Research

Causal Inference Methods for Investigating the Health Impacts of Contaminant Exposures.

Much of my recent research has been focused on the development of and application of statistical methods for causal inference to study the health impacts of contaminant exposures, including air pollution, natural gas exposures, and point-sources of contamination. One paper from this work addresses the causal inference assumption of overlap in confounder distributions across treatment groups. Evidence in the literature suggests that this assumption is commonly unmet by data in many applications, including environmental health, due to strong confounding. In our work (Nethery et al., 2019a), we introduce a novel method for diagnosing violations of this assumption and a machine-learning based method that can provide less biased population average causal effects in this setting. We apply this method to perform the first-ever investigation of the effects of exposure to natural gas compressor stations on cancer outcomes. In collaboration with EPA scientists, I have also developed a causal machine learning method that leverages nationwide Medicare health records data to estimate the number of adverse health events prevented due to large-scale air quality regulations like the 1990 Clean Air Act Amendments (Nethery et al., 2020b). In another recent paper (Nethery et al., 2020a), we propose a causal inference approach to epidemiological cancer cluster investigations. The methods currently used in these investigations induce enormous inflation of Type-1 error rates due to poorly-defined statistical hypotheses and fail to adjust for most sources of confounding. Our approach, which relies entirely on publicly available data, resolves these issues by framing the statistical hypothesis around a point source environmental exposure (e.g. did exposure to this point source cause increased cancer incidence in the exposed population?) and applying causal matching techniques for confounding adjustment.

Related Publications:

  1. Nethery R.C., Mealli F., Dominici F. (2019a). Estimating Population Average Causal Effects in the Presence of Non-Overlap: The Effect of Natural Gas Compressor Station Exposure on Cancer Mortality. Annals of Applied Statistics 13(2), 1242-1267. PMID: 31346355, PMCID: PMC6658123. [CODE]
  2. Nethery R.C., Dominici F. (2019b) Estimating pollution-attributable mortality at the regional and global scales: challenges in uncertainty estimation and causal inference. European Heart Journal 40(20), 1597-1599.
  3. Nethery R.C., Yang Y., Brown A., Dominici F. (2020a). A Causal Inference Framework for Cancer Cluster Investigations Using Publicly Available Data. In press, Journal of the Royal Statistical Society, Series A. arXiv preprint number: 1811.05997. [CODE]
  4. Nethery R.C., Mealli F., Sacks J.D., Dominici F. (2020b) Evaluation of the Health Impacts of the 1990 Clean Air Act Amendments Using Causal Inference and Machine Learning. Journal of the American Statistical Association, DOI: 10.1080/01621459.2020.1803883 [CODE]
  5. Yitshak-Sade, M., Nethery, R.C., Abu Awad, Y., Mealli, F., Dominici, F., Kloog, I., Zanobetti, A. (2020a). Lowering Air Pollution Levels in Massachusetts May Prevent Cardiovascular Hospital Admissions. Journal of the American College of Cardiology, 75(20), 2642-2644.
  6. Yitshak-Sade, M., Nethery, R.C., Schwartz, J. D., Mealli, F., Dominici, F., Di, Q., Abu Awad, Y., Ifergane G., Zanobetti, A. (2020b). PM2.5 and Hospital Admissions Among Medicare Enrollees with Chronic Debilitating Brain Disorders. Science of The Total Environment, 142524.

 

Statistical and Machine Learning Methods for Climate and Disaster Epidemiology.

To minimize the health threats presented by extreme weather and climate events, we must generate high-precision insights and tools to better inform strategic preparedness efforts. I am leading several ongoing projects to develop statistical and machine learning methods for this purpose. In one recent paper (Nethery et al., 2020), we propose an integrated causal and predictive machine learning modeling approach that enables standardized, high-resolution quantification of the health impacts of recent historic tropical cyclones and characterizes how features of the cyclone and the impacted communities explain variation in cyclone-related health risks. The resulting model not only provides unprecedented insights into tropical cyclone epidemiology but can also aid in identifying high-risk communities in advance of future cyclones. Additionally, my team is working on causal inference approaches to investigate potential adverse interactions between medications, heat, and air pollution exposures using Medicare claims data. I also collaborate with investigators leading the Botswana Birth Outcomes Surveillance Study to assess how prenatal heat and air pollution exposures impact birth outcomes in developing countries.

Related Publications:

  1. Nethery, R. C., Katz-Christy, N., Kioumourtzoglou, M. A., Parks, R. M., Schumacher, A., & Anderson, G. B. (2020). Integrated Causal-Predictive Machine Learning Models for Tropical Cyclone Epidemiology. In preprint: https://arxiv.org/abs/2010.11330 [CODE]

 

Data Reduction with Complex Spatial and Temporal Correlation Structures.

I am interested in the development and applications of latent variable models, which are statistical models often used for reduction/summarization of large, noisy datasets. Specifically, I work on spatio-temporal latent variable models that provide reliable inference when applied to data with certain complex spatial and temporal structures. We developed extensions to spatial factor analysis, a latent variable model, that address common complications in geo-spatial data. One such extension (Nethery et al., 2015) strengthens the ability of this method to predict values of the extracted features of interest for unobserved spatial locations. In another paper (Nethery et al., 2018), we present a novel variation of spatial factor analysis that allows a small number of features of interest to be extracted from spatially misaligned data. We applied this method to spatially misaligned social data to create a measure that identifies communities in the state of Louisiana that are at the highest risk during natural disasters. I received a student paper award from the American Statistical Association for this work.

Related Publications:

  1. Nethery R.C., Warren J.L., Herring A.H., Moore K.A.B., Evenson K.R., Diez-Roux A.V. (2015). A Common Spatial Factor Analysis Model for Measured Neighborhood-Level Characteristics: The Multi-Ethnic Study of Atherosclerosis. Health & Place 36, 35-46. PMID: 26372887, PMCID: PMC4679666.
  2. Nethery R.C., Sandler D.P., Zhao S., Engel L.S., Kwok R.K. (2018). A Joint Spatial Factor Analysis Model to Accommodate Data from Misaligned Areal Units with Application to Louisiana Social Vulnerability. Biostatistics 20(3), 468-484. PMID: 29659722, PMCID: PMC6659171. [CODE]

 

Statistical Challenges in Studies of Health Inequities

In collaboration with experts at Harvard, Emory, and Imperial College London, I have recently published several papers addressing statistical challenges that commonly arise in studies of health inequities-- including systematic errors and uncertainties in group-specific population counts and the problematic data features brought about by residential segregation. One paper on this topic (Nethery et al., 2021a) evaluated and compared numerous alternative sources of small-area population counts for real-time monitoring of inequities in disease incidence. Another paper investigated how residential segregation gives rise to data features that violate assumptions in standard disease mapping models used in health inequities research, which can lead to misleading characterizations of inequities (Nethery et al., 2021b). Our team is also actively investigating the impacts of new US Census privacy-preserving algorithms on health inequity studies (Krieger et al., 2021).

  1. Nethery R.C., Rusovich T., Peterson E., Chen J., Waterman P., Krieger N., Waller L., Coull B. (2021a). Comparing Denominator Sources for Real-Time Disease Incidence Modeling: American Community Survey and WorldPop. Social Science and Medicine- Population Health 14: 100786. [CODE]
  2. Nethery R.C., Chen J.T., Krieger N., Waterman P.D., Peterson E., Waller L.A., Coull B.A. (2021b). Statistical Implications of Endogeneity Induced by Residential Segregation in Small-Area Modelling of Health Inequities. In press, The American Statistician. [PAPER]
  3. Krieger N., Nethery R.C., Chen J.T., Waterman P.D., Wright E., Rushovich T, Coull B. (2021). Impact of Differential Privacy and Small Numbers on the Monitoring of Health Inequities Using US Census Tract Sources. American Journal of Public Health 111 (2): 265-268. [CODE]

 

Applied/Collaborative Environmental Health Studies.

Maternal gestational phenotypes have long been known to be associated with numerous pregnancy complications and childhood disorders; however, much remains unknown due to the complicated nature of the biological processes at play during pregnancy. I have collaborated with physicians and epidemiologists to study the complex relationships between maternal gestational conditions/exposures and childhood health outcomes. In an ongoing series of studies with Dr. Stephanie Engel’s lab at UNC Chapel Hill, we are investigating the relationships between material gestational phthalate exposure and thyroid hormone levels, and childhood neurodevelopmental outcomes. In one paper, we discovered an association between maternal gestational phthalate exposure and child’s risk of ADHD (Engel et al., 2018). Digging deeper into the potential biological mechanisms behind this association, we also studied how phthalate exposures were associated with maternal thyroid hormone levels (Villanger et al., 2020), which could in turn impact the developing fetus. I have also been a key contributor to research that detected novel associations between several maternal lipoprotein particle concentrations and risk of pre-term birth (Grace et al., 2018). The findings of these works could lead to new environmental and dietary guidelines to minimize pregnancy risks. In another vein of research, I have collaborated with cell biologists investigating the impact of flavored cigars and e-cigarettes on lung health. This work contributed novel evidence that the use of certain e-cigarette liquid flavors might lead to much worse health consequences than others, an important finding in a highly understudied research area (Rowell et al., 2017). For all of the studies cited here, I was the primary statistician, responsible for developing the statistical analysis plan and conducting all analyses.

Related Publications:

  1. Engel S.M., Villanger G.D., Nethery R.C., Thomsen C., Sakhi A.K., Drover S.S., Hoppin J.A., Zeiner P., Knudsen G.P., Reichborn-Kjennerud T., Herring A.H. (2018). Prenatal Phthalates, Maternal Thyroid Function, and Risk of Attention-Deficit Hyperactivity Disorder in the Norwegian Mother and Child Cohort. Environmental Health Perspectives, 126(5). PMID: 29790729, PMCID: PMC6071976.
  2. Villanger G.D., Drover S.S.M., Nethery R.C., Thomsen C., Sakhi A.K., Overgaard K.R., Zeiner P., Hoppin J.A., Reichborn-Kjennerud T., Aase H., Engel S.M. (2020). Associations between Urine Phthalate Metabolites and Thyroid Function in Pregnant Women and the Influence of Iodine Status. In press, Environment International. PMCID: In progress.
  3. Grace M.R., Vladutiu C.J., Nethery R.C., Siega-Riz A.M., Manuck T.A., Herring A.H., Savitz D., Thorp J.T. (2018). Lipoprotein Particle Concentration Measured by Nuclear Magnetic Resonance Spectroscopy and Preterm Birth: A Prospective Cohort Study. BJOG: An International Journal of Obstetrics and Gynaecology 125, 895–903. PMID: 28886230, PMCID: PMC6582364.
  4. Rowell T.R., Reeber S.L., Lee S.L., Harris R.A., Nethery R.C., Herring A.H., Glish G.L., Tarran R. (2017). E-Cigarette Liquids Reduce Proliferation and Viability in the CALU3 Airway Epithelial Cell Line. American Journal of Physiology- Lung Cellular and Molecular Physiology 313(1), 52-66. PMID: 28428175, PMCID: PMC5538872.