Ian Buller, PhD, MA

American Geophysical Union 2025 Annual Meeting

Thu, 18 Dec 2025 00:00:00 +0000

I presented a poster at the American Geophysical Union 2025 Annual Meeting entitled “Temporal clustering of air pollution and heat wave events at domestic U.S. Military facilities.”

I presented initial results comparing three statistical approaches (supervised and unsupervised machine learning approaches) to evaluate the potential relationship between heat and air pollution wave events at U.S. Military Facilities within 4 km of U.S. Environmental Protection Agency Air Quality System sampling locations. We also used publicly-available PRISM Climate Group data from Oregon State University with our custom-built open-source software.

Ian Buller, PhD, MA

Thu, 18 Dec 2025 00:00:00 +0000

Disclaimer: All content is my own and does not represent my employer

I am an Epidemiologist within Public Health & Scientific Research at DLH, LLC (formerly known as Social and Scientific Systems, Inc.), focusing on the (geo)spatial and environmental epidemiology of various health outcomes, including cancer and infectious disease.

I was a Postdoctoral Cancer Prevention Fellow in the Cancer Prevention Fellowship Program at the National Cancer Institute (Preceptor: Rena Jones, PhD, MS) working within the Occupational and Enviornmental Epidemiology Branch of the Division of Cancer Epidemiology and Genetics, part of the Intramural Research Program at the National Institutes of Health. I received a PhD in Environmental Health Sciences at Emory University (Advisor: Lance Waller, PhD) after completing a concurrent BA/MA in Ecology and Evolutionary Biology from the University of Colorado at Boulder (Advisor: Pieter Johnson, PhD).

My research is published in (the):

NIH Research Festival 2025

Fri, 12 Sep 2025 00:00:00 +0000

I co-instructed a workshop at the NIH Research Festival 2025 entitled “Geospatial for Everyone: Enhancing Your ‘Non-spatial’ Research with Geospatial Data” with Dr. Nathaniel MacNell.

In our workshop, we presented an interactive open-source workflow of geographically linking publicly available data to simulated locations and demonstrated how these data can be used for clinical study design and analysis. We used the National Heart, Lung, and Blood Institute (NHLBI) & National Cancer Institute (NCI) Neighborhood Deprivation Index (NDI) throughout our examples, which is also calculated by my R package ndi (click here for more information). We were one of twenty-six vendor workshops held at the NIH Research Festival chosen out of hundreds of applicants.

Military Health System Research Symposium 2025

Wed, 06 Aug 2025 00:00:00 +0000

I presented a poster at the Military Health System Research Symposium 2025 entitled “Open-source estimation of shade availability at military bases in the United States.”

I presented an initial workflow of estimating the shadow footprint (shade) of obstacles on United States (US) Military facilities using the US Naval Academy campus as an example. We use publicly-available US Geological Survey 3D Elevation Program point cloud data with custom-built open-source software.

New Publication in Frontiers in Public Health

Sun, 13 Jul 2025 00:00:00 +0000

I co-authored an article in Frontiers in Public Health entitled “Descriptive analysis of municipal policies addressing shade in eight southwest and northeast states in the United States” led by Dr. David Buller.

We assessed the municipal codes, planning documents, and manuals/guidelines from n=48 municipalities in eight states across the northeastern and southwestern U.S. We found that many municipalities had policies that mentioned shade, but only a few had policies that indicated their purpose was sun or heat protection.

New Publication in Environmental Research

Wed, 28 May 2025 00:00:00 +0000

I co-authored an article in Environmental Research entitled “Ambient air pollution in critical windows of exposure and spontaneous miscarriage in a preconception cohort” led by Dr. Anne Marie Jukic.

We assessed the potential relationship between ambient air pollution and risk of spontaneous pregnancy loss for participants within the Time to Conceive study. We linked the U.S. Environmental Protection Agency (EPA) Community Multi-Scale Air Quality model (CMAQ) and Fused Air Quality Surface Using Downscaling model for ambient air pollutant concentrations during multiple exposure windows during conception cycles to participant residential addresses at study enrollment. We found that residing in areas with higher ozone concentrations were associated with a small increase in spontaneous pregnancy loss risk. Ozone, carbon monoxide, and nitric oxide had stronger associations with spontaneous pregnancy loss risk among those with low vitamin D. Mixtures of ambient air pollutants were weakly associated with spontaneous pregnancy loss risk.

New Publication in the Science of The Total Environment

Thu, 15 May 2025 00:00:00 +0000

I co-authored an article in the Science of The Total Environment entitled “Prevalence of cardiovascular disease risk factors associated with residential natural hazard risk” led by Dr. Kaitlyn Lawrence.

We assessed the potential relationship between CVD-related risk factors (i.e., diabetes, obesity, hypertension) and risk of natural hazards for participants within the Gulf Long-Term Follow-up Study. We linked the Federal Emergency Management Agency (FEMA) National Risk Index (NRI) for Natural Hazards to participant residential addresses at home visit. We found that residing in areas prone to all or specific natural disasters is associated with higher prevalence of cardiovascular disease risk factors.

New Publication in Cancer Research

Wed, 30 Apr 2025 00:00:00 +0000

I co-authored an article in Cancer Research entitled “Integration of Germline and Somatic Variation Improves Chronic Lymphocytic Leukemia Risk Stratification” led by Dr. Aubrey Hubbard.

We investigated whether inclusion of a chronic lymphocytic leukemia (CLL)-associated polygenic score and two common types of clonal hematopoiesis (i.e., autosomal mosaic chromosomal alterations and clonal hematopoiesis indeterminate potential), could improve CLL risk stratification in the UK Biobank. We also replicated our investigation in the Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial. We obtained enhanced ability to identify individuals at high risk of CLL when integrating germline and somatic data derived from peripheral blood.

New Publication in the International Journal of Health Geographics

Fri, 11 Apr 2025 00:00:00 +0000

I co-authored an article in the International Journal of Health Geographics entitled “Geographic patterns in wildland fire exposures and county-level lung cancer mortality in the United States” led by Dr. Richard Remigio.

We applied geospatial statistical methods to describe patterns in lung cancer mortality rates (2016-2020) in relation to patterns in various wildland fire metrics (1997-2003) by sex at the US county level after accounting cigarette smoking prevalences (1997-2003). Our findings identified counties outside the western US with wildfires associated with lung cancer mortality and where further study is needed. This analysis used similar approaches to a previous co-authorship in Cancer Epidemiology, Biomarkers & Prevention Volume 32 Issue 2.

New Publication in the Journal of the National Cancer Institute

Thu, 20 Mar 2025 00:00:00 +0000

I co-authored an article in the Journal of the National Cancer Institute entitled “Associations of self-identified race and ethnicity and genetic ancestry with mortality among cancer survivors” led by Dr. Jacqueline Vo and Dr. Derek Brown.

We studied self-identified race and ethnicity (SIRE) and genetic ancestry in the Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial. We found that SIRE and genetic ancestry do not solely reflect biologic variation; rather, social factors may drive mortality differences by SIRE and genetic ancestry. This project was funded, in-part, by our 2021 William G. Coleman Minority Health and Health Disparities Research Innovation Award.

New Commentary in Epidemiology

Wed, 22 Jan 2025 00:00:00 +0000

I co-authored a commentary in the Epidemiology entitled “Advancing reproducible research through version control technology” led by Dr. Ghassan Hamra.

We introduce version control and explain why we believe all epidemiologists should use it within their workflows to support reproducible research. This is my first published non-invited commentary.

American Geophysical Union 2024

Mon, 09 Dec 2024 00:00:00 +0000

I presented a lightning talk at the American Geophysical Union Annual Meeting 2024 entitled “Ambient PM2.5 exposure and risk of non-Hodgkin lymphoma in a large U.S.-based cohort.”

I presented initial results from a project conducted by Inam Ghulamhussain who I mentored while I was a postdoctoral Cancer Prevention Fellow and he was a National Institutes of Health Summer Intern at the National Cancer Institute.

New Publication in Spatial and Spatio-temporal Epidemiology

Sat, 02 Nov 2024 00:00:00 +0000

I first authored an article in Spatial and Spatio-temporal Epidemiology entitled “Multiple ‘spaces’: using wildlife surveillance, climatic variables, and spatial satitics to identify and map a climatic niche for endemic plague in California.”

We estimated plague-suitable local climates and mapped them in California using coyotes (Canis latrans) tested for plague exposure by the California Department of Public Health - Vector-Bourne Disease Section. We applied spatial point processes within a space composed to two principal components of a principal component analysis of Oregon State University PRISM Climate Group 30-year average climate variables. The analysis was conducted using the envi package in R that I previously published. This is the first publication from my doctoral dissertation research.

New Publication in BMC Cancer

Tue, 06 Aug 2024 00:00:00 +0000

I co-first authored an article in BMC Cancer entitled “Neighborhood-Level Deprivation and Survival in Lung Cancer” co-led with Dr. Kathleen Kennedy and Dr. Ignacio Jusue-Torres.

We assessed the potential associations of neighborhood deprivation, DNA methylation, and lung cancer mortality in a multicenter retrospective cohort study. We found HOXA7 and DNA methylation were significantly correlated with neighborhood deprivation and participants residing in high-deprivation neighborhoods had a significantly shorter survival than those residing in low-deprivation neighborhoods. My largest contributions to the study was creating the code companion for the manuscript and designing, computing, and performing of the neighborhood deprivation index linkages to participant locations using the ndi package in R that I previously published.

New Publication in the American Journal Of Epidemiology

Fri, 19 Jul 2024 00:00:00 +0000

I co-authored an article in the American Journal of Epidemiology entitled “Residential natural hazard risk and mental health effects” led by Dr. Kaitlyn Lawrence.

We assessed the potential relationship between multiple mental health outcomes (mood disorders: generalized anxiety disorder [GAD], major depressive disorder [MDD], and post-traumatic stress disorder [PTSD]) and risk of natural hazards for participants within the Gulf Long-Term Follow-up Study. We linked the Federal Emergency Management Agency (FEMA) National Risk Index (NRI) for Natural Hazards to participant residential addresses at home visit. Using adjusted log-binomial regression we observed the highest levels of risk for hurricane and flooding were associated with higher levels of PTSD. We also observed higher levels of heatwave risk was associated with higher levels of anxiety and depression. My largest contribution to the study was designing and performing the majority of the data linkages.

New Publication in Science of The Total Environment

Tue, 02 Jul 2024 00:00:00 +0000

I co-authored an article in Science of the Total Environment entitled “Residential air pollution, greenspace, and adverse mental health outcomes in the U.S. Gulf Long-term Follow-up Study” led by Dr. Emily Werder.

We assessed the potential relationship between multiple mental health outcomes (mood disorders: depression and distress) and exposure to ambient air pollution (annual average concentrations of particulate matter less than 2.5 microns in diameter [PM_2.5] and nitrogen dioxide [NO₂]) and greenness (Enhanced Vegetation Index [EVI]) for participants within the Gulf Long-Term Follow-up Study. We linked National Aeronautics and Space Administration (NASA) Sociodemographic Data and Applications Center (SEDAC) and Moderate Resolution Imaging Spectroradiometer (MODIS) data to participant residential addresses at home visit. Using adjusted log-binomial regression we observed the highest level of PM_2.5 exposure was associated with higher levels of depression. We also observed the highest level of greenness was associated with lower levels of depression and appears to mitigate the impact of PM_2.5 on depression. We also linked (and adjusted for) U.S. Department of Agriculture (USDA) Rural-urban commuting area codes (RUCA) and Area Deprivation Index (ADI) from the University of Wisconsin. My largest contribution to the study was designing and performing the majority of the data linkages.

Society for Epidemiologic Research 2024

Tue, 18 Jun 2024 00:00:00 +0000

I co-instructed a workshop at the Society for Epidemiologic Research Annual Meeting 2024 entitled “Git’n up to speed on versioning control.”

The workshop was an introduction to the Git language and other Git-enabled technology for epidemiologists to version control their software code. I developed the course with other DLHers, including Dr. Ghassan Hamra, Dr. Nat MacNell, and Audrey Brown.

I was also a co-author on a poster abstract entitled “Revisiting the modifiable areal unit problem in the era of exposome-wide association studies: Assessing the performance of the CDC/ATSDR Social Vulnerability Index at privacy-protecting spatial scales” led by Jonathan Lewis.

New Publication in the Journal of Exposure Science & Environmental Epidemiology

Wed, 12 Jun 2024 00:00:00 +0000

I co-authored an article in the Journal of Exposure Science & Environmental Epidemiology entitled “Association between solar radiation and mood disorders among Gulf Coast residents” led by Dr. Xinlei Deng.

We assessed the potential relationship between multiple mental health outcomes (mood disorders: depression and distress) and multiple windows of exposure to solar radiation (7-, 14-, and 30-days prior to home visit) for participants within the Gulf Long-Term Follow-up Study. We linked Oak Ridge National Laboratory (ORNL) Daymet data to participant residential addresses at home visit. Using adjusted generalized linear mixed models we observed higher levels of 7-days prior solar radiation exposure were associated with lower levels of mood disorders. This is my first peer-reviewed publication as an Epidemiologist at DLH, LLC.

New Publication in the Journal of Agricultural, Biological and Environmental Statistics

Mon, 03 Apr 2023 00:00:00 +0000

I co-authored an article in the Journal of Agricultural, Biological and Environmental Statistics entitled “A Shared Latent Process Model to Correct for Preferential Sampling in Disease Surveillance Systems” led by Dr. Brian Conroy.

We designed an approach to address preferential sampling in infectious disease surveillance that incorporates both locational and disease-related components. The locational component describes the spatial pattern of sampled locations in terms of a latent spatial process, which is shared with the disease-related component of the model that describes the abundances of disease positive and negative specimens. We used a simulation-based study and a real-world case study of wildlife rodent specimen in California, U.S.A., to compare our proposed method against two conventional infectious disease surveillance methods. Our proposed method identified the focal high-risk areas and large low-risk areas (e.g., Central Valley) of plague better than the two benchmark methods.

Accessible geospatial socio-economic metrics

Tue, 24 Jan 2023 00:00:00 +0000

I gave an invited talk at the National Cancer Institute Division of Cancer Prevention Multi-Cancer Detection Working Group entitled “Accessible geospatial socio-economic metrics.”

The talk was an introduction to my open-source R package in the Comprehensive R Archive Network (CRAN) named ndi that I developed as a postdoctoral Cancer Prevention Fellow. The host, Dr. Amanda Skarlupka, performed a tutorial of the package and how it can be used for the evaluation and design of clinical trial recruitment. See more details about the ndi package.

New Publication in the Journal of the National Cancer Institute

Thu, 12 Jan 2023 00:00:00 +0000

I co-authored an article in the Journal of the National Cancer Institute entitled “Ethylene oxide emissions and incident breast cancer and non-Hodgkin lymphoma in a U.S. cohort” led by Dr. Rena Jones.

We constructed two GIS-based exposure metrics of ethylene oxide (EtO) from the U.S. Environmental Protection Agency Toxics Release Inventory to estimate the risk of incident breast cancer and non-Hodgkin lymphoma (NHL) in the NIH-AARP Diet and Health Study. We observed an increased risk of breast cancer associated with EtO-emitting facilities within 10 kilometers of residence (stronger for in situ than invasive disease) and higher risk of breast cancer in situ in top exposure quartiles. No differences in breast cancer risk were observed by tumor estrogen receptor status and we found no clear pattern of increased NHL risk.

New Publication in the Proceedings of the National Academy of Sciences

Tue, 03 Jan 2023 00:00:00 +0000

I co-authored an article in the Proceedings of the National Academy of Sciences entitled “Burkitt lymphoma risk shows geographic and temporal associations with Plasmodium falciparum infections in Uganda, Tanzania, and Kenya” led by Kelly Broen and Dr. Sam Mbulaiteye.

We used high-resolution long-term serial malaria exposure data (2000-2017; 5km resolution) and granular pediatric endemic Burkitt lymphoma (eBL) case data from the EMBLEM study (2010-2016) across regions of Kenya, Tanzania, and Uganda to find geographic, temporal, and age-related associations between lifetime exposure to Plasmodium falciparum and eBL risk. Our findings suggest that malaria reduction may substantially reduce eBL incidence and mortality. You can find the R langauge scripts for the manuscript analysis in this GitHub repository.

New Affiliation

Mon, 05 Dec 2022 00:00:00 +0000

I completed my postdoctoral fellowship at the National Cancer Institute through the Cancer Prevention Fellowship Program working with Dr. Rena Jones and Dr. Mary Ward in the Occupational and Environmental Epidemiology Branch in the Division of Cancer Epidemiology and Genetics in Rockville, Maryland.

I am now an Epidemiologist at DLH, LLC (formerly Social and Scientific Systems, Inc.), in Bethesda, Maryland.

New Publication in Cancer Epidemiology, Biomarkers & Prevention

Tue, 22 Nov 2022 00:00:00 +0000

I co-authored an article in Cancer Epidemiology, Biomarkers & Prevention entitled “Geographic Patterns in U.S. Lung Cancer Mortality and Cigarette Smoking” led by Alaina Shreves and Dr. Rena Jones.

We applied geospatial statistical methods to describe patterns in lung cancer mortality rates (2005-2018) in relation to patterns in cigarette smoking prevalences (1997-2003) by sex at the US county level. Our findings identified counties where lung carcinogens other than smoking may be driving lung cancer mortality and where further study is needed. You can find the R langauge scripts to recreate the manuscript figures in this GitHub repository.

Update - 2023-02-06: Manuscript published in Cancer Epidemiology, Biomarkers & Prevention Volume 32 Issue 2

Find me on Bluesky and Mastodon

Sat, 29 Oct 2022 00:00:00 +0000

Migrating my professional updates from Twitter to Bluesky and Mastodon. Find me there!

New CRAN Package {ndi}

Sat, 13 Aug 2022 00:00:00 +0000

My fourth R package is published in the Comprehensive R Archive Network named ndi. It computes various geospatial neighborhood deprivation indices (NDI) and other metrics of social vulnerability in the United States.

Two types of NDI are available in the initial version:

based on Messer et al. (2006)
based on Andrews et al. (2020) and Slotman et al. (2022) who use variables chosen by Roux and Mair (2010).

Both are a decomposition of various demographic characteristics from the U.S. Census Bureau American Community Survey 5-year estimates (ACS-5; 2010-2020) pulled by the tidycensus package.

Update - 2022-08-15: The package computes two additional metrics using ACS-5 data (2009-2020) that account for values in nearby (i.e., adjacent) census geographies (i.e., census tracts or counties):

the spatial Racial Isolation Index (RI) based on Anthopolos et al. (2011)
the spatial Educational Isolation Index (EI) based on Bravo et al. (2021)

And retrieves the Gini Index based on Gini (1921).

Update - 2022-08-18: The package link was added to the NCI GIS Portal for Cancer Research Accumulating Data to Optimally Predict Obesity Treatment (ADOPT) Core Measures Environmental Domain site under the “Socioeconomic Deprivation” tab.

Update - 2022-10-01: The package computes an additional metric using ACS-5 data (2009-2020) to compute the Index of Concentration at the Extremes (ICE) based on Feldman et al. (2015) and Krieger et al. (2016)

Update - 2022-12-01: The package computes an additional metric using ACS-5 data (2009-2020) to compute the Dissimilarity Index (DI) based on Duncan & Duncan (1955)

Update - 2023-02-01: The package computes two additional metrics using ACS-5 data (2009-2020):

the aspatial income or racial/ethnic Atkinson Index (AI) based on Atkinson (1970) for specified counties/tracts 2009 onward
the aspatial racial/ethnic Isolation Index (II) based on Shevky & Williams (1949; ISBN-13:978-0837156378) and Bell (1954)
the aspatial racial/ethnic Correlation Ratio (V) based on Bell (1954) and White (1986)
the aspatial racial/ethnic Location Quotient (LQ) based on Merton (1939) and Sudano et al. (2013)
the aspatial racial/ethnic Local Exposure and Isolation metric (LEx/Is) based on Bemanian & Beyer (2017)

New Publication in Environmental Research

Tue, 19 Jul 2022 00:00:00 +0000

I co-authored an article in Environmental Research entitled “Urinary nitrate and sodium in a high-risk area for upper gastrointestinal cancers: Golestan Cohort Study” led by Dr. Arash Etemadi.

We found urinary nitrate was associated with living at a higher elevation and using groundwater for drinking and a positive correlation between median urinary nitrate and sodium and esophageal cancer incidence rates in a rural population in northeastern Iran. My contribution included testing the data for spatial autocorrelation, evaluating elevation and precipitation, formulating the random intercept model, and consulting on the design of Figure 1. This is my first career international study and first international study as a Cancer Prevention Fellow at the National Cancer Institute.

Plotting a Neighborhood Network with ggplot2

Sun, 26 Jun 2022 00:00:00 +0000

Here is an example of plotting a neighborhoods network with ggplot2 using the sf and sfdep packages for counties in a U.S. state.

We can display the “weights” feature of the neighborhoods network as the size of the line segments by scaling the size aesthetic with + scale_size_identity().

# Packages
loadedPackages <- c("ggplot2", "sf", "sfdep", "tigris")
invisible(lapply(loadedPackages, require, character.only = TRUE))

# County geometries of Georgia, U.S.A.
shp_ga <- counties(state = "Georgia", cb = TRUE)

# NAD83/UTM zone 17N geospatial projection
proj_ga <- st_transform(shp_ga, crs = 26917)

# First order contiguity (Queen's case by default)
nb <- st_contiguity(st_geometry(proj_ga)) 

# Contiguity-based spatial weights matrix
nbw <- st_weights(nb)

# County centroids
centroids <- st_centroid(proj_ga)

# Assign latitude and longitude for centroid connections in a dataframe
da <- data.frame(
  from = rep(1:length(nbw), attributes(nbw)$comp$d),
  to = unlist(nb),
  weight = unlist(nbw)
)
da <- cbind(
  da, 
  st_coordinates(centroids)[da$from, 1:2], 
  st_coordinates(centroids)[da$to, 1:2]
)
colnames(da)[4:7] <- c("longitude", "latitude", "long_to", "lat_to")

Plot counties and first order contiguity line segments with size scaled by “weights” feature.

ggplot() +
  geom_sf(data = proj_ga, fill = "white", color = "black") + 
  geom_sf(data = centroids, color = "blue", size = 1) + 
  geom_segment(
    data = da,
    aes(
      x = longitude, 
      y = latitude, 
      xend = long_to, 
      yend = lat_to, 
      size = weight
    ),
    color = "red", 
    alpha = 0.5
  ) +
  scale_size_identity() +
  theme_minimal()

This answer was also posted to Stack Overflow. Some code modified from code by @ StupidWolf’s answer.

First Cover Article

Wed, 04 May 2022 00:00:00 +0000

My first manuscript that was featured as the cover article of a scientific journal was published today in Cancer Epidemiology, Biomarkers & Prevention Volume 31, Issue 5.

I co-first authored the article entitled “A National Map of NCI-Designated Cancer Center Catchment Areas on the 50th Anniversary of the Cancer Centers Program” with Dr. Peter DelNero. The cover image is an artistically modified version of Figure 1 in the manuscript. You can find more details in a previous post.

2021 NCI Director's Award for Emerging Leader

Mon, 14 Feb 2022 00:00:00 +0000

I received a 2021 National Cancer Institute (NCI) Director’s Award for Emerging Leader along with the other members of the Fellows Advisory Board of the Cancer Prevention Fellowship Program (CPFP) in recognition of spearheading innovative opportunities for fellows to develop skills and community during the COVID-19 pandemic.

We were nominated by the CPFP Director, Dr. Lisa Signorello, and were the only recipients of a 2021 NCI Director’s Award within the Division of Cancer Prevention (DCP). Watch the recorded announcement by DCP Director Dr. Philip Castle if you have access to the myNCI portal.

Putting Cancer and Environmental Exposures in (Geo)spatial Perspective

Mon, 07 Feb 2022 00:00:00 +0000

I gave an invited talk at the Center for Science and Technology in the Radiation Protection Division of the Office of Air and Radiation of the U.S. Environmental Protection Agency entitled “Putting cancer and environmental exposures in (geo)spatial perspective.”

The talk was an overview of my work in progress over the past year in the Cancer Prevention Fellowship Program.

New Web Application "Catchment Areas of NCI-Designated Cancer Centers" launched

Tue, 01 Feb 2022 00:00:00 +0000

Catchment Areas of NCI-Designated Cancer Centers is a web-based application that serves as a visualization tool for the geographically defined catchment areas of NCI-Designated Cancer Centers.

The web application provides a geographic scope for:

monitoring cancer trends
identifying pronounced socioeconomic- and health-related disparities
informing high-impact translational science
guiding the implementation of evidence-based interventions in clinical and community settings

The web application will continue to evolve, geographically linking Cancer Centers’ catchment areas to publicly available data sources to aggregate sociodemographic and epidemiologic characteristics across the NCI Cancer Centers Program. I co-led the tool development with Dr. Peter DelNero with guidance from Dr. Zaria Tatalovich and Dr. Robin Vanderpool and technical assistance from Information Management Services, Inc..

New Publication in Cancer Epidemiology, Biomarkers & Prevention

Mon, 31 Jan 2022 00:00:00 +0000

I co-first authored an article in Cancer Epidemiology, Biomarkers & Prevention entitled “A National Map of NCI-Designated Cancer Center Catchment Areas on the 50th Anniversary of the Cancer Centers Program” with Dr. Peter DelNero.

We geographically define and display a national map of 63 NCI-designated Cancer Center catchment areas as well as demonstrate how publicly available data sets can be linked to each catchment area. You can find the R langauge scripts to recreate the manuscript figures in this GitHub repository. This is my first data product during my training as a Cancer Prevention Fellow at the National Cancer Institute.

2022 NCI Director Intramural Innovation Award Program Career Development Award

Tue, 11 Jan 2022 00:00:00 +0000

I received an annual Career Development Award from the National Cancer Institute (NCI) Director Intramural Innovation Award Program that is designed to support development of highly innovative approaches and technology aimed at significant cancer–related problems.

Winning proposals showed potential for significant scientific or public health impact, as well as approach, innovation, and programmatic relevance to the mission of the NCI Division of Cancer Epidemiology and Genetics (DCEG). Recipients will use their award in fiscal year 2022 with an upper limit of $15,000. Four other fellows and staff scientists within DCEG received an award.

New Invited Commentary in the American Journal of Epidemiology

Wed, 08 Dec 2021 00:00:00 +0000

I first authored an invited commentary in the American Journal of Epidemiology entitled “Invited Commentary: Predicting incidence rates of rare cancers: adding epidemiologic and spatial contexts.”

My co-author, Dr. Rena Jones, and I commented on Salmerón et al. (2021a) estimation of county-level incidences rates of rare cancers in Europe. Salmerón et al. (2021b) responded to our invited commentary. This is my first published invited commentary.

Hosted Dr. David Wheeler for the Cancer Prevention & Control Colloquia Series

Tue, 12 Oct 2021 00:00:00 +0000

I hosted Dr. David Wheeler, Associate Professor in the Department of Biostatistics within the School of Medicine at Virginia Commonwealth University (VCU), for the National Cancer Institute Cancer Prevention Fellowship Program Cancer Prevention and Control Colloquia Series.

He gave a talk entitled, “Estimating Neighborhood Disadvantage Indices for Health Outcomes.” I moderated the hour-long virtual session.

Update - 2024-06-29: Sadly, Dr. Wheeler passed away in 2024 after a hard-fought battle with brain cancer.

Update - 2025-04-04: He is honored by the VCU Department of Biostatistics in the School of Public Health with its Dr. David Wheeler Memorial Lecture in Spatial and Cancer Statistics.

14th AACR Conference on The Science of Cancer Health Disparities in Racial/Ethnic Minorities and the Medically Underserved

Fri, 08 Oct 2021 00:00:00 +0000

I was a co-author on a poster entitled “Comparing the association of self-reported race-ethnicity and genetic ancestry with all-cause mortality: A pan-cancer survivor analysis in the PLCO Screening Trial” at the 14th AACR Conference on The Science of Cancer Health Disparities in Racial/Ethnic Minorities and the Medically Underserved by Dr. Derek Brown and Dr. Jacqueline Vo held virtually due to the global COVID-19 pandemic.

This effort was funded by our 2021 William G. Coleman Minority Health and Health Disparities Research Innovation Award.

2021 FDA-NIH-NIST-USDA Joint Agency Microbiome (JAM) Symposium

Mon, 30 Aug 2021 00:00:00 +0000

I presented a poster entitled “Geographic variation in the oral microbiome of NIH-AARP Diet and Health Study participants” at the 2021 FDA-NIH-NIST-USDA Joint Agency Microbiome (JAM) Symposium held virtually due to the global COVID-19 pandemic. This poster was previously presented at the International Society for Environmental Epidemiology 2020 Annual meeting.

International Society for Environmental Epidemiology 2021

Thu, 26 Aug 2021 00:00:00 +0000

I presented my oral abstract entitled “Roadway Proximity and Lung Cancer Risk in NIH-AARP Diet and Health Study Participants” at the 33rd Annual Conference of the International Society for Environmental Epidemiology held virtually due to the global COVID-19 pandemic.

I was also a co-author on three abstracts:

“Residential proximity to animal feeding operations and mortality among postmenopausal women in the Iowa Women’s Health Study” led by Dr. Jessica Madrigal
“Residential proximity to animal feeding operations and risk of lymphohematopoietic cancers in the Iowa Women’s Health Study” led by Dr. Jared Fisher
“Ethylene oxide emissions and risk of breast cancer and Non-Hodgkin lymphoma in a large U.S. cohort” led by Dr. Rena Jones

New Publication in the International Journal of Environmental Research and Public Health

Fri, 25 Jun 2021 00:00:00 +0000

I first authored an article in the International Journal of Environmental Research and Public Health entitled “Ingestion of Nitrate and Nitrite and Risk of Stomach and Other Digestive System Cancers in the Iowa Women’s Health Study.”

We assessed risk of certain digestive cancers and ingestion of nitrate and nitrite from diet and drinking water in a prospective cohort of post-menopausal women in Iowa, USA. This is my first epidemiologic analysis for the Occupational and Environmental Epidemiology Branch at the National Cancer Institute.

Society for Epidemiologic Research 2021

Wed, 23 Jun 2021 00:00:00 +0000

I presented my invited talk entitled “Estimating environmental mixtures in a geospatial context” in a symposium entitled “Implementing geolocation-based exposure assessments” at the Society for Epidemiologic Reserach 2021 annual meeting held virtually due to the global COVID-19 pandemic.

2021 Cancer Prevention Fellowship Merit Award

Tue, 08 Jun 2021 00:00:00 +0000

I received an annual Merit Award from the National Cancer Institute (NCI) Division of Cancer Prevention Cancer Prevention Fellowship Program (CPFP) that recognizes scientific productivity, leadership, and exemplary service to the CPFP, NCI, National Institutes of Health, and other outside organizations in cancer prevention.

Three other postdoctoral Cancer Prevention Fellows also won an award this year: Dr. Jacqueline Vo, Dr. Derek Brown, and Dr. Sydney O’Connor

New Web Application "Spatial Power" launched

Thu, 20 May 2021 00:00:00 +0000

Spatial Power is a web application that calculates the power for spatial statistics. Currently, the tool has one module for the spatial relative risk function.

Dr. Derek Brown and I co-designed the web application with a development team from the NCI Center for Biomedical Informatics & Information Technology. We received funding from a DCEG Informatics Tool Challenge Award and a DCP Trans-Fellowship Research Award Supplement. Please contact me or Dr. Brown with any suggestions for improvements, especially additional spatial statistics you would like to see in the web application.

New Publication in the International Journal of Health Geographics

Fri, 26 Feb 2021 00:00:00 +0000

I co-first authored an article in the International Journal of Health Geographics entitled “sparrpowR: a flexible R package to estimate statistical power to identify spatial clustering of two groups and its application.”

The article introduces the open-source R package sparrpowR and demonstrates how it can be used to calculate statistical power for the spatial relative risk function for etiologic and surveillance research questions. Dr. Derek Brown is my fellow co-first author. See my earlier post about the sparrpowR package that we co-developed.

Lightning Talk at NIEHS Workshop

Wed, 24 Feb 2021 00:00:00 +0000

I presented a lightning talk entitled “Geographic Variation in the Oral Microbiome of NIH-AARP Diet and Health Study Participants” at a National Institute of Environmental Health Sciences workshop entitled Impact of Environmental Exposures on the Microbiome and Human Health held virtually due to the global COVID-19 pandemic.

New Publication in the International Journal of Environmental Research and Public Health

Wed, 10 Feb 2021 00:00:00 +0000

I co-authored an article in the International Journal of Environmental Research and Public Health entitled “Spatial Heterogeneity in Positional Errors: A Comparison of Two Residential Geocoding Efforts in the Agricultural Health Study,” which is included in the Special Issue Spatial Data Uncertainty in Public Health Research.

The article identifies areas with positional error in residential addresses between two geocoding efforts compared to Geographic Positioning System recordings in the Agricultural Health Study, a prospective cancer cohort of licensed pesticide applicators and their spouses in Iowa and North Carolina. My contribution was the data visualization and spatial analysis of the positional error such as a weighted spatial density estimation of the positional error to identify areas of significant improvement in positional accuracy between geocoding efforts while protecting personally identifiable information.

2021 Trans-Fellowship Research Award Supplement

Tue, 09 Feb 2021 00:00:00 +0000

Dr. Derek Brown and I received a Trans-Fellowship Research Award from the National Cancer Institute (NCI) Division of Cancer Prevention Cancer Prevention Fellowship Program (CPFP) to supplement our 2020 DCEG Informatics Tool Challenge Award.

The award ($17,000) will be used to complete the development of our “Spatial Power” webtool for the NCI Division of Cancer Epidemiology and Genetics and fund enhancements such as, for example, a Geographic Information System interface to see the calculated results displayed on an interactive map.

Update: Spatial Power was successfully launched on May 20, 2021. See more details.

Invited Presentation for the DCEG Geospatial Analysis Working Group

Thu, 14 Jan 2021 00:00:00 +0000

Dr. Derek Brown and I presented our “Spatial Power” webtool for the NCI Division of Cancer Epidemiology and Genetics January 2021 monthly Geographic Analysis Working Group meeting. Spatial Power is an upcoming webtool to calculate power for spatial statistics. Stay tuned for its launch.

Update: Spatial Power was successfully launched on May 20, 2021. See more details.

New CRAN Package {envi}

Mon, 21 Dec 2020 00:00:00 +0000

My third R package is published in the Comprehensive R Archive Network named envi. It estimates the ecological niche using presence/absence data and the spatial relative risk function via the sparr package. See the public Github repository for more details.

Thanks to major contributions from Dr. Lance Waller.

Society for Epidemiologic Research 2020

Wed, 16 Dec 2020 00:00:00 +0000

I presented my abstract entitled “Geographic variation in the oral microbiome of Agricultural Health Study applicators” at the Society for Epidemiologic Reserach 2020 annual meeting held virtually due to the global COVID-19 pandemic.

New CRAN Package {gateR}

Tue, 10 Nov 2020 00:00:00 +0000

My second R package is published in the Comprehensive R Archive Network named gateR. It estimates clustering of cytometry cells using markers and the spatial relative risk function via the sparr package. See the public Github repository for more details.

Thanks to major contributions from Dr. Elena Hsieh, Dr. Debashis Ghosh, and Dr. Lance Waller.

2021 William G. Coleman Minority Health and Health Disparities Research Innovation Award

Mon, 02 Nov 2020 00:00:00 +0000

I received a William G. Coleman Minority Health and Health Disparities Research Innovation Award from the National Institute on Minority Health and Health Disparities (NIMHD) for a joint research proposal entitled “HDoCS in PLCO: Health Disparities of Cancer Survivors in the Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial.”

I join a stellar team of three co-awardees Dr. Jacqueline Vo, Dr. Naoise Synnott, and Dr. Derek Brown who are also NCI Cancer Prevention Fellows in the intramural Division of Cancer Epidemiology and Genetics. NIMHD designed this competitive award to support the development of innovative research ideas and concepts contributed by post-doctoral fellows within the NIH Intramural Research Program who have the potential for high impact in any area of minority health and health disparities research.

DCEG Fellows Award for Research Excellence

Tue, 27 Oct 2020 00:00:00 +0000

I received a DCEG Fellows Award for Research Excellence from the NCI Division of Cancer Epidemiology and Genetics for an abstract entitled “Geographic variation in the oral microbiome of NIH-AARP Diet and Health Study participants.”

I was one of nine recipients. The award provides funding for travel to scientific meetings or conferences to fellows who have made exceptional contributions to research projects at DCEG. The award was conferred virtually due to the coronavirus pandemic.

Co-Editor-in-Chief of DFEB

Thu, 01 Oct 2020 00:00:00 +0000

I took over for Dr. Sarah Jackson as Co-Editor-in-Chief of the DCEG Fellows Editorial Board (DFEB). I join Dr. Maeve Bailey-Whyte in this role.

DFEB is a fellow-led group that offers fellows a free scientific document-reviewing service and the opportunity to build scientific editing and review skills of their own. The group provides confidential, anonymous feedback on grammar, organization, and flow of scientific manuscripts, abstracts, proposals, posters, and slide decks.

International Society for Environmental Epidemiology 2020

Wed, 26 Aug 2020 00:00:00 +0000

I presented my abstract entitled “Geographic variation in the oral microbiome of NIH-AARP Diet and Health Study Participants” at the 32nd Annual Conference of the International Society for Environmental Epidemiology held virtually due to the global COVID-19 pandemic.

I was also a co-author on an abstract entitled “Ingestion of nitrate and nitrite from drinking water and diet and cancers of the gastrointestinal tract in Iowa women” led by Dr. Mary Ward.

Creating a hexsticker for the {sparrpowR} package

Thu, 11 Jun 2020 00:00:00 +0000

I present code to create the hexsticker for the sparrpowR package using the hexSticker and spatstat packages. The sparrpowR calculated the statistical power for a spatial relative risk function from the sparr package.

Update - 2020-06-15: The sparrpowR sticker is posted in the hexSticker README.md file

# Packages
loadedPackages <- c("hexSticker", "spatstat")
invisible(lapply(loadedPackages, require, character.only = TRUE))
set.seed(1234) # for reproducibility
par(pty = "s") # for equal proportions

The point pattern to plot within the hexsticker is created using the spatstat package.

# Create windows for separate point patterns
angle <- pi / 3
x <- c(0, 0.5, 0.5 * cos(angle), 0) + 1
y <- c(0, 0, 0.5 * sin(angle), 0) + 1
y1 <- c(0, 0, 0.5 * -sin(angle), 0) + 1
x2 <- c(1 - 0.33, 1.1, 1.25, 1.6, 1.5 + 0.33, 1 - 0.33)
y2 <- c(1.5, 1, 1.433013, 1, 1.5, 1.5)

# Create point patterns
g1 <- rsyst(nx = 20, win = owin(poly = list(x = x, y = y)))
g2 <- rsyst(nx = 20, win = owin(poly = list(x = rev(x), y = rev(y1))))
g3 <- rpoispp(lambda = 20, win = owin(poly = list(x = x2, y = y2)))

# Set marks for point patterns (for colors)
marks(g1) <- 1
marks(g2) <- 2
marks(g3) <- 1

# Combine point patterns and set marks to be factors
g4 <- superimpose(g1, g2, g3)
marks(g4) <- as.factor(marks(g4))

I use the hexSticker package to create the hexsticker which layers the sticker elements into a hexagon window. In order to hide points within the g4 object that reside outside the hexagon window I set the white_around_sticker argument as TRUE. I choose the Balance Within color palette.

# Create hexSticker

hexSticker::sticker(
  ~plot.ppp(
    g4, 
    cols = c("#ee9a00", "#698b69"),
    show.all = F, cex = 0.5,
    pch = c(8, 8)
  ),
  package = "sparrpowR",
  p_size = 5,
  p_color = "#ffe4b5",
  s_x = 0.7, s_y = 0.1,
  s_width = 3.33, s_height = 3.33,
  h_fill = "#344960",
  h_color = "#8b3a3a",
  dpi = 1000,
  white_around_sticker = TRUE
)

After removing the white image background in your favorite image editor you can replicate the final hexsticker.

And when you connect three stickers you can create a larger design:

New CRAN Package {sparrpowR}

Wed, 10 Jun 2020 00:00:00 +0000

My first R package is in the Comprehensive R Archive Network named sparrpowR. It provides a statistical power calculation for the spatial relative risk function via the sparr package. See the public Github repository for more details. The featured image comes from the package vignette.

Thanks to Dr. Derek Brown as my co-first author as well as major contributions from Tim Myers, and Dr. Mitch Machiela

Animating Spatio-Temporal COVID-19 Data

Mon, 08 Jun 2020 00:00:00 +0000

I present an update to my previous posts #1 and #2. This update can also be found on a public GitHub repository.

Starting May 17, 2020 the DC Mayoral Office began releasing testing information by neighborhood on their coronavirus data portal. Molly Tolzmann zmotoly added this feature to her publicly available Google Sheet presented at the DC health planning neighborhood level.

Here, I present the daily 5-day rolling average test positivity rate (TPR) for COVID-19 cases from May 25 to June 17, 2020 in an animated Graphics Interchange Format (GIF) using the open-source software language R. A TPR is the proportion of positive tests for every administered test. For example, the World Health Organization recommends a test positivity rate below 5% (1 out of 20 tests, TPR = 0.05) before reopening. A 5-day rolling average is the mean value of a daily TPR and the TPR of the four days prior.

Important Note: There are missing data for certain dates. Case information is missing for May 22nd and 27th. Testing information is missing for May 22nd, 23rd, 24th, and 27th. Missing data are considered NA but included in the 5-day rolling averages. Therefore, 5-day rolling averages that include these dates are more unstable. Because reporting case information at the DC health planning neighborhood level became available on May 7th, daily incident case information is available starting May 8th and 5-day rolling average cases are available starting May 12th. Because testing information became available on May 20th, daily testing information is available starting May 21st and 5-day averages are available starting May 25th.

Necessary packages and settings for the exercise include:

# Packages
loadedPackages <- c("dplyr", "gganimate", "ggplot2", "googlesheets4", "sf", "tidyr", "transformr")
invisible(lapply(loadedPackages, require, character.only = T))

# Settings
gs4_deauth() # no Google authorization necessary because we are not reading a public repo

Data Importation

I merge the District of Columbia Health Planning Neighborhoods boundaries and Molly Tolzmann’s zmotoly collation of the cumulative cases and tests from start of the SARS-CoV-2 outbreak. I clean up variable names that are easier for R to use.

# District of Columbia Health Planning Neighborhoods
gis_path <- "https://opendata.arcgis.com/datasets/de63a68eb7674548ae0ac01867123f7e_13.geojson"
dc <- st_read(gis_path)

## Reading layer `DC_Health_Planning_Neighborhoods' from data source 
##   `https://opendata.arcgis.com/datasets/de63a68eb7674548ae0ac01867123f7e_13.geojson' 
##   using driver `GeoJSON'
## Simple feature collection with 51 features and 9 fields
## Geometry type: POLYGON
## Dimension:     XY
## Bounding box:  xmin: -77.11976 ymin: 38.79165 xmax: -76.9094 ymax: 38.99556
## Geodetic CRS:  WGS 84

# District of Columbia SAR-CoV-2 Data prepared by @zmotoly
covid_path <- "https://docs.google.com/spreadsheets/d/1u-FlJe2B1rYV0obEosHBks9utkU30-C2TSkHka6AVS8/edit#gid=1923705378"
covid <- read_sheet(
  ss = covid_path,
  sheet = 2, # second sheet
  skip = 2
)  # skip 1st row of annotation

# Fix column names
names(covid) <- sub("\n", "", names(covid)) # remove extra line in column names
names(covid) <- gsub(" ", "_", names(covid)) # replace spaces with underscore
names(covid) <- gsub("Total_cases", "Cumulative", names(covid)) # Change to one word
names(covid) <- gsub("Total_tests", "Tested_", names(covid)) # Change to one word
names(covid) <- gsub("Cases_per_1000", "Case.Rate", names(covid)) # Change to one word
names(covid) <- gsub("Tests_per_1000", "Test.Rate_", names(covid)) # Change to one word
names(covid) <- gsub("__", "_", names(covid)) # replace double underscores

# Merge
dc_covid <- left_join(dc, covid, by = join_by("CODE" == "NB_Code"))

# Spatial Projection
## UTM zone 18N (Washington, DC)
dc_covid_proj <- st_transform(dc_covid, crs = 32618)

Data Management

The 5-day rolling average test positivity rate is not provided in Molly Tolzmann’s zmotoly Google Sheet and I calculate it for every day data are available. First, I find the daily incident cases from the reported cumulative cases.

# Uses ggplot2 package
## helpful material: https://cengel.github.io/rspatial/4_Mapping.nb.html

# Step 1) Daily incident cases
CoV_DC_loop <- dc_covid_proj %>% 
  select(starts_with("cumulative")) %>% 
  st_drop_geometry()
CoV_DC_loop <- CoV_DC_loop[, rev(1:length(CoV_DC_loop))] # reorder dates in ascending order
CoV_DC_loop$Cumulative_May_22 <- NA # missing data
CoV_DC_loop$Cumulative_May_27 <- NA # missing data
CoV_DC_loop$Cumulative_Jun_8 <- NA # missing data
CoV_DC_loop$Cumulative_Jun_10 <- NA # missing data
CoV_DC_loop <- CoV_DC_loop[, c(1:15, 39, 16:19, 40, 20:30, 41, 31, 42, 32:38)] # reorder for consistent dates
i <- NULL # initiate indexing

# Empty matrices
mat_inc <- matrix(ncol = length(CoV_DC_loop)-1, nrow = nrow(CoV_DC_loop)) # for values
col_lab <- vector(mode = "character", length = length(CoV_DC_loop)-1)     # for names

for (i in 1:length(CoV_DC_loop)-1) {
  mat_inc[ , i] <- ifelse(
    CoV_DC_loop[ , i+1] - CoV_DC_loop[ , i] < 0,
    NA, 
    CoV_DC_loop[ , i+1] - CoV_DC_loop[ , i]
  )
  col_lab[i] <- paste(
    sub("Cumulative", names(CoV_DC_loop[i+1]), replacement = "incident")
  )
  if(i == length(CoV_DC_loop))
    mat_inc <- data.frame(mat_inc)
    colnames(mat_inc) <- col_lab
}

CoV_DC_df <- cbind(dc_covid_proj, mat_inc) # merge with original dataset

Then I calculate 5-day rolling average cases (rate included, too).

# Step 2) 5-Day Average Incident Case and Case Rate
CoV_DC_loop <-  CoV_DC_df %>% 
  select(starts_with("incident")) %>%
  st_drop_geometry()
i <- NULL # reinitiate indexing

# Empty matrices
mat_5d <- matrix(ncol = length(CoV_DC_loop)-4, nrow = nrow(CoV_DC_loop)) # for values
mat_5dr <- matrix(ncol = length(CoV_DC_loop)-4, nrow = nrow(CoV_DC_loop)) # for values
col_lab <- vector(mode = "character", length = length(CoV_DC_loop)-4)     # for names
col_labr <- vector(mode = "character", length = length(CoV_DC_loop)-4)    # for names

for (i in 1:(length(CoV_DC_loop)-4)) {
  mat_5d[ , i] <- rowMeans(CoV_DC_loop[ , i:(i+4)], na.rm = T)
  mat_5dr[ , i] <-  mat_5d[ , i] / CoV_DC_df$Population_.2018_ACS. * 1000
  col_lab[i] <- paste(sub("incident", names(CoV_DC_loop[(i+4)]), replacement = "average"))
  col_labr[i] <- paste(sub("incident", names(CoV_DC_loop[(i+4)]), replacement = "average.rate"))
  if(i == (length(CoV_DC_loop)-4))
    mat_5d <- data.frame(mat_5d)
    mat_5dr <- data.frame(mat_5dr)
    colnames(mat_5d) <- col_lab
    colnames(mat_5dr) <- col_labr
    out <- cbind(mat_5d, mat_5dr)
}

CoV_DC_df <- cbind(CoV_DC_df, out) # merge with original dataset

I also find the daily tests administered from the reported cumulative tests (daily test positivity rate included, too), and hen I calculate a 5-day rolling average tests.

# Step 3) Daily Tests
CoV_DC_loop <- CoV_DC_df %>% 
  select(starts_with("Tested"), starts_with("incident")) %>%
  st_drop_geometry()
CoV_DC_loop <- CoV_DC_loop[,c(rev(1:23), 36:64)]   # reorder dates in ascending order
CoV_DC_loop$Tested_May_22 <- NA # missing data
CoV_DC_loop$Tested_May_23 <- NA # missing data
CoV_DC_loop$Tested_May_24 <- NA # missing data
CoV_DC_loop$Tested_May_27 <- NA # missing data
CoV_DC_loop$Tested_Jun_9 <- NA  # missing data
CoV_DC_loop$Tested_Jun_10 <- NA # missing data
CoV_DC_loop <- CoV_DC_loop[,c(1:2, 53:55, 3:4, 56, 5:16, 57:58, 17:52)] # reorder for consistent dates
i <- NULL # reinitiate indexing

# Empty matrices
mat_inc <- matrix(ncol = length(CoV_DC_loop)/2-1, nrow = nrow(CoV_DC_loop))  # for values
mat_inc2 <- matrix(ncol = length(CoV_DC_loop)/2-1, nrow = nrow(CoV_DC_loop)) # for values
col_lab <- vector(mode = "character", length = length(CoV_DC_loop)/2-1)      # for names
col_lab2 <- vector(mode = "character", length = length(CoV_DC_loop)/2-1)     # for names

for (i in 1:(length(CoV_DC_loop)/2-1)) {
  mat_inc[ , i] <- ifelse(
    CoV_DC_loop[ , (i+1)] - CoV_DC_loop[ , i] < 0,
    NA, 
    CoV_DC_loop[ , (i+1)] - CoV_DC_loop[ , i]
  )
  mat_inc2[ , i] <-  CoV_DC_loop[ , (length(CoV_DC_loop)/2+i)] / mat_inc[ , i]
  col_lab[i] <- paste(sub("Tested", names(CoV_DC_loop[i+1]), replacement = "testing"))
  col_lab2[i] <- paste(sub("Tested", names(CoV_DC_loop[i+1]), replacement = "positivity"))
  if(i == (length(CoV_DC_loop)/2-1))
    mat_inc <- data.frame(mat_inc)
    mat_inc2 <- data.frame(mat_inc2)
    colnames(mat_inc) <- col_lab
    colnames(mat_inc2) <- col_lab2
    out <- cbind(mat_inc, mat_inc2)
}

CoV_DC_df <- cbind(CoV_DC_df, out) # merge with original dataset

A 5-day rolling average test positivity is calculated by dividing the 5-day rolling average cases by the 5-day rolling average tests.

# Step 4) 5-day average testing and positivity
CoV_DC_loop <-  CoV_DC_df %>% 
  select(starts_with("testing_"), starts_with("average_")) %>%
  st_drop_geometry()
CoV_DC_loop <- CoV_DC_loop[, -c(29:37)]   # restrict to dates with testing
i <- NULL # reinitiate indexing

# Empty matrices
mat_5d <- matrix(ncol = 28-4, nrow = nrow(CoV_DC_loop))  # for values
mat_5dr <- matrix(ncol = 28-4, nrow = nrow(CoV_DC_loop)) # for values
col_lab <- vector(mode = "character", length = 28-4)     # for names
col_labr <- vector(mode = "character", length = 28-4)    # for names

for (i in 1:(28-4)) {
  mat_5d[ , i] <- rowMeans(CoV_DC_loop[ , i:(i+4)], na.rm = T)
  mat_5dr[ , i] <-  CoV_DC_loop[ , 28+i] / mat_5d[ , i]
  col_lab[i] <- paste(sub("testing", names(CoV_DC_loop[(i+4)]), replacement = "test.avg"))
  col_labr[i] <- paste(sub("testing", names(CoV_DC_loop[(i+4)]), replacement = "posit.avg"))
  if(i == (28-4))
    mat_5d <- data.frame(mat_5d)
    mat_5dr <- data.frame(mat_5dr)
    colnames(mat_5d) <- col_lab
    colnames(mat_5dr) <- col_labr
    out <- cbind(mat_5d, mat_5dr)
}

CoV_DC_df <- cbind(CoV_DC_df, out) # merge with original dataset

# Conserve memory
rm(out, col_lab, col_labr, CoV_DC_loop, mat_5d, mat_5dr, mat_inc, mat_inc2)

I use the ggplot2 and gganimate packages to create a GIF of the daily 5-day rolling average TPR for COVID-19 cases from May 25 to June 17, 2020. The packages require converting the data from wide- to long-format.

# Convert to long format
CoV_DC_5dtest <-  CoV_DC_df %>% 
  select(1:25, starts_with("posit.avg")) %>%
  pivot_longer(
    cols = starts_with("posit.avg"),
    values_to = "posit.avg",
    names_to = c("Posit.Avg", "month", "day"),
    names_sep = "_"
  ) %>%  
  unite("date_reported", month:day) %>%
  select(-Posit.Avg) %>%
  mutate(date_reported = as.Date(date_reported, format="%B_%d"))

Finally, some 5-day rolling average TPR are invalid for two possible reasons. If any district had zero (0) administered tests, then the TPR (a ratio) would be undefined. Also, if there were more reported positive cases than administered tests (e.g., reporting backlog) then the TPR would be above one (1.0). In these cases, I present these instances as NA values.

# Clean-up
## Any Inf values (zero tests in denominator) set as NA
CoV_DC_5dtest$posit.avg <- ifelse(
  is.infinite(CoV_DC_5dtest$posit.avg), NA, CoV_DC_5dtest$posit.avg
)

## Any values above 1.0 (more positive tests than werer administered) set as NA
CoV_DC_5dtest$posit.avg <- ifelse(
  CoV_DC_5dtest$posit.avg > 1, NA, CoV_DC_5dtest$posit.avg
)

Animated Map

The following is the code to create a GIF of the daily 5-day rolling average TPR for COVID-19 cases from May 25 to June 17, 2020. I chose non-alarmist colors similar to previous posts #1 and #2. Grey-colored areas have missing (NA) or no information.

# 5-day average test positivity
g1 <- CoV_DC_5dtest %>%                   # data
  ggplot() +                              # initial plot
  geom_sf(
    aes(fill = posit.avg),
    color = NA
  ) +                                     # add polygons
  scale_fill_gradient(
    name = "5-day average rate",          # color fill
    low = "lavenderblush",
    high = "navyblue", 
    na.value = "grey80",
    breaks = range(CoV_DC_5dtest$posit.avg, na.rm = T)
  ) +
  theme_void() +
  theme(
    plot.margin = margin(t = 20, r = 50, b = 40, l = 10, unit = "pt"),
    legend.position = "bottom",           # legend position
    text = element_text(size = 10)        # set font size
  ) +     
  coord_sf() +                            # force equal dimensions
  transition_time(date_reported) +        # animate by date
  labs(
    title = "5-Day Average SARS-CoV-2 Test Positivity\nDate: {frame_time}"
  )                                       # add title

animate(g1, end_pause = 30)               # animate

Overall, my qualitative assessment of the daily 5-day rolling average TPR for COVID-19 cases from May 25 to June 17, 2020 is a decrease in TPR overtime, especially after June 3rd. Ward 4 (Northern-most section) appears to have consistently high 5-day rolling average TPR overtime.

The provided maps are not intended to inform decision-making. Instead, I provide the the open-source code to download, manage, and visualize publicly available data. Future steps (in addition to one listed in my previous post) include, displaying TPR above or below the the 5% WHO recommendation and conducting spatial statistical analysis such as, for example, assessing the presence of spatio-temporal clustering. Other values (e.g., daily TPR, incident cases per 1,000) can be animated by modifying the above code, which is also provided in my public GitHub repository.

Thanks, again, to Molly Tolzmann zmotoly for the data collation, management, and custodianship.

A Place for Spatial Analysis in the Cancer Control Continuum

Fri, 05 Jun 2020 00:00:00 +0000

I gave an invited talk at the National Cancer Institute entitled “A place for spatial analysis in the Cancer Control Continuum.”

The talk was an overview of my work in progress over the past year in the Cancer Prevention Fellowship Program. The talk was recorded, so please contact me if you are interested in viewing the talk.

DCEG Informatics Tool Challenge Award

Tue, 02 Jun 2020 00:00:00 +0000

I received an Informatics Tool Challenge Award from the NCI Division of Cancer Epidemiology and Genetics for a project entitled “sparrpowR: A flexible R package and webtool to estimate statistical power of a spatial cluster detection technique.”

My co-Principal Investigator, Dr. Derek Brown (DCEG/OEEB), is another postdoctoral Cancer Prevention Fellow and we are joined by our supervisors Dr. Rena Jones (DCEG/OEEB) and Dr. Mitch Machiela (DCEG/ITEB), respectively, as well as Tim Myers (DCEG/ITEB). The funds ($20,000) will be used by the NCI Center for Biomedical Informatics & Information Technology to develop a web application of our R package called sparrpowR.

Update #1: sparrpowR was successfully published in the Comprehensive R Archive Network on June 10, 2020. See more details.

Update #2: Spatial Power was successfully launched on May 20, 2021. See more details.

Cluster Detection in PCoA Space using Kernel Density Estimation

Sat, 30 May 2020 00:00:00 +0000

I present code to identify relative spatial clustering in multidimensional scaling / principal coordinate analysis (MDS/PCoA) space. I use a spatial relative risk function risk from the sparr package. I use a spatial segregation model relrisk from the spatstat package.

See my previous post about using the spatial segregation model.

I use microbiome data from the dietswap data set in the microbiome package part of Bioconductor. The diet swap data set represents a study with African and African American groups undergoing a two-week diet swap reported in O’Keefe et al. Nat. Comm. 6:6342, 2015. I follow the Ordination Analysis example by Leo Lahti, Sudarshan Shetty et al. 2020.

MDS/PCoA summarizes and attempts to represent inter-object dissimilarity in a low-dimensional space, not necessarily using Euclidean Distances (used by Principal Component Analysis)

Necessary packages and settings for the exercise include:

# Packages
loadedPackages <- c("ggplot2", "microbiome", "phyloseq", "sf", "sparr", "spatstat")
invisible(lapply(loadedPackages, require, character.only = T))
set.seed(4235421) # for reproducibility

I prepare the data by choosing limiting to taxa with 90% prevalence and a detection limit above 0.1%. I then ordinate Bray-Curtis dissimilarity matrix by MDS/PCoA.

# data
data(dietswap)
pseq <- dietswap

# Convert to compositional data
pseq.rel <- transform(pseq, "compositional")

# Pick core taxa with with the given prevalence and detection limits
pseq.core <- core(pseq.rel, detection = 0.1/100, prevalence = 90/100)

# Use relative abundances for the core
pseq.core <- transform(pseq.core, "compositional")

# Ordinate the data
ord <- ordinate(pseq.core, "MDS", "bray")

Two Group Comparison

As an example of relative spatial clustering between two groups I compare the two levels of the nationality variable “AMR” (America) and “AFR” (Africa). A plot of the samples by the two groups is displayed below:

# Plot ordination
plot_ordination(pseq.core, ord, color = "nationality") +
  geom_point(size = 1) + 
  theme(
    panel.grid.major = element_blank(),
    panel.grid.minor = element_blank(),
    panel.background = element_blank(),
    axis.line = element_line(colour = "black")
  )

To detect significant relative spatial clustering between the two nationalities, I use the risk function from the sparr package. The function requires data as a planar point process (ppp) object. Here, I use the default settings for the risk function except I use adaptive-bandwidth selection. Significant areas are defined as areas that exceed a two-tailed 0.05 alpha level.

# Convert to ppp object
test_df <- data.frame(
  "x" = ord$vectors[, 1],
  "y" = ord$vectors[, 2],
  "g" = pseq@sam_data@.Data[[3]]
)
test_sf <- st_as_sf(test_df, coords = c("x", "y"))
test_pp <- as.ppp(test_sf)

# Spatial Relative Risk Function
f0 <- risk(f = test_pp, tolerate = TRUE, adapt = TRUE)

# Plot of significant areas
f0_p <- f0$P
f0_p$v <- factor(
  ifelse(
    f0_p$v < 0.025, 
    "AAM",
    ifelse(f0_p$v > 0.975, "AFR","insignificant")
  ),
  levels = c("AFR", "insignificant", "AAM"))
plot.ppp(
  test_pp, 
  ann = TRUE, 
  axes = TRUE,
  leg.side = "left",
  xlab = "Axis.1 [54.5%]",
  ylab = "Axis.2 [20.4%]",
  main = "Significant area, adaptive bandwidth, alpha = 0.05",
  cols = c("coral1", "cadetblue3"),
  pch = 16
)
plot.im(
  f0_p, add = TRUE, show.all = TRUE, main = "", col = c("cadetblue4","grey80","coral4")
)
plot.ppp(
  test_pp, add = TRUE, main = "", cols = c("coral1", "cadetblue3"), pch = 16
)

The African group has one significant cluster relative to the American Group (colored blue), which has two significant clusters (colored red). The areas in grey are where the African and American groups spatially covary together that is indiscernible from spatial randomness.

Multi-Group Comparison

As an example of relative spatial clustering between three groups I compare the two levels of the bmi_group variable “lean”, “overweight’ and “obese”. A plot of the samples by the three groups is displayed below:

# Plot ordination
plot_ordination(pseq.core, ord, color = "bmi_group") +
  geom_point(size = 1) +
  labs(color = "BMI Group") +
  theme(
    panel.grid.major = element_blank(),
    panel.grid.minor = element_blank(),
    panel.background = element_blank(),
    axis.line = element_line(colour = "black")
  )

To detect significant relative spatial clustering between three BMI groups, I use a spatial segregation model relrisk from the spatstat package. The function requires data as a planar point process (ppp) object. Here, I use the default settings for the relrisk function. I present 1) the most common BMI group across the MDS/PCoA space, and 2) areas that are significantly different from the null expectation for each BMI group. See my previous post about using the spatial segregation model.

# Convert to ppp object
test_df <- data.frame(
  "x" = ord$vectors[, 1],
  "y" = ord$vectors[, 2],
  "g" = pseq@sam_data@.Data[[8]]
)
test_sf <- st_as_sf(test_df, coords = c("x", "y"))
test_pp <- as.ppp(test_sf)

# Spatial Segregation model
f1 <- relrisk.ppp(test_pp, se = T)

# Plot of most common BMI Group
wh <- im.apply(f1$estimate, which.max)
types <- levels(marks(test_pp))
wh <- eval.im(types[wh]) # most common 

plot.ppp(
  test_pp,
  ann = TRUE,
  axes = TRUE,
  leg.side = "left",
  xlab = "Axis.1 [54.5%]",
  ylab = "Axis.2 [20.4%]",
  main = "Most common type",
  cols = c("coral1", "chartreuse3", "cadetblue3"),
  pch = 16)
plot.im(
  wh, add = TRUE, show.all = TRUE, main = "", col = c("coral4","cadetblue4","chartreuse4")
)
plot.ppp(
  test_pp, add = TRUE, main = "", cols = c("coral1", "chartreuse3", "cadetblue3"), pch = 16
)

The “lean” BMI group appears most common in one area of the MDS/PCoA space (colored red), while the other two groups have numerous clusters. This plot does not indicate where eat group are significantly different from their null expectations. For example, our null expectation for the “lean” BMI group is a probability of about 0.25 (56/222).

# plot.im(wh, main="Most common type", ribargs = list(las = 1))

# Plot of significantly different areas
alpha <- 0.05                           # alpha
z <- qnorm(alpha/2, lower.tail = F)     # z-statistic
f1$u <- f1$estimate + z*f1$SE           # Upper CIs
f1$l <- f1$estimate - z*f1$SE           # Lower CIs
mu_0 <- as.vector(table(marks(test_pp))/test_pp$n) # null expectations by type
f1$p <- f1$estimate # copy structure of pixels, replace values
for (i in 1:length(f1$p)) {
  f1$p[[i]]$v <- factor(
    ifelse(
      mu_0[i] > f1$u[[i]]$v,
      "lower",
      ifelse( mu_0[i] < f1$l[[i]]$v, "higher", "none")
    ),
    levels = c("lower", "none", "higher")
  )
}

# Plot significant p-values
plot(f1$p, main = "Significant difference from null?")

The “lean” BMI group is significantly higher than expected in one large cluster. The “overweight” BMI group is significantly higher than expected in one large cluster and one small cluster. The “obese” BMI group is significantly higher in numerous small clusters.

Test Positivity Rate of Cumulative SARS-CoV-2 Cases in the District of Columbia

Sat, 23 May 2020 00:00:00 +0000

I present an update to my previous post. Starting May 17, 2020 the DC Mayoral Office began releasing testing information by neighborhood on their coronavirus data portal. Molly Tolzmann zmotoly added this feature to her publicly available Google Sheet presented at the DC health planning neighborhood level. The update here can also be found on a public GitHub repository.

Here, I present the test positivity rate (TPR) for cumulative COVID-19 cases as of May 21, 2020. A TPR is the proportion of positive tests for every administered test. For example, the World Health Organization recommends a test positivity rate below 5% (1 out of 20 tests, TPR = 0.05) before reopening.

Important Note: The following show cumulative case rate per 1,000 since the beginning of the COVID-19 outbreak in DC. Therefore, the the stats do not reflect the number of people currently infected with SARS-CoV-2.

Necessary packages and settings for the exercise include:

# Packages
loadedPackages <- c("dplyr", "ggplot2", "googlesheets4", "leaflet", "sf", "stringr", "widgetframe")
invisible(lapply(loadedPackages, require, character.only = T))

# Settings
gs4_deauth() # no Google authorization necessary because we are not reading a public repo

I merged the District of Columbia Health Planning Neighborhoods boundaries and Molly Tolzmann’s zmotoly collation of the cumulative cases and tests from start of the SARS-CoV-2 outbreak. I created a new variable for the test positivity rate test_positivity_May_21.

# District of Columbia Health Planning Neighborhoods
gis_path <- "https://opendata.arcgis.com/datasets/de63a68eb7674548ae0ac01867123f7e_13.geojson"
dc <- st_read(gis_path)

## Reading layer `DC_Health_Planning_Neighborhoods' from data source 
##   `https://opendata.arcgis.com/datasets/de63a68eb7674548ae0ac01867123f7e_13.geojson' 
##   using driver `GeoJSON'
## Simple feature collection with 51 features and 9 fields
## Geometry type: POLYGON
## Dimension:     XY
## Bounding box:  xmin: -77.11976 ymin: 38.79165 xmax: -76.9094 ymax: 38.99556
## Geodetic CRS:  WGS 84

# District of Columbia SAR-CoV-2 Data prepared by @zmotoly
covid_path <- "https://docs.google.com/spreadsheets/d/1u-FlJe2B1rYV0obEosHBks9utkU30-C2TSkHka6AVS8/edit#gid=1923705378"
covid <- read_sheet(
  ss = covid_path,
  sheet = 2, # second sheet
  skip = 2 # skip 1st row of annotation
)  
names(covid) <- sub("\n", "", names(covid))   # remove extra line in column names
names(covid) <- gsub(" ", "_", names(covid))  # replace spaces with underscore

# Test Positivity Rate
covid$test_positivity_May_21 <- covid$Total_cases_May_21 / covid$Total_tests_May_21

# Merge
dc_covid <- left_join(dc, covid, by = join_by("CODE" == "NB_Code"))

# Spatial Projection
## UTM zone 18N (Washington, DC)
dc_covid_proj <- st_transform(dc_covid, crs = 32618)

Static Map

I use the ggplot2 package to plot the test positivity rate (as of May 21).

# Uses ggplot2 package
## helpful material: https://cengel.github.io/rspatial/4_Mapping.nb.html
## Plot of cumulative cases per 1,000
ggplot() +                                                # initialize ggplot object
  geom_sf(                                                # make a polygon
    data = dc_covid_proj,                                     # data frame
    aes(fill = cut_interval(test_positivity_May_21, 5)),  # variable to use for filling
    colour = "white"
  ) +                                     # color of polygon borders
  scale_fill_brewer(
    "Test positivity rate",               # title of colorkey 
    palette = "Purples",                  # fill with brewer colors 
    na.value = "grey67",                  # color for NA (The National Mall)
    direction = 1,                        # reverse colors in colorkey
    guide = guide_legend(reverse = T)     # reverse order of colokey
  ) +  
  ggtitle(
    "Test Positivity Rate of cumulative SARS-CoV-2 cases as of May 21, 2020"
  ) +                                     # add title
  theme(
    line = element_blank(),               # remove axis lines
    axis.text = element_blank(),          # remove tickmarks
    axis.title = element_blank(),         # remove axis labels
    panel.background = element_blank(),   # remove background gridlines
    text = element_text(size = 10)
  ) +                                     # set font size
  coord_sf()                              # both axes the same scale

And an interactive map with the leaflet package.

# Work with unprojected spatialpolygonsdataframe
## Project to WGS84 EPSG:4326
CoV_DC_WGS84 <- st_transform(dc_covid_proj, crs = 4326)

## Create Popups
dc_health <- str_to_title(CoV_DC_WGS84$Neighborhood_Name)
dc_health[c(11, 25, 41, 49)] <- c(
  "DC Medical Center", "GWU", "National Mall", "SW/Waterfront"
)
CoV_DC_WGS84$popup1 <- paste(
  dc_health, 
  ": ",
  format(round(CoV_DC_WGS84$Total_cases_May_21, digits = 0), big.mark = ",", trim = T),
  " cumulative cases",
  sep = ""
)
CoV_DC_WGS84$popup2 <- paste(
  dc_health, 
  ": ",
  format(round(CoV_DC_WGS84$Cases_per_1000_May_21, digits = 0), big.mark = ",", trim = T),
  " cumulative cases per 1,000", 
  sep = ""
)
CoV_DC_WGS84$popup3 <- paste(
  dc_health, 
  ": ",
  format(round(CoV_DC_WGS84$test_positivity_May_21, digits = 2), big.mark = ",", trim = T),
  " test positivity rate", 
  sep = ""
)

## Set Palettes
pal_cum <- colorNumeric(
  palette = "Purples", domain = CoV_DC_WGS84$Total_cases_May_21, na.color = "#555555"
)
pal_rate <- colorNumeric(
  palette = "Purples", domain = CoV_DC_WGS84$Cases_per_1000_May_21, na.color = "#555555"
)
pal_test <- colorNumeric(
  palette = "Purples", domain = CoV_DC_WGS84$Tests_per_1000_May_21, na.color = "#555555"
)
pal_weight <- colorNumeric(
  palette = "Purples", domain = CoV_DC_WGS84$test_positivity_May_21, na.color = "#555555"
)

## Create leaflet plot
lf <- leaflet(CoV_DC_WGS84) %>%                        # initial data
  setView(lng = -77, lat = 38.9, zoom = 11) %>%                  # starting coordinates
  addTiles() %>% # basemap
  addPolygons(
    data = CoV_DC_WGS84,
    color = "black", 
    weight = 1, 
    smoothFactor = 0.5, 
    opacity = 1,
    fillOpacity = 0.67, 
    fillColor = ~pal_cum(Total_cases_May_21), 
    popup = ~popup1,
    highlightOptions = highlightOptions(color = "white", weight = 2, bringToFront = TRUE),
    group = "Cases"
  ) %>%
  addPolygons(
    data = CoV_DC_WGS84, 
    color = "black", 
    weight = 1, 
    smoothFactor = 0.5, 
    opacity = 1,
    fillOpacity = 0.67, 
    fillColor = ~pal_rate(Cases_per_1000_May_21), 
    popup = ~popup2,
    highlightOptions = highlightOptions(color = "white", weight = 2, bringToFront = TRUE),
    group = "Cases per 1,000"
  ) %>%
  addPolygons(
    data = CoV_DC_WGS84, 
    color = "black", 
    weight = 1, 
    smoothFactor = 0.5, 
    opacity = 1,
    fillOpacity = 0.67, 
    fillColor = ~pal_test(Tests_per_1000_May_21), popup = ~popup3,
    highlightOptions = highlightOptions(color = "white", weight = 2, bringToFront = TRUE),
    group = "Tests per 1,000"
  ) %>%
  addPolygons(
    data = CoV_DC_WGS84, 
    color = "black", 
    weight = 1, 
    smoothFactor = 0.5, 
    opacity = 1,
    fillOpacity = 0.67, 
    fillColor = ~pal_weight(test_positivity_May_21), 
    popup = ~popup3,
    highlightOptions = highlightOptions(color = "white", weight = 2, bringToFront = TRUE),
    group = "Test positivity rate"
  ) %>%
  addLayersControl(
    overlayGroups = c(
      "Cases", "Cases per 1,000", "Tests per 1,000", "Test positivity rate"
    ), # layer selection
    options = layersControlOptions(collapsed = FALSE)
  ) %>%     
  addLegend(
    "topright", 
    pal = pal_cum, 
    values = ~Total_cases_May_21,                  # legend for cases
    title = "Cumulative COVID-19 cases", 
    opacity = 1, 
    na.label = "No Data", 
    group = "Cases"
  ) %>%
  addLegend(
    "topright", 
    pal = pal_rate, 
    values = ~Cases_per_1000_May_21,              # legend for rate
    title = "Cumulative cases per 1,000", 
    opacity = 1, 
    na.label = "No Data", 
    group = "Cases per 1,000"
  ) %>%
  addLegend(
    "topright", 
    pal = pal_test, 
    values = ~Tests_per_1000_May_21,              # legend for test positivity rate
    title = "Cumulative tests per 1,000", 
    opacity = 1, 
    na.label = "No Data", 
    group = "Tests per 1,000"
  ) %>%
  addLegend(
    "topright", 
    pal = pal_weight, 
    values = ~test_positivity_May_21,              # legend for test positivity rate
    title = "Cumulative test positivity rate", 
    opacity = 1, 
    na.label = "No Data", 
    group = "Test positivity rate"
  ) %>%
  hideGroup(
    c("Cases", "Cases per 1,000", "Tests per 1,000", "Test positivity rate")
  ) %>% # no data shown (default)
  addMiniMap(position = "bottomleft") # add mini map

frameWidget(lf, width='100%')

As of May 21, 2020 the highest positive testing rate of cumulative SARS-CoV-2 cases has occurred in the Stadium Armory (almost 1 out of 2 tests return positive). The DC Jail is located in the Stadium Armory as previously noted. An East/West divide is evident in the case rate after accounting for testing. Also, no neighborhood is below the 5% WHO recommendation; however, this metric is more appropriately used for incident (or recent) tests such as, for example, a 7 day average than for cumulative cases, so interpret these findings with caution.

The provided maps are not intended to inform decision-making. Instead, I provide the the open-source code to download, manage, and visualize publicly available data. Future steps are similar to my previous post, linking other demographic information to each DC Health Planning Neighborhood (e.g., housing occupancy) and assessing their relationships with disease occurrence or creating an automatic workflow to update these figures daily.

Thanks, again, to Molly Tolzmann zmotoly for the data management.

Cumulative SARS-CoV-2 Cases in the District of Columbia by Health Planning Neighborhoods

Sat, 16 May 2020 00:00:00 +0000

After moving to DC last year, PoPville has been a personal favorite for local scoop. A post on May 11, 2020 captured my attention. Molly Tolzmann zmotoly adjusted the daily coronavirus data publicly released by the DC Mayoral Office at the DC health planning neighborhood level by the 2018 American Community Survey (ACS) census tract data and demographic data from Open Data DC. This post is a replication of the data visualization using R and can be found on a public GitHub repository.

Important Notes:

The following show cumulative case rate per 1,000 since the beginning of the COVID-19 outbreak in DC. Therefore, the the stats do not reflect the number of people currently infected with SARS-CoV-2.
The following does not account for the degree of COVID-19 testing in each neighborhood.

COVID-19 in DC by Molly Tolzmann

The following are the necessary packages and settings for the exercise.

# Packages
loadedPackages <- c("dplyr", "ggplot2", "googlesheets4", "leaflet", "sf", "stringr", "widgetframe")
invisible(lapply(loadedPackages, require, character.only = T))

# Settings
gs4_deauth() # no Google authorization necessary because we are not reading a public repo

We import the District of Columbia Health Planning Neighborhoods boundaries and Molly Tolzmann’s zmotoly collation of the cumulative cases from start of the SARS-CoV-2 outbreak. The latter is hosted on a public Google Sheet and is accessible in R using the googlesheets4 package. After cleaning up the column names of the disease data, we merge the two data sets together and spatially project the polygons contained in the data to EPSG:32618.

# District of Columbia Health Planning Neighborhoods
gis_path <- "https://opendata.arcgis.com/datasets/de63a68eb7674548ae0ac01867123f7e_13.geojson"
dc <- st_read(gis_path)

## Reading layer `DC_Health_Planning_Neighborhoods' from data source 
##   `https://opendata.arcgis.com/datasets/de63a68eb7674548ae0ac01867123f7e_13.geojson' 
##   using driver `GeoJSON'
## Simple feature collection with 51 features and 9 fields
## Geometry type: POLYGON
## Dimension:     XY
## Bounding box:  xmin: -77.11976 ymin: 38.79165 xmax: -76.9094 ymax: 38.99556
## Geodetic CRS:  WGS 84

# District of Columbia SAR-CoV-2 Data prepared by @zmotoly
covid_path <- "https://docs.google.com/spreadsheets/d/1u-FlJe2B1rYV0obEosHBks9utkU30-C2TSkHka6AVS8/edit#gid=1923705378"
covid <- read_sheet(
  ss = covid_path,
  sheet = 2, # second sheet
  skip = 2 # skip 1st row of annotation
)  
names(covid) <- sub("\n", "", names(covid))   # remove extra line in column names
names(covid) <- gsub(" ", "_", names(covid))  # replace spaces with underscore

# Merge
dc_covid <- left_join(dc, covid, by = join_by("CODE" == "NB_Code"))

# Spatial Projection
## UTM zone 18N (Washington, DC)
dc_covid_proj <- st_transform(dc_covid, crs = 32618)

Static Map

We can then plot the cumulative rate using various plotting techniques. Here, I demonstrate the ggplot2 package. We can customize our plot in various ways. Here I present cumulative cases per 1,000 (May 15, since start of the outbreak) choosing a non-alarmist color palette from ColorBrewer2.0.

## Plot of cumulative cases per 1,000
ggplot() +                                                    # initialize ggplot object
  geom_sf(                                                    # make a polygon
    data = dc_covid_proj,                                     # data frame
    aes(fill = cut_interval(Cases_per_1000_May_15, 10)),      # variable to use for filling
    colour = "white") +                                       # color of polygon borders
  scale_fill_brewer(
    "Cumulative cases per 1,000",         # title of colorkey 
    palette = "Purples",                  # fill with brewer colors 
    na.value = "grey67",                  # color for NA (The National Mall)
    direction = 1,                        # reverse colors in colorkey
    guide = guide_legend(reverse = T)) +  # reverse order of colokey
  ggtitle(
    "Cumulative SARS-CoV-2 cases per 1,000 in Washington, D.C. as of May 15, 2020"
  ) +                                     # add title
  theme(
    line = element_blank(),               # remove axis lines
    axis.text = element_blank(),          # remove tickmarks
    axis.title = element_blank(),         # remove axis labels
    panel.background = element_blank(),   # remove background gridlines
    text = element_text(size = 10)) +     # set font size
  coord_sf()                              # both axes the same scale

Interactive Map

In addition to a static map, the leaflet package provides capabilities to create an interactive map with customizable features such as basemaps and overlapping layers. First, we need to spatially project the data to EPSG:4326. We can also create custom popups when scrolling mouse over each neighborhood. Here, I use the same color palette as the static map and provide an example of the raw cumulative cases (May 15, 2020) in DC to demonstrate the layer overlapping feature of the leaflet package.

# Work with unprojected spatialpolygonsdataframe
## Project to WGS84 EPSG:4326
CoV_DC_WGS84 <- st_transform(dc_covid_proj, crs = 4326)

## Create Popups
dc_health <- str_to_title(CoV_DC_WGS84$Neighborhood_Name)
dc_health[c(11, 25, 41, 49)] <- c(
  "DC Medical Center", "GWU", "National Mall", "SW/Waterfront"
)
CoV_DC_WGS84$popup1 <- paste(
  dc_health, 
  ": ", 
  format(round(CoV_DC_WGS84$Total_cases_May_15, digits = 0), big.mark = ",", trim = T),
  " cumulative cases",
  sep = ""
)
CoV_DC_WGS84$popup2 <- paste(
  dc_health, 
  ": ",
  format(round(CoV_DC_WGS84$Cases_per_1000_May_15, digits = 0), big.mark = ",", trim = T),
  " cumulative cases per 1,000",
  sep = ""
)

## Set Palettes
pal_cum <- colorNumeric(
  palette = "Purples",
  domain = CoV_DC_WGS84$Total_cases_May_15,
  na.color = "#555555"
)
pal_rate <- colorNumeric(
  palette = "Purples",
  domain = CoV_DC_WGS84$Cases_per_1000_May_15,
  na.color = "#555555"
)

The following is code for the leaflet plot. I set the starting parameters including available basemaps and then add each layer of COVID-19 data as polygons. After specifying the legend for each layer I finish up with a mini map to show scale.

## Create leaflet plot
lf <- leaflet(CoV_DC_WGS84) %>%                        # initial data
  setView(lng = -77, lat = 38.9, zoom = 11) %>%                  # starting coordinates
  addTiles() %>% # basemap
  addPolygons(
    data = CoV_DC_WGS84, 
    color = "black", 
    weight = 1, 
    smoothFactor = 0.5, 
    opacity = 1,
    fillOpacity = 0.67, 
    fillColor = ~pal_cum(Total_cases_May_15), 
    popup = ~popup1,
    highlightOptions = highlightOptions(color = "white", weight = 2, bringToFront = TRUE),
    group = "Cumulative Cases"
  ) %>%
  addPolygons(
    data = CoV_DC_WGS84, 
    color = "black", 
    weight = 1, 
    smoothFactor = 0.5, 
    opacity = 1,
    fillOpacity = 0.67, 
    fillColor = ~pal_rate(Cases_per_1000_May_15), 
    popup = ~popup2,
    highlightOptions = highlightOptions(color = "white", weight = 2, bringToFront = TRUE),
    group = "Cumulative Rate"
  ) %>%
  addLayersControl(
    overlayGroups = c("Cumulative Cases", "Cumulative Rate"), # layer selection
    options = layersControlOptions(collapsed = FALSE)
  ) %>%     
  addLegend(
    "topright",
    pal = pal_cum,
    values = ~Total_cases_May_15,                  # legend for cases
    title = "Cumulative Cases", 
    opacity = 1, 
    group = "Cumulative Cases"
  ) %>%
  addLegend(
    "topright", 
    pal = pal_rate, 
    values = ~Cases_per_1000_May_15,              # legend for rate
    title = "Cumulative Rate per 1,000", 
    opacity = 1, 
    group = "Cumulative Rate"
  ) %>%
  hideGroup(c("Cumulative Cases", "Cumulative Rate")) %>% # no data shown (default)
  addMiniMap(position = "bottomleft") # add mini map

frameWidget(lf, width='100%')

As of May 15, 2020 the highest cumulative rate of SARS-CoV-2 cases have occurred in the Stadium Armory and Molly Tolzmann zmotoly noted the DC Jail is located in this neighborhood in her original post.

The provided maps are not intended to inform decision-making. Instead, I provide the the open-source code to download, manage, and visualize publicly available data. Future steps include linking other demographic information to each DC Health Planning Neighborhood (e.g., housing occupancy) and assessing their relationships with disease occurrence or creating an automatic workflow to update these figures daily.

Thanks, again, to Molly Tolzmann zmotoly and PoPville.

Areas of a spatial segregation model significantly different from null expectations

Sun, 10 May 2020 00:00:00 +0000

I present code to identify areas of a spatial segregation model that exceed our null expectations using the relrisk function in the spatstat package and an assumption of normality of the estimated probabilities.

A spatial segregation model was originally proposed by Diggle, Zheng, & Durr in 2005 which estimates spatially-varying probabilities of an event of a certain type to occur in an area accounting for other types. The original method uses a Monte Carlo-based simulation, which is computationally expensive. Instead, Adrian Baddeley and the spatstat team adapted the relrisk function for a multitype (m > 2) point pattern that has an option to compute the standard error of the probability estimates based on asymptotic theory, assuming a Poisson process.

Here, I use the standard errors to compute a 95% confidence interval (CI) at all gridded pixels (“knots”) for each type. Knots with a CI that does not capture the null expectation for each type are identified. I use the provided lansing dataset from the spatstat.data package. Created with assistance from Dr. Lance Waller and Barry Rowlingson.

# Packages
loadedPackages <- c("spatstat")
invisible(lapply(loadedPackages, require, character.only = T))

# Data
  lansing

## Marked planar point pattern: 2251 points
## Multitype, with levels = blackoak, hickory, maple, misc, redoak, whiteoak 
## window: rectangle = [0, 1] x [0, 1] units (one unit = 924 feet)

# Convert the lansing data to a ppp object
  ppp_lansing <- ppp(
    x = lansing$x, 
    y = lansing$y,
    window = unit.square(),
    marks = as.factor(marks(lansing))
  )

# Plot input
  plot.ppp(ppp_lansing, main = "Lansing Woods", cex = 0.5)

# Estimate nonparametric spatially-varying probabilities of type
  f1 <- relrisk.ppp(ppp_lansing, casecontrol = F, diggle = T, se = T, sigma = bw.diggle)
  
# Default plots
  plot(f1$estimate, main = "Probability of an event by type", zlim = c(0,1)) # probabilities

  plot(f1$SE, main = "Standard error of probability", zlim = c(0,0.03)) # standard errors

  wh <- im.apply(f1$estimate, which.max)
  types <- levels(marks(lansing))
  wh <- eval.im(types[wh]) # most common 
  plot.im(wh, main="Most common species", ribargs = list(las = 1))

# Significant p-values assuming normality of the Poisson process
## relrisk() computes standard errors based on asymptotic theory, assuming a Poisson process
  alpha <- 0.05                           # alpha
  z <- qnorm(alpha/2, lower.tail = F)     # z-statistic
  f1$u <- f1$estimate + z*f1$SE           # Upper CIs
  f1$l <- f1$estimate - z*f1$SE           # Lower CIs
  mu_0 <- as.vector(table(marks(ppp_lansing))/ppp_lansing$n) # null expectations by type
  f1$p <- f1$estimate # copy structure of pixels, replace values
  for (i in 1:length(f1$p)) {
    f1$p[[i]]$v <- factor(
      ifelse(
        mu_0[i] > f1$u[[i]]$v, "lower",
             ifelse(mu_0[i] < f1$l[[i]]$v, "higher", "none")
      ),
      levels = c("lower", "none", "higher"))
  }

  # Plot significant p-values
  plot(f1$p, main = "Significant difference from null?")

The final plot identifies where the spatially-varying probabilities exceed the null expected probability of each type. Areas of higher-than-expected probability could be considered “hot-spots” and areas of lower-than-expected probability could be considered “cold-spots.” Areas that are not significantly different from the null expectation could suggest additional sampling is necessary to determine if these areas are hot or cold spots.

2020 William G. Coleman Minority Health and Health Disparities Research Innovation Award

Thu, 30 Apr 2020 00:00:00 +0000

I will be taking over a project entitled “Evaluating the impact of concentrated animal feeding operations on Campylobacter jejuni infections in rural agricultural communities” previously led by Dr. Joseph Shearer as he is departing the National Cancer Institute.

The project is funded by a 2020 National Institute for Minority Health and Health Disparities William G. Coleman Jr., Ph.D., Minority Health and Health Disparities Research Innovation Award.

Interim Co-Editor-in-Chief of DFEB

Wed, 18 Dec 2019 00:00:00 +0000

I took over for Dr. Maeve Bailey-Whyte as Interim Co-Editor-in-Chief of the DCEG Fellows Editorial Board (DFEB) while she is on maternity leave until March 2020.

I join Dr. Sarah Jackson in this role. DFEB is a fellow-led group that offers fellows a free scientific document-reviewing service and the opportunity to build scientific editing and review skills of their own. The group provides confidential, anonymous feedback on grammar, organization, and flow of scientific manuscripts, abstracts, proposals, posters, and slide decks.

New Postdoctoral Mentor

Sat, 10 Aug 2019 00:00:00 +0000

I started working with Dr. Rena Jones in the Occupational and Environmental Epidemiology Branch within the Division of Cancer Epidemiology and Genetics at the National Cancer Institute part of the National Institutes of Health Intramural Research Program as my Cancer Prevention Fellowship Program preceptor.

New Postdoctoral Cancer Prevention Fellow

Mon, 10 Jun 2019 00:00:00 +0000

I started as a Postdoctoral Cancer Prevention Fellow in the Cancer Prevention Fellowship Program at the National Cancer Institute.