Socioeconomic Correlates of Air Pollution and Heart Disease

Abstract

This chapter begins an investigation that occupies the rest of the book: examining how the causal analysis methods discussed in Part 2, along with the methodological points about interpretation of regression coefficients raised in earlier chapters, especially Chaps. 1, 2, 7, and 8, can be applied to the important public health topic of health risks caused by exposure to fine particulate matter (PM2.5) air pollution. The specific concept of causation of greatest interest is manipulative or interventional causation (Chap. 9), addressing how changing air pollution would change death and disease rates. This contrasts with other concepts of causation that seek to attribute some number of “premature” deaths or diseases per year to air pollution without addressing the extent, if any, to which these attributed deaths would be prevented by reducing pollution levels (see Chaps. 9 and 18).

Appendix: Guide to Replication

For investigators who wish to replicate or extend the analyses presented in this chapter, the following instructions provide the required details.

• 1. The full BRFSS data set with PM2.5 data appended, for which Table 13.1 shows the first few records, can be downloaded from the following site: http://cox-associates.com/CausalAnalytics/BRFSS.csv

• 2. The data can be prepared for analysis via the following R script

  df<- read.csv("BRFSS.csv",stringsAsFactors=FALSE)

• 3. Linear and logistic regression models without income can be fit as follows:

  LinR <- lm(df$HeartDiseaseEver ~ df$Sex + df$Smoking + df$Age + df$PM2.5); summary(LinR)

  LogR <- glm(df$HeartDiseaseEver ~ df$Sex + df$Smoking + df$Age + df$PM2.5, family=binomial(link=’logit’)); exp(cbind(OR = coef(LR), confint(LR))); summary(LogR)

• 4. Linear and logistic regression models with income can be fit as follows:

  LinR <- lm(df$HeartDiseaseEver ~ df$Sex + df$Smoking + df$Age + df$PM2.5 + df$Income); summary(LinR)

  LogR<- glm(df$HeartDiseaseEver ~ df$Sex + df$Smoking + df$Age + df$PM2.5 + df$Income, family=binomial(link=’logit’)); exp(cbind(OR = coef(LR), confint(LR))); summary(LogR)

• 5. For more computationally intensive analyses, a random sample from the full data set (small enough so that the randomForest algorithm can be used) can be created using the following commands:

  set.seed(1)

  brfss<- df[sample(1:nrow(df), 5000, replace = FALSE),]

  write.csv(brfss, “brfss2.csv”)

• 6. Open the CAT software at http://cloudcat.cox-associates.com:8899/ (registering for free first if needed). Select the Data tab from the menu of commands, then browse to and upload the file “brfss2.csv”. From the data tab, select the variables HeartDiseaseEver, Income, Age, Sex, Smoking, PM2.5, Education, MaritalStatus. To reproduce Fig. 13.9, select Bayes and specify Age and Sex and sources and HeartDiseaseEver as a sink. To reproduce Fig. 13.10, select Income as the cause and HeartDiseaseEver as the target effect in the interactive Bayesian network. Select direct or total to generate corresponding causal PDPs.