Toward a panoramic perspective of the association between environmental factors and cardiovascular disease: An environment-wide association study from National Health and Nutrition Examination Survey 1999–2014

Design and methodology challenges of environment-wide association studies: A systematic review

“…The total numbers of participants in the EWAS included in this systematic review ranged from 322 to 77,086, and the number of environmental factors included ranged from 8 to 14,663. Out of the 23 included studies, 15 stated that a hypothesis-generating or data-driven approach was used (Balazard et al, 2016;Hall et al, 2014;Lapidus et al, 2013;Lind et al, 2013;Lynch et al, 2017;Patel et al, 2010;Patel et al, 2018;Patel et al, 2012;Patel et al, 2017;Patel et al, 2013b;Patel et al, 2014;Wulaningsih et al, 2017;Zhong et al, 2016;Zhuang et al, 2018a;Zhuang et al, 2018b).…”

“…population density, percentage of college graduates, crime rates. The National Health and Nutrition Examination Survey (NHANES, https://www.cdc.gov/nchs/nhanes/ index.htm) was used as the main data source in 10 studies (McGinnis et al, 2016;Patel et al, 2010;Patel et al, 2012;Patel et al, 2015;Patel et al, 2017;Patel et al, 2013b;Patel et al, 2014;Wulaningsih et al, 2017;Zhuang et al, 2018a;Zhuang et al, 2018b), and 9 studies obtained data from other sources (Balazard et al, 2016;Chung et al, 2019;Hall et al, 2014;Lapidus et al, 2013;Lind et al, 2013;Lynch et al, 2017;Mooney et al, 2017;Patel et al, 2018;Zhong et al, 2016). The other 4 studies involved primary data collections based on interviews or questionnaires (Hovi et al, 2016;Jiménez-Cruz et al, 2013;Lenters et al, 2015;Zhou et al, 2013).…”

“…A variety of outcomes were examined in these studies, including childhood type 1 diabetes (Balazard et al, 2016), type 2 diabetes (Hall et al, 2014;Patel et al, 2010), metabolic syndrome (Lind et al, 2013), serum lipid levels (Patel et al, 2012), preterm birth (Patel et al, 2014), reproductive function (Chung et al, 2019;Lenters et al, 2015), hematocrit (Zhong et al, 2016), blood pressure (McGinnis et al, 2016), leukocyte telomere length (Patel et al, 2017), obesity (Jiménez-Cruz et al, 2013;Wulaningsih et al, 2017), physical activity (Mooney et al, 2017;Zhou et al, 2013), household income (Patel et al, 2015), prostate cancer (Lynch et al, 2017), peripheral arterial diseases (PADs) (Zhuang et al, 2018b), cardiovascular diseases (CVDs) (Zhuang et al, 2018a), respiratory or gastrointestinal tract infection (RTI or GTI) (Hovi et al, 2016), H1N1 virus (Lapidus et al, 2013), and human immunodeficiency viruses (HIV) (Patel et al, 2018). In 19 out of the 21 EWAS using top-down approaches, outcomes and environmental exposures were assessed using the same data source.…”

“…Nineteen of the 23 EWAS controlled for multiple testing using a strict Bonferroni correction or the Benjamini-Hochberg procedure. Eleven studies included internal validation by dividing the data into training and testing sets using different approaches, such as random split (Patel et al, 2018;Zhuang et al, 2018a;Zhuang et al, 2018b), splitting by calendar year (McGinnis et al, 2016;Patel et al, 2010;Patel et al, 2012;Patel et al, 2015;Patel et al, 2013b;Wulaningsih et al, 2017), and cross-validation (Lind et al, 2013;Zhong et al, 2016).…”

“…In addition, different approaches for variable selection or importance ranking were employed. For example, among the 21 top-down EWAS, one study used a random forest model to identify important variables with potential interaction effects (Zhuang et al, 2018a), and another study conducted a random effects meta-analysis by combining results from both training and testing datasets (McGinnis et al, 2016). The 2 bottom-up EWAS studies explored additional and more machine learning oriented methods: Mooney et al (2017) performed multivariable regression using the least absolute shrinkage and selection operator (LASSO) as well as the random forest, and Lynch et al 2017…”