Identifying County-Level All-Cause Mortality Rate Trajectories and Their Spatial Distribution Across the United States

What is the relationship between mortality and satellite images as elucidated through the use of Convolutional Neural Networks?Background: Following a century of increase, life expectancy in the United States has stagnated and begun to decline in recent decades. Using satellite images and street view images, prior work has demonstrated associations of the built environment with income, education, access to care, and health factors such as obesity. However, assessment of learned image feature relationships with variation in crude mortality rate across the United States has been lacking.Objective: We sought to investigate if county-level mortality rates in the U.S. could be predicted from satellite images.Methods: Satellite images of neighborhoods surrounding schools were extracted with the Google Static Maps application programming interface for 430 counties representing ~68.9% of the US population. A convolutional neural network was trained using crude mortality rates for each county in 2015 to predict mortality. Learned image features were interpreted using Shapley Additive Feature Explanations, clustered, and compared to mortality and its associated covariate predictors.Results: Predicted mortality from satellite images in a held-out test set of counties was strongly correlated to the true crude mortality rate (Pearson r = 0.72). Direct prediction of mortality using a deep learning model across a cross-section of 430 U.S. counties identified key features in the environment (e.g., sidewalks, driveways, and hiking trails) associated with lower mortality. Learned image features were clustered, and we identified 10 clusters that were associated with education, income, geographical region, race, and age.Conclusions: The application of deep learning techniques to remotely-sensed features of the built environment can serve as a useful predictor of mortality in the United States. Although we identified features that were largely associated with demographic information, future modeling approaches that directly identify image features associated with health-related outcomes have the potential to inform targeted public health interventions.

Download Full-text

Using Satellite Images and Deep Learning to Identify Associations Between County-Level Mortality and Neighborhood Features

10.1101/2020.10.12.20211755 ◽

2020 ◽

Author(s):

Joshua J. Levy ◽

Rebecca M. Lebeaux ◽

Anne G. Hoen ◽

Brock C. Christensen ◽

Louis J. Vaickus ◽

...

Keyword(s):

United States ◽

Deep Learning ◽

Mortality Rate ◽

Satellite Images ◽

The United States ◽

Mortality Rates ◽

Image Features ◽

County Level ◽

Crude Mortality Rate ◽

Crude Mortality

AbstractWhat is the relationship between mortality and satellite images as elucidated through the use of Convolutional Neural Networks?BackgroundFollowing a century of increase, life expectancy in the United States has stagnated and begun to decline in recent decades. Using satellite images and street view images, prior work has demonstrated associations of the built environment with income, education, access to care and health factors such as obesity. However, assessment of learned image feature relationships with variation in crude mortality rate across the United States has been lacking. We sought to investigate if county-level mortality rates in the U.S. could be predicted from satellite images.MethodsSatellite images were extracted with the Google Static Maps application programming interface for 430 counties representing approximately 68.9% of the US population. A convolutional neural network was trained using crude mortality rates for each county in 2015 to predict mortality. Learned image features were interpreted using Shapley Additive Feature Explanations, clustered, and compared to mortality and its associated covariate predictors.ResultsPredicted mortality from satellite images in a held-out test set of counties was strongly correlated to the true crude mortality rate (Pearson r=0.72). Learned image features were clustered, and we identified 10 clusters that were associated with education, income, geographical region, race and age. Direct prediction of mortality using a deep learning model across a cross-section of 430 U.S. counties identified key features in the environment (e.g. sidewalks, driveways and hiking trails) associated with lower mortality.ConclusionsThe application of deep learning techniques to remotely-sensed features of the built environment can serve as a useful predictor of mortality in the United States. Although we identified features that were largely associated with demographic information, future modeling approaches that directly identify image features associated with health-related outcomes have the potential to inform targeted public health interventions.

Download Full-text

Exploring the Power of County-Level Regional Classifications on Predicting Sex-Specific All-Cause Mortality in the United States

10.33915/etd.6968 ◽

2017 ◽

Author(s):

Julia Kay Wolf

Keyword(s):

United States ◽

The United States ◽

County Level ◽

All Cause Mortality

Download Full-text

All-Cause and Inhibitor-Related Mortality in Non-Severe Hemophilia Α Patients in the United States

Blood ◽

10.1182/blood-2019-122991 ◽

2019 ◽

Vol 134 (Supplement_1) ◽

pp. 902-902 ◽

Cited By ~ 1

Author(s):

Ming Y. Lim ◽

Dunlei Cheng ◽

Christine L. Kempton ◽

Nigel S. Key

Keyword(s):

United States ◽

At Risk ◽

Mortality Rate ◽

Hemophilia A ◽

The United States ◽

Severe Hemophilia ◽

All Cause Mortality ◽

Related Mortality ◽

Observation Period

Introduction: The majority of published studies evaluating inhibitors have focused mainly on patients with severe hemophilia A. In non-severe hemophilia A (NSHA) patients, the development of inhibitors can have a profound clinical impact, with major bleeding complications similar to that of patients with severe or acquired hemophilia. Yet, epidemiological data on inhibitors in NSHA patients, specifically mortality, is scarce and currently limited to the European and Australian cohort [Eckhardt CL, et al. J Thromb Haemost. 2015 Jul;13(7):1217-251]. Objectives: To determine the all-cause and inhibitor-related mortality in NSHA patients in the United States using the ATHNdataset Methods: Subjects and study design The ATHNdataset is a 'limited dataset' as defined under the United States Health Insurance Portability and Accountability Act (HIPAA) to be free of protected health information, with data collection by more than 130 hemophilia treatment centers (HTC) across the United States. It includes patients with congenital bleeding disorders in the United States who have authorized the sharing of their demographic and clinical information for research. Data collection and definitions The ATHNdataset was queried on December 31, 2018 to extract the following information on NSHA patients: Patient demographics, inhibitor status, date of death, and primary cause of death. The presence of inhibitors was defined as: (i) ≥ 2 positive Bethesda inhibitor assay titers of ≥ 1.0 BU/mL; or (ii) a decrease in plasma FVIII coagulant activity (FVIII:C) to at least 50% of baseline activity and/or spontaneous bleeding symptoms in patients with inhibitor titers between 0.6 and 1.0 BU/mL. Patients who had a negative inhibitor history or have never been tested for FVIII inhibitors were classified as negative for inhibitors. Statistical analyses The person-year mortality rate was calculated as the ratio of the number of deaths to the number of person-years at risk, presented as rates per 1000 person-years. Person-years at risk was calculated for each patient as the time between the start of the observation period (January 1, 2010 or date of birth for patients who are born later) and the end of the observation period (date of death, loss-to follow-up or December 31, 2018). Patients who were deceased or lost to follow-up before January 1, 2010 were not included in the analysis. Inhibitor person-years at risk for inhibitor patients was calculated from January 1, 2010 if the first positive inhibitor test occurred prior to January 1, 2010 or from the date of the first positive inhibitor test that occurred during the observation period until the end of the observation period. Inhibitor-related death was attributed to all patients who had a positive inhibitor history. Mortality rates were compared between inhibitor and non-inhibitor patients using z- test. Results: Between 1/1/2010 and 12/31/2018, the ATHNdataset included 6,606 NSHA patients who were born between 1920 and 2018. Patients were observed for a total of 56,064 person-years. 85.57% (n = 5,653) of these patients were observed for the full nine years. The average follow-up time per patient was almost 8.5 years. Inhibitors developed in 171 (2.59%) NSHA patients. The median age for inhibitor development was 13 years (IQR, 6 - 37 years) and the mean age was 22 years. Demographics characteristics of the patients are listed in Table 1. All-cause mortality At the end of follow-up, there was a total of 136 deaths in the NSHA population, occurring at a median age of 63 years (IQR, 51 - 75 years). The overall all-cause mortality rate was 2.43 per 1,000 person-years (95% CI: 2.02 - 2.83). The most common primary cause of death was cancer (n=27, 19.9%) (Table 2). Inhibitor-related mortality Three deaths were associated with inhibitors. Inhibitor-related mortality rate was 2.40 per 1,000 person-years, whereas among the never inhibitor group, the mortality rate was 2.44 per 1,000 person-years (p = 0.790). Mortality risk ratio between inhibitor and never inhibitor was 0.98 (95% CI: 0.31 - 3.08). Conclusion: In NSHA patients, the development of inhibitors occurred at a relatively early age and was not associated with increased mortality. Disclosures Kempton: Novo Nordisk: Research Funding; Octapharma: Honoraria; Genentech: Honoraria; Spark Therapeutics: Honoraria. Key:Uniqure BV: Research Funding.

Download Full-text

COVID-19 and excess mortality in the United States: A county-level analysis

PLoS Medicine ◽

10.1371/journal.pmed.1003571 ◽

2021 ◽

Vol 18 (5) ◽

pp. e1003571

Author(s):

Andrew C. Stokes ◽

Dielle J. Lundberg ◽

Irma T. Elo ◽

Katherine Hempstead ◽

Jacob Bor ◽

...

Keyword(s):

United States ◽

Formal Education ◽

Excess Mortality ◽

Time Lag ◽

The United States ◽

County Level ◽

Level Variation ◽

Cross Sectional ◽

All Cause Mortality ◽

Excess Deaths

Background Coronavirus Disease 2019 (COVID-19) excess deaths refer to increases in mortality over what would normally have been expected in the absence of the COVID-19 pandemic. Several prior studies have calculated excess deaths in the United States but were limited to the national or state level, precluding an examination of area-level variation in excess mortality and excess deaths not assigned to COVID-19. In this study, we take advantage of county-level variation in COVID-19 mortality to estimate excess deaths associated with the pandemic and examine how the extent of excess mortality not assigned to COVID-19 varies across subsets of counties defined by sociodemographic and health characteristics. Methods and findings In this ecological, cross-sectional study, we made use of provisional National Center for Health Statistics (NCHS) data on direct COVID-19 and all-cause mortality occurring in US counties from January 1 to December 31, 2020 and reported before March 12, 2021. We used data with a 10-week time lag between the final day that deaths occurred and the last day that deaths could be reported to improve the completeness of data. Our sample included 2,096 counties with 20 or more COVID-19 deaths. The total number of residents living in these counties was 319.1 million. On average, the counties were 18.7% Hispanic, 12.7% non-Hispanic Black, and 59.6% non-Hispanic White. A total of 15.9% of the population was older than 65 years. We first modeled the relationship between 2020 all-cause mortality and COVID-19 mortality across all counties and then produced fully stratified models to explore differences in this relationship among strata of sociodemographic and health factors. Overall, we found that for every 100 deaths assigned to COVID-19, 120 all-cause deaths occurred (95% CI, 116 to 124), implying that 17% (95% CI, 14% to 19%) of excess deaths were ascribed to causes of death other than COVID-19 itself. Our stratified models revealed that the percentage of excess deaths not assigned to COVID-19 was substantially higher among counties with lower median household incomes and less formal education, counties with poorer health and more diabetes, and counties in the South and West. Counties with more non-Hispanic Black residents, who were already at high risk of COVID-19 death based on direct counts, also reported higher percentages of excess deaths not assigned to COVID-19. Study limitations include the use of provisional data that may be incomplete and the lack of disaggregated data on county-level mortality by age, sex, race/ethnicity, and sociodemographic and health characteristics. Conclusions In this study, we found that direct COVID-19 death counts in the US in 2020 substantially underestimated total excess mortality attributable to COVID-19. Racial and socioeconomic inequities in COVID-19 mortality also increased when excess deaths not assigned to COVID-19 were considered. Our results highlight the importance of considering health equity in the policy response to the pandemic.

Download Full-text

County-Level Basics of Living Affects Overall Mortality Rate in the United States – 2000

Applied Research in Quality of Life ◽

10.1007/s11482-006-9008-9 ◽

2006 ◽

Vol 1 (1) ◽

pp. 97-106

Author(s):

Sarah J. Shao ◽

Shu-Chih Su ◽

Norma Kanarek

Keyword(s):

United States ◽

Mortality Rate ◽

The United States ◽

County Level ◽

Overall Mortality

Download Full-text

Evaluation of County-Level Heterogeneity in Excess Mortality in Colorado from March to September 2020

10.1101/2021.04.10.21255235 ◽

2021 ◽

Author(s):

Jay Chandra ◽

Marie Charpignon ◽

Mathew C Samuel ◽

Anushka Bhaskar ◽

Saketh Sundar ◽

...

Keyword(s):

Public Health ◽

Mortality Rate ◽

Excess Mortality ◽

Ethnic Disparities ◽

The United States ◽

The State ◽

Mortality Data ◽

County Level ◽

All Cause Mortality ◽

Excess Deaths

Importance: Tracking the direct and indirect impact of the coronavirus disease 2019 (COVID-19) pandemic on all-cause mortality in the United States has been hindered by the lack of testing and by reporting delays. Evaluating excess mortality, or the number of deaths above what is expected in a given time period, provides critical insights into the true burden of the COVID-19 pandemic caused by the novel Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2). Stratifying mortality data by demographics such as age, sex, race, ethnicity, and geography helps quantify how subgroups of the population have been differentially affected. Similarly, stratifying mortality data by cause of death reveals the public health effects of the pandemic in terms of other acute and chronic diseases. Objective: To provide stratified estimates of excess mortality in Colorado from March to September 2020. Design, Setting, and Population: This study evaluated the number of excess deaths both directly due to SARS-CoV-2 infection and from all other causes between March and September 2020 at the county level in Colorado. Data were obtained from the Vital Statistics Program at the Colorado Department of Public Health and Environment. These estimates of excess mortality were derived by comparing population- adjusted mortality rates in 2020 with rates in the same months from 2015 to 2019. Results: We found evidence of excess mortality in Colorado between March and September 2020. Two peaks in excess deaths from all causes were recorded in the state, one mid-April and the other at the end of June. Since the first documented SARS-CoV-2 infection on March 5th, we estimated that the excess mortality rate in Colorado was two times higher than the officially reported COVID-19 mortality rate. State-level cumulative excess mortality from all causes reached 71 excess deaths per 100k residents (~4000 excess deaths in the state); in contrast, 35 deaths per 100k directly due to SARS-CoV-2 were recorded in the same period (~1980 deaths. Excess mortality occurred in 52 of 64 counties, accounting for 99% of the state's population. Most excess deaths recorded from March to September 2020 were associated with acute events (estimated at 44 excess deaths per 100k residents and at 9 after excluding deaths directly due to SARS-CoV-2) rather than with chronic conditions (~21 excess deaths per 100k). Among Coloradans aged 14-44, 1.4 times more deaths occurred in those months than during the same period in the five previous years. Hispanic White males died of COVID-19 at the highest rate during this time (~90 deaths from COVID-19 per 100k residents); however, Non-Hispanic Black/African American males were the most affected in terms of overall excess mortality (~204 excess deaths per 100k). Beyond inequalities in COVID-19 mortality per se, these findings signal considerable regional and racial-ethnic disparities in excess all-cause mortality that need to be addressed for a just recovery and in future public health crises.

Download Full-text