25 years and older (mean [SD] age, 73.6 [14.6] years; 435 943 female [44.6%]; 156 948 Hispanic [16.1%], 140 513 non-Hispanic Black [14.4%], and 629 578 non-Hispanic White [64.4%]) that included a mention of COVID-19. The proportion of COVID-19 deaths among adults residing in nonmetropolitan areas increased from 5944 of 110 526 deaths (5.4%) during the initial wave to a peak of 40 360 of 172 515 deaths (23.4%) during the Delta wave; the proportion was 45 183 of 210 554 deaths (21.5%) during the Omicron wave. The national disparity in age-standardized COVID-19 death rates per 100 000 person-years for non-Hispanic Black compared with non-Hispanic White adults decreased from 339 to 45 deaths from the initial to Omicron wave, or by 293 deaths. After standardizing for age and racial and ethnic differences by metropolitan vs nonmetropolitan residence, increases in death rates among non-Hispanic White adults explained 120 deaths/100 000 person-years of the decrease (40.7%); 58 deaths/100 000 person-years in the decrease (19.6%) were explained by shifts in mortality to nonmetropolitan areas, where a disproportionate share of non-Hispanic White adults reside. The remaining 116 deaths/100 000 person-years in the decrease (39.6%) were explained by decreases in death rates in non-Hispanic Black adults.Conclusions and RelevanceThis study found that most of the national decrease in racial and ethnic disparities in COVID-19 mortality between the initial and Omicron waves was explained by increased mortality among non-Hispanic White adults and changes in the geographic spread of the pandemic. These findings suggest that despite media reports of a decline in disparities, there is a continued need to prioritize racial health equity in the pandemic response.

Prior research has established that American Indian, Alaska Native, Black, Hispanic, and Pacific Islander populations in the United States have experienced substantially higher mortality rates from Covid-19 compared to non-Hispanic white residents during the first year of the pandemic. What remains less clear is how mortality rates have changed for each of these racial/ethnic groups during 2021, given the increasing prevalence of vaccination. In particular, it is unknown how these changes in mortality have varied geographically. In this study, we used provisional data from the National Center for Health Statistics (NCHS) to produce age-standardized estimates of Covid-19 mortality by race/ethnicity in the United States from March 2020 to February 2022 in each metro-nonmetro category, Census region, and Census division. We calculated changes in mortality rates between the first and second years of the pandemic and examined mortality changes by month. We found that when Covid-19 first affected a geographic area, non-Hispanic Black and Hispanic populations experienced extremely high levels of Covid-19 mortality and racial/ethnic inequity that were not repeated at any other time during the pandemic. Between the first and second year of the pandemic, racial/ethnic inequities in Covid-19 mortality decreased—but were not eliminated—for Hispanic, non-Hispanic Black, and non-Hispanic AIAN residents. These inequities decreased due to reductions in mortality for these populations alongside increases in non-Hispanic white mortality. Though racial/ethnic inequities in Covid-19 mortality decreased, substantial inequities still existed in most geographic areas during the pandemic’s second year: Non-Hispanic Black, non-Hispanic AIAN, and Hispanic residents reported higher Covid-19 death rates in rural areas than in urban areas, indicating that these communities are facing serious public health challenges. At the same time, the non-Hispanic white mortality rate worsened in rural areas during the second year of the pandemic, suggesting there may be unique factors driving mortality in this population. Finally, vaccination rates were associated with reductions in Covid-19 mortality for Hispanic, non-Hispanic Black, and non-Hispanic white residents, and increased vaccination may have contributed to the decreases in racial/ethnic inequities in Covid-19 mortality observed during the second year of the pandemic. Despite reductions in mortality, Covid-19 mortality remained elevated in nonmetro areas and increased for some racial/ethnic groups, highlighting the need for increased vaccination delivery and equitable public health measures especially in rural communities. Taken together, these findings highlight the continued need to prioritize health equity in the pandemic response and to modify the structures and policies through which systemic racism operates and has generated racial health inequities.

Female genital mutilation/cutting (FGM/C) describes several procedures that involve injury to the vulva or vagina for nontherapeutic reasons. Though at least 200 million women and girls living in 30 countries have undergone FGM/C, there is a paucity of studies focused on public perception of FGM/C. We used machine learning methods to characterize discussion of FGM/C on Twitter in English from 2015 to 2020. Twitter has emerged in recent years as a source for seeking and sharing health information and misinformation. We extracted text metadata from user profiles to characterize the individuals and locations involved in conversations about FGM/C. We extracted major discussion themes from posts using correlated topic modeling. Finally, we extracted features from posts and applied random forest models to predict user engagement. The volume of tweets addressing FGM/C remained fairly stable across years. Conversation was mostly concentrated among the United States and United Kingdom through 2017, but shifted to Nigeria and Kenya in 2020. Some of the discussion topics associated with FGM/C across years included Islam, International Day of Zero Tolerance, current news stories, education, activism, male circumcision, human rights, and feminism. Tweet length and follower count were consistently strong predictors of engagement. Our findings suggest that (1) discussion about FGM/C has not evolved significantly over time, (2) the majority of the conversation about FGM/C on English-speaking Twitter is advocating for an end to the practice, (3) supporters of Donald Trump make up a substantial voice in the conversation about FGM/C, and (4) understanding the nuances in how people across cultures refer to and discuss FGM/C could be important for the design of public health communication and intervention.Author SummaryFemale genital mutilation/cutting (FGM/C) refers to procedures that involve injury to the vulva and vagina for nontherapeutic reasons. At least 200 million women and girls living in 30 countries have undergone FGM/C. In many countries, there is not public perception research on FGM/C. When research is conducted, survey respondents may not answer truthfully about their opinions. We used Twitter data from 2015 to 2020 to better understand how the public thinks about FGM/C. We found that the conversation has generally stayed the same in this time period. Early in the study, users discussing FGM/C were mostly from the United States and United Kingdom, but later in the study, users were mostly from Nigeria and Kenya. Many people posted about stopping FGM/C, especially on the International Day of Zero Tolerance for Female Genital Mutilation and using the hashtag #endfgm. Some people used FGM/C to justify Islamophobia and connected it with other practices like honor killings and acid attacks. We hope that this research can be used to inform communication about FGM/C in different countries to improve understanding about the practice and, design culturally effective campaigns to end FGM/C.

Data Science can be used to address racial health inequities. However, a wealth of scholarship has shown that there are many ethical challenges with using Data Science to address social problems. To develop a Data Science focused on racial health equity, we need the data, methods, application, and communication approaches to be antiracist and focused on serving minoritized groups that have long-standing worse health indicators than majority groups. In this perspective, we propose eight tenets that could shape a Data Science for Racial Health Equity research framework.

Physical distancing has been the primary strategy to control COVID-19 in the United States. We used mobility data from a large, anonymized sample of smartphone users to assess the relationship between neighbourhood income and physical distancing during the pandemic. We found a strong gradient between neighbourhood income and physical distancing. Individuals in high-income neighbourhoods increased their days at home substantially more than individuals in low-income neighbourhoods did. Residents of low-income neighbourhoods were more likely to work outside the home, compared to residents in higher-income neighbourhoods, but were not more likely to visit locations such as supermarkets, parks and hospitals. Finally, we found that state orders were only associated with small increases in staying home in low-income neighbourhoods. Our findings indicate that people in lower-income neighbourhoods have faced barriers to physical distancing, particularly needing to work outside the home, and that state physical distancing policies have not mitigated these disparities.

ImportanceRacist policies (such as redlining) create inequities in the built environment, producing racially and ethnically segregated communities, poor housing conditions, unwalkable neighborhoods, and general disadvantage. Studies on built environment disparities are usually limited to measures and data that are available from existing sources or can be manually collected.ObjectiveTo use built environment indicators generated from online street-level images to investigate the association among neighborhood racial and ethnic composition, the built environment, and health outcomes across urban areas in the US.Design, Setting, and ParticipantsThis cross-sectional study was conducted using built environment indicators derived from 164 million Google Street View images collected from November 1 to 30, 2019. Race, ethnicity, and socioeconomic data were obtained from the 2019 American Community Survey (ACS) 5-year estimates; health outcomes were obtained from the Centers for Disease Control and Prevention 2020 Population Level Analysis and Community Estimates (PLACES) data set. Multilevel modeling and mediation analysis were applied. A total of 59 231 urban census tracts in the US were included. The online images and the ACS data included all census tracts. The PLACES data comprised survey respondents 18 years or older. Data were analyzed from May 23 to November 16, 2022.Main Outcomes and MeasuresModel-estimated association between image-derived built environment indicators and census tract (neighborhood) racial and ethnic composition, and the association of the built environment with neighborhood racial composition and health.ResultsThe racial and ethnic composition in the 59 231 urban census tracts was 1 160 595 (0.4%) American Indian and Alaska Native, 53 321 345 (19.5%) Hispanic, 462 259 (0.2%) Native Hawaiian and other Pacific Islander, 17 166 370 (6.3%) non-Hispanic Asian, 35 985 480 (13.2%) non-Hispanic Black, and 158 043 260 (57.7%) non-Hispanic White residents. Compared with other neighborhoods, predominantly White neighborhoods had fewer dilapidated buildings and more green space indicators, usually associated with good health, and fewer crosswalks (eg, neighborhoods with predominantly minoritized racial or ethnic groups other than Black residents had 6% more dilapidated buildings than neighborhoods with predominantly White residents). Moreover, the built environment indicators partially mediated the association between neighborhood racial and ethnic composition and health outcomes, including diabetes, asthma, and sleeping problems. The most significant mediator was non–single family homes (a measure associated with homeownership), which mediated the association between neighborhoods with predominantly minority racial or ethnic groups other than Black residents and sleeping problems by 12.8% and the association between unclassified neighborhoods and asthma by 24.2%.Conclusions and RelevanceThe findings in this cross-sectional study suggest that large geographically representative data sets, if used appropriately, may provide novel insights on racial and ethnic health inequities. Quantifying the impact of structural racism on social determinants of health is one step toward developing policies and interventions to create equitable built environment resources.

Objectives Access to safe and nutritious food is essential for good health. However, food can become unsafe due to contamination with pathogens, chemicals or toxins, or mislabeling of allergens. Illness resulting from the consumption of unsafe foods is a global health problem. Here, we develop a machine learning approach for detecting reports of unsafe food products in consumer product reviews from Amazon.com. Materials and Methods We linked Amazon.com food product reviews to Food and Drug Administration (FDA) food recalls from 2012 to 2014 using text matching approaches in a PostGres relational database. We applied machine learning methods and over- and under-sampling methods to the linked data to automate the detection of reports of unsafe food products. Results Our data consisted of 1 297 156 product reviews from Amazon.com. Only 5149 (0.4%) were linked to recalled food products. Bidirectional Encoder Representation from Transformations performed best in identifying unsafe food reviews, achieving an F1 score, precision and recall of 0.74, 0.78, and 0.71, respectively. We also identified synonyms for terms associated with FDA recalls in more than 20 000 reviews, most of which were associated with nonrecalled products. This might suggest that many more products should have been recalled or investigated. Discussion and Conclusion Challenges to improving food safety include, urbanization which has led to a longer food chain, underreporting of illness and difficulty in linking contaminated food to illness. Our approach can improve food safety by enabling early identification of unsafe foods which can lead to timely recall thereby limiting the health and economic impact on the public.

IntroductionPeople in low-income households face a disproportionate burden of health and economic consequences brought on by the COVID-19 pandemic, including COVID-19 and food insufficiency. State minimum wage and paid sick leave policies may affect whether people are vulnerable to employment and health shocks to income and affect food insufficiency.MethodsWe evaluated the relationship between state minimum wage policies and the outcome of household food insufficiency among participants younger than 65 during the COVID-19 pandemic. We used data from biweekly, state representative Census Pulse surveys conducted between August 19 and December 21, 2020. We conducted analyses in the full population under age 65 years, who are most likely to work, and in households with children. The primary exposure was state minimum wage policies in four categories: less than $8.00, $8.00 to $9.99, $10.00 to $11.99, and $12.00 or more. A secondary exposure was missing work due to COVID-19, interacted with whether participants reported not having paid sick leave. Food insufficiency was defined as sometimes or often not having enough to eat in the past seven days. Very low child food sufficiency was defined as children sometimes or often not eating enough in the past seven days because of inability to afford food. We conducted a multivariable modified Poisson regression analysis to estimate adjusted prevalence ratios and marginal effects. We clustered standard errors by state. To adjust for state health and social programs, we adjusted for health insurance and receipt of supplemental nutrition assistance program benefits, unemployment insurance, and stimulus payments as well as for population demographic characteristics associated with food insufficiency. We conducted subgroup analyses among populations most likely to be affected by minimum wage policies: Participants who reported any work in the past seven days, who reported $75,000 in 2019 household income.ResultsIn states with a minimum wage of less than $8.00, 14.3% of participants under age 65 and 16.6% of participants in households with children reported household food insufficiency, while 10.3% of participants reported very low child food sufficiency. A state minimum wage of $12 or more per hour was associated with a 1.83 percentage point reduction in the proportion of households reporting food insufficiency relative to a minimum wage of less than $8.00 per hour (95% CI: −2.67 to −0.99 percentage points). In households with children, a state minimum wage of $12 or more per hour was associated with a 2.13 percentage point reduction in household food insufficiency (95% CI: −3.25 to −1.00 percentage points) and in very low child food sufficiency (−1.16 percentage points, 95% CI: −1.69 to −0.63 percentage points) relative to a state minimum wage of less than $8.00 per hour. Minimum wages of $8.00 to $9.99 and $10.00 to $11.99 were not associated with changes in child food insufficiency or very low child food sufficiency relative to less than $8.00 per hour. Subgroup analyses and sensitivity analyses were consistent with the main results. Estimates were of a lesser magnitude (DiscussionFood insufficiency is high in all households and even more so in households with children during the COVID-19 pandemic. Living in a state with at least a $12 minimum wage was associated with a decrease in the proportion of people reporting food insufficiency during the COVID-19 pandemic. Not having paid leave was associated with increases in food insufficiency among people who reported missing work due to COVID-19 illness. Policymakers may wish to consider raising the minimum wage and paid sick leave as approaches to reducing food insufficiency during and after the COVID-19 pandemic.

Background The epidemic of misinformation about COVID-19 transmission, prevention, and treatment has been going on since the start of the pandemic. However, data on the exposure and impact of misinformation is not readily available. Objective We aim to characterize and compare the start, peak, and doubling time of COVID-19 misinformation topics across 8 countries using an exponential growth model usually employed to study infectious disease epidemics. Methods COVID-19 misinformation topics were selected from the World Health Organization Mythbusters website. Data representing exposure was obtained from the Google Trends application programming interface for 8 English-speaking countries. Exponential growth models were used in modeling trends for each country. Results Searches for “coronavirus AND 5G” started at different times but peaked in the same week for 6 countries. Searches for 5G also had the shortest doubling time across all misinformation topics, with the shortest time in Nigeria and South Africa (approximately 4-5 days). Searches for “coronavirus AND ginger” started at the same time (the week of January 19, 2020) for several countries, but peaks were incongruent, and searches did not always grow exponentially after the initial week. Searches for “coronavirus AND sun” had different start times across countries but peaked at the same time for multiple countries. Conclusions Patterns in the start, peak, and doubling time for “coronavirus AND 5G” were different from the other misinformation topics and were mostly consistent across countries assessed, which might be attributable to a lack of public understanding of 5G technology. Understanding the spread of misinformation, similarities and differences across different contexts can help in the development of appropriate interventions for limiting its impact similar to how we address infectious disease epidemics. Furthermore, the rapid proliferation of misinformation that discourages adherence to public health interventions could be predictive of future increases in disease cases.

Data from non-traditional data sources, such as social media, search engines, and remote sensing, have previously demonstrated utility for disease surveillance. Few studies, however, have focused on countries in Africa, particularly during the SARS-CoV-2 pandemic. In this study, we use searches of COVID-19 symptoms, questions, and at-home remedies submitted to Google to model COVID-19 in South Africa, and assess how well the Google search data forecast short-term COVID-19 trends. Our findings suggest that information seeking trends on COVID-19 could guide models for anticipating COVID-19 trends and coordinating appropriate response measures.

BACKGROUND The prevalence of chronic conditions such as obesity, hypertension, and diabetes is increasing in African countries. Many chronic diseases have been linked to risk factors such as poor diet and physical inactivity. Data for these behavioral risk factors are usually obtained from surveys, which can be delayed by years. Behavioral data from digital sources, including social media and search engines, could be used for timely monitoring of behavioral risk factors. OBJECTIVE The objective of our study was to propose the use of digital data from internet sources for monitoring changes in behavioral risk factors in Africa. METHODS We obtained the adjusted volume of search queries submitted to Google for 108 terms related to diet, exercise, and disease from 2010 to 2016. We also obtained the obesity and overweight prevalence for 52 African countries from the World Health Organization (WHO) for the same period. Machine learning algorithms (ie, random forest, support vector machine, Bayes generalized linear model, gradient boosting, and an ensemble of the individual methods) were used to identify search terms and patterns that correlate with changes in obesity and overweight prevalence across Africa. Out-of-sample predictions were used to assess and validate the model performance. RESULTS The study included 52 African countries. In 2016, the WHO reported an overweight prevalence ranging from 20.9% (95% credible interval [CI] 17.1%-25.0%) to 66.8% (95% CI 62.4%-71.0%) and an obesity prevalence ranging from 4.5% (95% CI 2.9%-6.5%) to 32.5% (95% CI 27.2%-38.1%) in Africa. The highest obesity and overweight prevalence were noted in the northern and southern regions. Google searches for diet-, exercise-, and obesity-related terms explained 97.3% (root-mean-square error [RMSE] 1.15) of the variation in obesity prevalence across all 52 countries. Similarly, the search data explained 96.6% (RMSE 2.26) of the variation in the overweight prevalence. The search terms yoga, exercise, and gym were most correlated with changes in obesity and overweight prevalence in countries with the highest prevalence. CONCLUSIONS Information-seeking patterns for diet- and exercise-related terms could indicate changes in attitudes toward and engagement in risk factors or healthy behaviors. These trends could capture population changes in risk factor prevalence, inform digital and physical interventions, and supplement official data from surveys.

Background Homicides are a major problem in Brazil. Drugs and arms trafficking, and land conflicts are three of the many factors driving homicide rates in Brazil. Understanding long-term spatiotemporal trends and social structural factors associated with homicides in Brazil would be useful for designing policies aimed at reducing homicide rates. Methods We obtained data from 2000 to 2014 from the Brazil Ministry of Health (MOH) Mortality Information System and sociodemographic data from the Brazil Institute of Geography and Statistics (IBGE). First, we quantified the rate of change in homicides at the municipality and state levels. Second, we used principal component regression and k-medoids clustering to examine differences in temporal trends across municipalities. Lastly, we used Bayesian hierarchical space-time models to describe spatio-temporal patterns and to assess the contribution of structural factors. Results There were significant variations in homicide rates across states and municipalities. We noted the largest decrease in homicide rates in the western and southeastern states of Sao Paulo, Rio de Janeiro and Espirito Santo, which coincided with an increase in homicide rates in the northeastern states of Ceará, Alagoas, Paraiba, Rio Grande Norte, Sergipe and Bahia during the fifteen-year period. The decrease in homicides in municipalities with populations of at least 250,000 coincided with an increase in municipalities with 25,000 people or less. Structural factors that predicted municipality-level homicide rates included crude domestic product, urbanization, border with neighboring countries and proportion of population aged fifteen to twenty-nine. Conclusions Our findings support both a dissemination hypothesis and an interiorization hypothesis. These findings should be considered when designing interventions to curb homicide rates.

Although acute respiratory infections are a leading cause of mortality in sub-Saharan Africa, surveillance of diseases such as influenza is mostly neglected. Evaluating the usefulness of influenza-like illness (ILI) surveillance systems and developing approaches for forecasting future trends is important for pandemic preparedness. We applied statistical and machine learning models to forecast 2012 to 2018 trends in ILI cases reported by the Cameroon Ministry of Health (MOH), using Google searches for influenza symptoms, treatments, natural or traditional remedies as well as, infectious diseases with a high burden (i.e., AIDS, malaria, tuberculosis). The variance explained by the models based on Google search data were 87.7%, 79.1% and 52.0% for the whole country, the Littoral and Centre regions respectively. Our study demonstrates the need for developing contextualized approaches when using digital data for disease surveillance and demonstrates the potential usefulness of search data for monitoring ILI in sub-Saharan African countries.

The robust estimate and forecast capability of random forests (RF) has been widely recognized, however this ensemble machine learning method has not been widely used in mosquito-borne disease forecasting. In this study, two sets of RF models were developed for the national and departmental levels in Colombia to predict weekly dengue cases at 12-weeks ahead. A national model based on artificial neural networks (ANN) was also developed and used as a comparator to the RF models. The various predictors included historic dengue cases, satellite-derived estimates for vegetation, precipitation, and air temperature, population counts, income inequality, and education. Our RF model trained on the national data was more accurate for department-specific weekly dengue cases estimation compared to a local model trained only on the department’s data. Additionally, the forecast errors of the national RF model were smaller to those of the national ANN model and were increased with the forecast horizon increasing from one-week ahead (mean absolute error, MAE: 5.80; root mean squared error, RMSE: 11.10) to 12-weeks ahead (MAE: 13.38; RMSE: 26.82). There was considerable variation in the relative importance of predictors dependent on forecast horizon. The environmental and meteorological predictors were relatively important for short-term dengue forecast horizons while socio-demographic predictors were relevant for longer-term forecast horizons. This study showed the potential of RF in dengue forecasting with also demonstrating the feasibility of using a national model to forecast at finer spatial scales. Furthermore, sociodemographic predictors are important to include to capture longer-term trends in dengue.Author summaryDengue virus has the highest disease burden of all mosquito-borne viral diseases, infecting 390 million people annually in 128 countries. Forecasting is an important warning mechanism that can help with proactive planning and response for clinical and public health services. In this study, we compare two different machine learning approaches to dengue forecasting: random forest (RF) and neural networks (NN). National and local (departmental-level) models were compared and used to predict dengue cases in the future. The results showed that the counts of future dengue cases were more accurately estimated by RF than by NN. It was also shown that environmental and meteorological predictors were more important for forecast accuracy for shorter-term forecasts while socio-demographic predictors were more important for longer-term forecasts. Finally, the national model applied to local data was more accurate in dengue forecasting compared to the local model. This research contributes to the field of disease forecasting and highlights different considerations for future forecasting studies.

Supplemental Digital Content is Available in the Text., Context: Foodborne illness affects 1 in 4 US residents each year. Few of those sickened seek medical care or report the illness to public health authorities, complicating prevention efforts. Citizens who report illness identify food establishments with more serious and critical violations than found by regular inspections. New media sources, including online restaurant reviews and social media postings, have the potential to improve reporting. Objective: We implemented a Web-based Dashboard (HealthMap Foodborne Dashboard) to identify and respond to tweets about food poisoning from St Louis City residents. Design and Setting: This report examines the performance of the Dashboard in its first 7 months after implementation in the City of St Louis Department of Health. Main Outcome Measures: We examined the number of relevant tweets captured and replied to, the number of foodborne illness reports received as a result of the new process, and the results of restaurant inspections following each report. Results: In its first 7 months (October 2015-May 2016), the Dashboard captured 193 relevant tweets. Our replies to relevant tweets resulted in more filed reports than several previously existing foodborne illness reporting mechanisms in St Louis during the same time frame. The proportion of restaurants with food safety violations was not statistically different (P = .60) in restaurants inspected after reports from the Dashboard compared with those inspected following reports through other mechanisms. Conclusion: The Dashboard differs from other citizen engagement mechanisms in its use of current data, allowing direct interaction with constituents on issues when relevant to the constituent to provide time-sensitive education and mobilizing information. In doing so, the Dashboard technology has potential for improving foodborne illness reporting and can be implemented in other areas to improve response to public health issues such as suicidality, spread of Zika virus infection, and hospital quality.

Although digital reports of disease are currently used by public health officials for disease surveillance and decision making, little is known about environmental factors and compositional characteristics that may influence reporting patterns. The objective of this study is to quantify the association between climate, demographic and socio-economic factors on digital reporting of disease at the US county level. We reference approximately 1.5 million foodservice business reviews between 2004 and 2014, and use census data, machine learning methods and regression models to assess whether digital reporting of disease is associated with climate, socio-economic and demographic factors. The results show that reviews of foodservice businesses and digital reports of foodborne illness follow a clear seasonal pattern with higher reporting observed in the summer, when most foodborne outbreaks are reported and to a lesser extent in the winter months. Additionally, factors typically associated with affluence (such as, higher median income and fraction of the population with a bachelor's degrees) were positively correlated with foodborne illness reports. However, restaurants per capita and education were the most significant predictors of illness reporting at the US county level. These results suggest that well-known health disparities might also be reflected in the online environment. Although this is an observational study, it is an important step in understanding disparities in the online public health environment.

Background Rapidly increasing volumes of news feeds from diverse data sources, such as online newspapers, Twitter, and online blogs, are proving to be extremely valuable resources in helping to anticipate, detect, and forecast outbreaks of rare diseases. The goal of this paper is to develop techniques that can effectively forecast the emergence and progression of rare infectious diseases by combining data from disparate data sources. Methods We introduce SourceSeer, a novel algorithmic framework that combines spatiotemporal topic models with source-based anomaly detection techniques. SourceSeer is capable of discovering the location focus of each source, allowing sources to be used as experts with varying degrees of authoritativeness. To fuse the individual source predictions into a final outbreak prediction, we employ a multiplicative weights algorithm taking into account the accuracy of each source. Results We evaluate the performance of SourceSeer using incidence data for hantavirus syndromes in multiple countries of Latin America provided by HealthMap over a time span of 15 months. We demonstrate that SourceSeer makes predictions of increased accuracy compared to several baselines and can forecast disease outbreaks in a timely manner even when no outbreaks were previously reported.

Obesity is a global epidemic affecting millions. Implementation of interventions to curb obesity rates requires timely surveillance. In this study, we estimated sex-specific obesity prevalence using social media, search queries, demographics and built environment variables. We collected 3,817,125 and 1,382,284 geolocated tweets on food and exercise respectively, from Twitter’s streaming API from April 2015 to March 2016. We also obtained searches related to physical activity and diet from Google Search Trends for the same time period. Next, we inferred the gender of Twitter users using machine learning methods and applied mixed-effects state-level linear regression models to estimate obesity prevalence. We observed differences in discussions of physical activity and foods, with males reporting higher intensity physical activities and lower caloric foods across 40 and 48 states, respectively. In addition, counties with the highest percentage of exercise and food tweets had lower male and female obesity prevalence. Lastly, our models separately captured overall male and female spatial trends in obesity prevalence. The average correlation between actual and estimated obesity prevalence was 0.797(95% CI, 0.796, 0.798) and 0.830 (95% CI, 0.830, 0.831) for males and females, respectively. Social media can provide timely community-level data on health information seeking and changes in behaviors, sentiments and norms. Social media data can also be combined with other data types such as, demographics, built environment variables, diet and physical activity indicators from other digital sources (e.g., mobile applications and wearables) to monitor health behaviors at different geographic scales, and to supplement delayed estimates from traditional surveillance systems.

ObjectivesWe examined the use of data from social media for surveillance of physical activity prevalence in the USA.MethodsWe obtained data from the social media site Twitter from April 2015 to March 2016. The data consisted of 1 382 284 geotagged physical activity tweets from 481 146 users (55.7% men and 44.3% women) in more than 2900 counties. We applied machine learning and statistical modelling to demonstrate sex and regional variations in preferred exercises, and assessed the association between reports of physical activity on Twitter and population-level inactivity prevalence from the US Centers for Disease Control and Prevention.ResultsThe association between physical inactivity tweet patterns and physical activity prevalence varied by sex and region. Walking was the most popular physical activity for both men and women across all regions (15.94% (95% CI 15.85% to 16.02%) and 18.74% (95% CI 18.64% to 18.88%) of tweets, respectively). Men and women mentioned performing gym-based activities at approximately the same rates (4.68% (95% CI 4.63% to 4.72%) and 4.13% (95% CI 4.08% to 4.18%) of tweets, respectively). CrossFit was most popular among men (14.91% (95% CI 14.52% to 15.31%)) among gym-based tweets, whereas yoga was most popular among women (26.66% (95% CI 26.03% to 27.19%)). Men mentioned engaging in higher intensity activities than women. Overall, counties with higher physical activity tweets also had lower leisure-time physical inactivity prevalence for both sexes.ConclusionsThe regional-specific and sex-specific activity patterns captured on Twitter may allow public health officials to identify changes in health behaviours at small geographical scales and to design interventions best suited for specific populations.

Text, images, geotags and other data from social media sites lend researchers a unique window into population health trends and disease spread. While these data provide the opportunity to track and measure health outcomes across geographic regions, over extended periods of time, and through complex social networks, they also present challenges. Most notably, these data carry significant biases due to demographic differences in who chooses to use each platform, and what they choose to share. While several publications have discussed the limitations of leveraging social media data for public health research, the amount of literature systematically investigating their demographic bias and exploring mitigation strategies is limited and ripe for interdisciplinary contributions. In this discussion paper, we highlight that understanding the strengths and limitations of these data sources would enable a rigorous assessment of their usefulness for public health research and provide a means for quantifying uncertainty in research findings.