## Estimate county-level Arsenic levels, directly-standardized lung cancer incidence and smoking rates for each county in the study area. 
## Estimate association and interaction between exposure, smoking, median income, and lung cancer incidence.

## As per PONE-D-11-06364
## Association of Arsenic Exposure with Lung Cancer Incidence Rates in the United States PLoS ONE
## Putila J and Guo NL
## MBR Cancer Center and WVU Department of Community Medicine
## Room 1814
## Morgantown WV, 26506

rm(list=ls())
gc()

## Import Population Data
## County-level estimates only from the 2000 U.S. Census
CApop <- read.csv("C:/Users/jputila/Desktop/Heavy Metal Data/Census/CA/2kh06_mod.csv")
CTpop <- read.csv("C:/Users/jputila/Desktop/Heavy Metal Data/Census/CT/2kh09_mod.csv")
IApop <- read.csv("C:/Users/jputila/Desktop/Heavy Metal Data/Census/IA/2kh19_mod.csv")
KYpop <- read.csv("C:/Users/jputila/Desktop/Heavy Metal Data/Census/KY/2kh21_mod.csv")
LApop <- read.csv("C:/Users/jputila/Desktop/Heavy Metal Data/Census/LA/2kh22_mod.csv")
NJpop <- read.csv("C:/Users/jputila/Desktop/Heavy Metal Data/Census/NJ/2kh34_mod.csv")
NMpop <- read.csv("C:/Users/jputila/Desktop/Heavy Metal Data/Census/NM/2kh35_mod.csv")
UTpop <- read.csv("C:/Users/jputila/Desktop/Heavy Metal Data/Census/UT/2kh49_mod.csv")

## Remove extraneous character from county name
IApop$Area <- ifelse(IApop$Area=="O'Brien County, Iowa", "OBrien County, Iowa", as.character(factor(IApop$Area)))
## Combine individual population files
Allpop <- rbind(CApop, CTpop, IApop, KYpop, LApop, NJpop, NMpop, UTpop)

## Total number of counties thus far
NUMCOUNT <- length(Allpop$Area)[1]
NUMCOUNT
## Should be 432

## Shortened name for the county
CountyShort <- as.character(substr(Allpop$Area, 1,8))
## Move county name to end of data frame
Area <- Allpop$Area
Allpop <- Allpop[,-1]
Allpop <- data.frame(Allpop, CountyShort, Area)

## Read population data for the U.S.
US <- read.csv("C:/Users/jputila/Desktop/Heavy Metal Data/Census/US/2kh00_mod.csv")
## Remove row name, total, males, females columns
US <- US[,-c(1:4)]

## Import Incidence Data from SEER
## IA,KY,LA,NJ,NM,UT,CA,CT
Allinc <- read.csv("C:/Users/jputila/Desktop/Heavy Metal Data/SEER/all_dca.txt")
## Only 2000+ CA entries
Allinc <- Allinc[substr(Allinc$StateCounty,1,2)!="CA" | Allinc$YearDx>=2000,]

## Missouri, MO - age adjusted
MOinc <- read.csv("C:/Users/jputila/Desktop/Heavy Metal Data/Other Inc/MO/mica_mod.csv")
MOinc$County <- paste(as.character(factor(MOinc$County)), ", Missouri", sep="")
MOinc$FIPSCode <- ifelse(as.numeric(MOinc$FIPSCode) < 10 , paste(0, MOinc$FIPSCode, sep=""), MOinc$FIPSCode)
MOinc$FIPSCode <- ifelse(as.numeric(MOinc$FIPSCode) < 100, paste(0, MOinc$FIPSCode, sep=""), MOinc$FIPSCode)
MOinc$FIPS <- paste("29", MOinc$FIPSCode, sep="")
MOpop <- read.csv("C:/Users/jputila/Desktop/Heavy Metal Data/Other Inc/MO/2kh29_mod.csv", header=TRUE)

## Ohio, OH - age adjusted
OHinc <- read.csv("C:/Users/jputila/Desktop/Heavy Metal Data/Other Inc/OH/OHinc.csv")
OHinc$County <- paste(as.character(factor(OHinc$County)), " County, Ohio", sep="")
OHinc$FIPSCode <- ifelse(as.numeric(OHinc$FIPSCode) < 10 , paste(0, OHinc$FIPSCode, sep=""), OHinc$FIPSCode)
OHinc$FIPSCode <- ifelse(as.numeric(OHinc$FIPSCode) < 100, paste(0, OHinc$FIPSCode, sep=""), OHinc$FIPSCode)
OHinc$FIPS <- paste("39", OHinc$FIPSCode, sep="")
OHpop <- read.csv("C:/Users/jputila/Desktop/Heavy Metal Data/Other Inc/OH/2kh39_mod.csv", header=TRUE)

## West Virginia, WV - age adjusted
WVinc <- read.table("C:/Users/jputila/Desktop/Heavy Metal Data/Other Inc/WV/wv_inc.txt", sep=" ", header=TRUE)
WVfips <- read.table("C:/Users/jputila/Desktop/Heavy Metal Data/Other Inc/WV/wv_fips.txt", sep="\t", header=TRUE)
WVpop <- read.csv("C:/Users/jputila/Desktop/Heavy Metal Data/Other Inc/WV/2kh54_mod.csv", header=TRUE)
WVinc <- data.frame(WVfips[,2], WVfips[,1], WVinc[,4])
colnames(WVinc) <- c("County", "FIPS", "AdjRate")
WVinc$County <- paste(WVinc$County, ", West Virginia", sep="")

## Pennsylvania, PA - age adjusted
PAinc <- read.table("C:/Users/jputila/Desktop/Heavy Metal Data/Other Inc/PA/pa_inc.txt", sep="\t", header=TRUE)
PAinc$County <- paste(PAinc$County, "County, Pennsylvania", sep="")
PApop <- read.csv("C:/Users/jputila/Desktop/Heavy Metal Data/Other Inc/PA/2kh42_mod.csv")
PAfips <- read.table("C:/Users/jputila/Desktop/Heavy Metal Data/Other Inc/PA/pa_fips.txt", sep="\t", header=TRUE)
PAinc <- data.frame(PAinc[,1], PAfips[,1], PAinc[,10])

#Fix extraneous issues in SEER naming of StateCounty
Allinc$StateCounty <- ifelse(Allinc$StateCounty=="CA: Los Angeles Registry (06037)" , "CA: Los Angeles County (06037)", as.character(factor(Allinc$StateCounty)))
Allinc$StateCounty <- ifelse(Allinc$StateCounty=="NM: Cibola County (35006) - 1982+", "NM: Cibola County (35006)", as.character(factor(Allinc$StateCounty)))
Allinc$StateCounty <- ifelse(Allinc$StateCounty=="NM: Valencia County (35061) - 1982+", "NM: Valencia County (35061)", as.character(factor(Allinc$StateCounty)))

## Create a FIPS code for the SEER data
FIPS <- substr(Allinc$StateCounty, nchar(as.character(Allinc$StateCounty))-5, nchar(as.character(Allinc$StateCounty))-1)
CountyShort <- substr(Allinc$StateCounty, 5, 12)

## Convert state abbreviation to full name
StateT <- substr(Allinc$StateCounty, 1, 2)
State <- ifelse(StateT=="CA", ", California",
			ifelse(StateT=="CT", ", Connecticut",
			ifelse(StateT=="LA", ", Louisiana",
			ifelse(StateT=="IA", ", Iowa",
			ifelse(StateT=="KY", ", Kentucky",
			ifelse(StateT=="NJ", ", New Jersey",
			ifelse(StateT=="UT", ", Utah",
			ifelse(StateT=="NM", ", New Mexico",
			"ERROR"
			))))))))
Area <- paste(substr(Allinc$StateCounty, 5, nchar(as.character(Allinc$StateCounty))-8), State, sep="")
Allinc <- data.frame(Allinc, CountyShort, FIPS, Area)
Allinc <- Allinc[Allinc$CountyShort!="Unknown ",]
rm(StateT)


########################
## Population Weights       ##
########################

## Combined Population for all 12 states.
Pop.1 <- rbind(CApop, CTpop, IApop, KYpop, LApop, NJpop, NMpop, UTpop, MOpop, OHpop, WVpop, PApop)
rm(CApop, CTpop, IApop, KYpop, LApop, NJpop, NMpop, UTpop, OHpop, MOpop, WVpop, PApop)

## Sum the population of each age group to get the total population
Pop <- matrix(NA, nrow=dim(Pop.1)[1], ncol=2)
for(i in 1:dim(Pop.1)[1]){
	Pop[i,1] <- as.character(factor(Pop.1[i,1]))
	Pop[i,2] <- sum(Pop.1[i,2:14])
}
colnames(Pop) <- c("PopCounty", "Population")

#############################
## Calculate Raw Incidence         ##
#############################
## Lung Cancer diagnosed between 1996 and 2005
## Obtained from the NCI-SEER database.

## Age Groups Pairs
agroups <- c(0,4,5,9,10,14,15,19,20,24,25,34,35,44,45,54,55,59,60,64,65,74,75,84,85,130)

crate <- matrix(rep(NA,NUMCOUNT*13), NUMCOUNT, 13)
County <- matrix(rep(NA, NUMCOUNT), NUMCOUNT, 1)
FIPS <- matrix(rep(NA, NUMCOUNT), NUMCOUNT, 1)

## Years of data available
## Either 6 or 10 depending upon when the state entered SEER
yearsdata <- NA
for(i in 1:NUMCOUNT){
yearsdata[i] <- ifelse(unique(substr(Allinc[as.character(Allinc$Area)==as.character(Allpop[i,]$Area),]$StateCounty,1,2)) %in% c("CT", "IA", "NM", "UT"), 10, 
					 ifelse(unique(substr(Allinc[as.character(Allinc$Area)==as.character(Allpop[i,]$Area),]$StateCounty,1,2)) %in% c("CA", "LA", "KY", "NJ"), 6, NA))
}

## SEER Areas
## Calculates the crude rate (crate) by taking the total number of cases and dividing that by the number of years (6 or 10)
## then multiplies by 100,000 to get the average incidence rate per 100,000 persons in each of the j=1:13 age groups
for(i in 1:NUMCOUNT){
	crateRow <- NA
	for(j in 1:13){
	crate[i,j] <- ((dim(Allinc[as.character(Allinc$Area)==as.character(Allpop[i,]$Area)
				& (Allinc$YearDx <= 2005 & Allinc$YearDx >= 1996)
				& (Allinc$AgeDx >= agroups[j+(j-1)] & Allinc$AgeDx <= agroups[j+j] ),
				])[1] / yearsdata[i]) / Allpop[i,j])*100000
	}
	County[i] <- as.character(Allpop[i,15])
	FIPS[i] <- as.character(unique(factor(Allinc[as.character(Allinc$Area)==as.character(factor(Allpop[i,]$Area)),]$FIPS)))
}

## Direct Standardization to the 2000 U.S. Population
Allstd <- matrix(NA, nrow=NUMCOUNT, ncol=13)
AdjRate <- NA

## Directly standardizes the rate to the 2000 U.S. population, using the US census data
for(i in 1:NUMCOUNT){
	for(j in 1:13){
		Allstd[i,j] <- (crate[i,j] * US[1,j])
	}
	AdjRate[i] <- (sum(Allstd[i,]) / sum(US[1,]))
}

## Combine the calculated rates with the published rates
temp <- data.frame(County, FIPS, AdjRate)
colnames(temp) <- c ("County", "FIPS", "AdjRate")
colnames(MOinc) <- c("County", "AdjRate", "Fipsshort", "FIPS")
colnames(OHinc) <- c("County", "AdjRate", "Fipsshort", "FIPS")
colnames(WVinc) <- c("County", "FIPS", "AdjRate")
colnames(PAinc) <- c("County", "FIPS", "AdjRate")


## Combine SEER and independent incidence
full <- rbind(temp, MOinc[,c(1,4,2)], OHinc[,c(1,4,2)], as.matrix(WVinc), as.matrix(PAinc))
## Add SES data
full <- cbind(full, Pop)
rm(temp)
## Update the number of counties, now 757
NUMCOUNT <- dim(full)[1]
NUMCOUNT
## Remove Sets
rm(MOinc, OHinc, WVinc, PAinc)


######################
## Read in SES data ##
######################

## Columns to read from each set
wide <- c(2 , 4, 127, 6, 53, 44, 23)
colname <- c("SFIPS", "CFIPS", "Junk1", "MedIncome", "Junk2", "SESName", "Junk3")

## Data sets
CAses <- read.fwf("C:/Users/jputila/Desktop/Heavy Metal Data/Census/CA/est00_CA.dat", widths = wide, col.names = colname)
CAses <- CAses[2:59,c(1,2,4,6)]

CTses <- read.fwf("C:/Users/jputila/Desktop/Heavy Metal Data/Census/CT/est00_CT.dat", widths = wide, col.names = colname)
CTses <- CTses[2:9,c(1,2,4,6)]

IAses <- read.fwf("C:/Users/jputila/Desktop/Heavy Metal Data/Census/IA/est00_IA.dat", widths = wide, col.names = colname)
IAses <- IAses[2:100,c(1,2,4,6)]

KYses <- read.fwf("C:/Users/jputila/Desktop/Heavy Metal Data/Census/KY/est00_KY.dat", widths = wide, col.names = colname)
KYses <- KYses[2:121,c(1,2,4,6)]

LAses <- read.fwf("C:/Users/jputila/Desktop/Heavy Metal Data/Census/LA/est00_LA.dat", widths = wide, col.names = colname)
LAses <- LAses[2:65,c(1,2,4,6)]

NJses <- read.fwf("C:/Users/jputila/Desktop/Heavy Metal Data/Census/NJ/est00_NJ.dat", widths = wide, col.names = colname)
NJses <- NJses[2:22,c(1,2,4,6)]

NMses <- read.fwf("C:/Users/jputila/Desktop/Heavy Metal Data/Census/NM/est00_NM.dat", widths = wide, col.names = colname)
NMses <- NMses[2:34,c(1,2,4,6)]

UTses <- read.fwf("C:/Users/jputila/Desktop/Heavy Metal Data/Census/UT/est00_UT.dat", widths = wide, col.names = colname)
UTses <- UTses[2:30,c(1,2,4,6)]

OHses <- read.fwf("C:/Users/jputila/Desktop/Heavy Metal Data/Other Inc/OH/est00_OH.dat", widths = wide, col.names = colname)
OHses <- OHses[2:89,c(1,2,4,6)]

MOses <- read.table("C:/Users/jputila/Desktop/Heavy Metal Data/Other Inc/MO/est00_MO.dat", sep="\t")
MOses <- MOses[,c(1,2,21,24)]
colnames(MOses) <- c("SFIPS", "CFIPS", "MedIncome", "SESName")

WVses <- read.fwf("C:/Users/jputila/Desktop/Heavy Metal Data/Other Inc/WV/est00_WV.dat", widths = wide, col.names = colname)
WVses <- WVses[2:56, c(1,2,4,6)]

PAses <- read.fwf("C:/Users/jputila/Desktop/Heavy Metal Data/Other Inc/PA/est00_PA.dat", widths = wide, col.names = colname)
PAses <- PAses[2:68, c(1,2,4,6)]

SES <- rbind(CAses, CTses, IAses, KYses, LAses, NJses, NMses, UTses, MOses, OHses, WVses, PAses)

## Add leading zeroes to the State/County FIPS Code
SES$CFIPS <- ifelse(as.numeric(SES$CFIPS) < 10, paste(0, SES$CFIPS, sep=""), SES$CFIPS)
SES$CFIPS <- ifelse(as.numeric(SES$CFIPS) < 100, paste(0, SES$CFIPS, sep=""), SES$CFIPS)

full <- cbind(full, SES)

#####################
## Read in BRFSS data ##
#####################
wide <- c(2, 95, 1, 1, 1, 2, 29, 3, 1158)  #max 1293

colname <- c("State", "Junk1", "Smoke100", "SmokeDay", "Quit", "Age", "Junk3", "County", "Junk4") 

## Read in the data
brfss.pre <- read.fwf("C:/Users/jputila/Desktop/Heavy Metal Data/BRFSS/Year08/CDBRFS08.ASC", widths = wide, col.names = colname)

## Remove extraneous variables and observations
brfss <- brfss.pre[,c(1,3,4,5,6,8)]
#rm(brfss.pre)
brfss$State <- ifelse(brfss$State < 10, paste("0", brfss$State, sep=""), brfss$State)
brfss <- brfss[(brfss$State=="06" | 
				brfss$State=="09" |
			    brfss$State=="19" | 
			    brfss$State=="21" | 
			    brfss$State=="22" |
				brfss$State=="29" |
			    brfss$State=="34" | 
			    brfss$State=="35" |
				brfss$State=="39" |
			    brfss$State=="49" |
				brfss$State=="54" |
				brfss$State=="42") &
				!is.na(brfss$State),]

## Only adults
brfss <- brfss[brfss$Age>=18,]

## Add leading zeroes to the County FIPS Code
brfss$County <- ifelse(as.numeric(brfss$County) < 10 , paste(0, brfss$County, sep=""), brfss$County)
brfss$County <- ifelse(as.numeric(brfss$County) < 100, paste(0, brfss$County, sep=""), brfss$County)

## Get rid entries with county suppressed due to small sample
brfss <- brfss[!is.na(brfss$County),]

## Code smoking status
## Smoked 100 cigarettes in lifetime, currently smokes daily, or currently smokes daily/some days
smk100 <- ifelse(brfss$Smoke100==1, 1, ifelse(brfss$Smoke100==2,0, NA))
smkDay <- ifelse(brfss$SmokeDay==1, 1, ifelse(smk100==0 | brfss$SmokeDay==2 | brfss$SmokeDay==3, 0, NA))
smkAny <- ifelse(brfss$SmokeDay==1 | brfss$SmokeDay==2, 1, ifelse(brfss$SmokeDay==3 | smk100==0, 0, NA))
scfips <- paste(brfss$State, brfss$County, sep="")
rateDay <- matrix(rep(NA, NUMCOUNT), NUMCOUNT)
rateAny <- matrix(rep(NA, NUMCOUNT), NUMCOUNT)
rate100 <- matrix(rep(NA, NUMCOUNT), NUMCOUNT)
for(i in 1:NUMCOUNT){ 
	rate100[i] <- (dim(brfss[scfips==as.character(factor(full[i,2])) & smk100==1 & !is.na(smk100),])[1] / dim(brfss[scfips==as.character(factor(full[i,2])),])[1])
}

full <- cbind(full, rate100)

###############################
## Calculate Metal Concentrations ##
###############################

## Import Metals data
metal <- read.csv("C:/Users/jputila/Desktop/Heavy Metal Data/geochem-2009-10-22-14-17-04.txt")
metal$FIPS <- substr(metal$FIPS, 2, 6)

## Get rid of invalid pH values
metal$PH <- ifelse(metal$PH == 88, NA, metal$PH)

## County FIPS Identifier to confirm accuracy of matching scheme
MetalName <- matrix(NA, nrow = NUMCOUNT, ncol = 1) 
for(i in 1:NUMCOUNT){ MetalName[i] <- as.character(factor(full[i,]$County))}

## Arsenic
## Calculate As values
## Do not include "-20" codes in estimates
Ascounty <- matrix(NA, nrow = NUMCOUNT, ncol = 1) 
for(i in 1:NUMCOUNT){
	Ascounty[i] <- mean(c(na.omit(metal[metal[,42] == as.character(factor(full[i,]$FIPS)) & metal$AS_ICP40 >= 0,]$AS_ICP40) , 
					na.omit(metal[metal[,42] == as.character(factor(full[i,]$FIPS)) & metal$AS_AA >= 0 ,]$AS_AA),
					na.omit(metal[metal[,42] == as.character(factor(full[i,]$FIPS)) & metal$AS_INAA >= 0 ,]$AS_INAA)))
}
Ascounty <- ifelse(Ascounty==0, NA, Ascounty)

## Final data set
full <- cbind(full, MetalName, Ascounty)

## Remove metal Dsets
#rm(MetalName, Ascounty)


#####################
## Statistical Analyses ##
#####################

## GGPlot package
library(ggplot2)

## Confidence intervals 
cconf <- function(model){
	numcov <- dim(coef(summary(model)))[1]
	uci <- NA
	lci <- NA
	for(i in 1:numcov){
	lci[i] <- exp(coef(summary(model))[i,1] - 1.96*coef(summary(model))[i,2]) 
	uci[i] <- exp(coef(summary(model))[i,1] + 1.96*coef(summary(model))[i,2])
	}
	outT <- data.frame(c(dimnames(coef(summary(model)))[1]), lci, exp(coef(summary(model))[,1]), uci)
	names(outT) <- c("Variable", "LowerCI", "Estimate", "UpperCI")
	outT	
}

############################
## Read in supplementary data ##
############################

## Read in data if saved previously
## full <- read.csv("C:/Users/jputila/Desktop/PLoS_Supplementary/full_8_18.csv")
## Remove Column Header
## full <- full[,-1]

AdjRatebak <- full$AdjRate
full$AdjRate <- round(as.numeric(full$AdjRate),0)

## Assign smoking rate (>=100 cigs over lifetime)
full$smkrate <- full$rate100

#######################
## Measure Associations ##
#######################

## Generate centered and transformed variables
full$lnAs <- log(full$Ascounty) - mean(na.omit(log(full$Ascounty)))
full$lnInc <- log(full$MedIncome) - mean(na.omit(log(full$MedIncome)))
full$Population <- as.numeric(as.character(full$Population))
full$lnsmk <- full$smkrate

## Bivariate, Untransformed
## Arsenic Levels and Lung Cancer Incidence, weighted
glm1 <- glm(full$AdjRate ~ full$Ascounty, family=poisson, weights=as.numeric(full$Population))
summary(glm1)
cconf(glm1)
plot(y=log(full$AdjRate), x=full$Ascounty, ylab="ln(Cancer Incidence per 10,000,000)", xlab="ln(As [ppm])", main="Association between Chromium and Lung Cancer Incidence")
abline(a=glm1$coef[1], b=glm1$coef[2], col=1)

glm1T <- data.frame(glm1$linear.predictors, glm1$y, glm1$model[3])
glm1gg <- ggplot(glm1T, aes(x=glm1.linear.predictors, y=log(glm1.y)))
glm1gg + geom_point(aes(size=X.weights.)) + 
		 geom_abline(coef(glm(log(full$AdjRate) ~ full$Ascounty, family=poisson, weights=as.numeric(full$Population))))


## Adjusted, Untransformed
## Arsenic, Smoking, SES
SESassmk <- glm(full$AdjRate ~ full$smkrate + full$Ascounty + full$MedIncome, family=poisson, weights=as.numeric(full$Population))
summary(SESassmk)
cconf(SESassmk)


## Estimate the 25, 50, amd 75% quartile points for each variable for the quartiles interaction models
AsCut <- NA
AsCut[1] <- as.numeric(summary(full$lnAs)[2])
AsCut[2] <- as.numeric(summary(full$lnAs)[3])
AsCut[3] <- as.numeric(summary(full$lnAs)[5])

SmkCut <- NA
SmkCut[1] <- as.numeric(summary(full$lnsmk)[2])
SmkCut[2] <- as.numeric(summary(full$lnsmk)[3])
SmkCut[3] <- as.numeric(summary(full$lnsmk)[5])

SESCut <- NA
SESCut[1] <- as.numeric(summary(full$lnInc)[2])
SESCut[2] <- as.numeric(summary(full$lnInc)[3])
SESCut[3] <- as.numeric(summary(full$lnInc)[5])


## Continuous Interaction Models
## Arsenic and Smoking
AsSmk <- full$lnAs * full$lnsmk
intAsSmk <- aov(full$AdjRate ~ full$lnsmk + full$lnAs + full$lnInc + AsSmk, weights=as.numeric(full$Population))
summary(intAsSmk)

## Arsenic and SES
AsSES <- full$lnAs * full$lnInc
intAsSES <- aov(full$AdjRate ~ full$lnsmk + full$lnAs + full$lnInc + AsSES, weights=as.numeric(full$Population))
summary(intAsSES)


## Calculate Strat Groups ##
## Smoking Quartiles
smkgrp <- ifelse(is.na(full$lnsmk), NA, ifelse(full$lnsmk < SmkCut[1], 1, ifelse(full$lnsmk >= SmkCut[1] & full$lnsmk < SmkCut[2], 2, ifelse(full$lnsmk >= SmkCut[2] & full$lnsmk < SmkCut[3], 3, 4))))
## SES Low-Income Cutoffs
SESgrp <- ifelse(is.na(full$MedIncome), NA, ifelse(full$MedIncome < 24000 & !is.na(full$MedIncome), 1, ifelse(full$MedIncome>= 24000 & full$MedIncome< 28700, 2, ifelse(full$MedIncome >= 28700 & full$MedIncome < 38300, 3, 4))))
## SES Quartiles
#SESgrp <- ifelse(full$lnInc< -0.158, 1, ifelse(full$lnInc>=-0.158 & full$lnInc< -0.00391, 2, ifelse(full$lnInc>=-0.00391 & full$lnInc <0.1478, 3, 4)))
## Arsenic Quartiles
AsQ <- ifelse(is.na(full$lnAs), NA, ifelse(full$lnAs < AsCut[1], 1, ifelse(full$lnAs >= AsCut[1] & full$lnAs < AsCut[2], 2, ifelse(full$lnAs >= AsCut[2] & full$lnAs < AsCut[3], 3, 4))))

## Quartile-Based Interaction Models
## Convert quartiles to factors
AsQf <- as.factor(AsQ)
smkgrpf <- as.factor(smkgrp)
smkgrpfbak <- smkgrpf
SESgrpf <- as.factor(SESgrp)


#############
## ARSENIC ##
#############
## Arsenic and Smoking
smkgrpf <- smkgrpfbak
smkgrpf <- ifelse(is.na(smkgrpf), NA, ifelse(smkgrpf==1 | smkgrpf==2, 1, 2))
intAsSmk <- aov(full$AdjRate ~ SESgrpf + AsQf*smkgrpf, weights=as.numeric(full$Population))
summary(intAsSmk)

## Without SES
intAsSmk2 <- aov(full$AdjRate ~ AsQf*smkgrpf, weights=as.numeric(full$Population))
summary(intAsSmk2)

## Arsenic and SES
intAsSES <- aov(full$AdjRate ~ smkgrpf + AsQf*SESgrpf, weights=as.numeric(full$Population))
summary(intAsSES)


## GLMS for smoking levels WITHOUT SES
## Bottom 50% vs Top 50%
smkgrpf <- smkgrpfbak
smkgrpf <- ifelse(is.na(smkgrpf), NA, ifelse(smkgrpf==1 | smkgrpf==2, 1, 2))
r1 <- glm(full[smkgrpf==1,]$AdjRate ~ full[smkgrpf==1,]$lnAs, family=poisson, weights=as.numeric(full[smkgrpf==1,]$Population))
summary(r1)
r2 <- glm(full[smkgrpf==2,]$AdjRate ~ full[smkgrpf==2,]$lnAs, family=poisson, weights=as.numeric(full[smkgrpf==2,]$Population))
summary(r2)


## Plot the Interaction between Arsenic and Smoking WITHOUT SES
data1 <- cbind( c(t(r1$model[2]),t(r2$model[2])),
				c(log(r1$fitted.values),log(r2$fitted.values)),
				c(r1$weights, r2$weights),
				c(rep("1",dim(r1$model[2])[1]), rep("2",dim(r2$model[2])[1])))
data1 <- as.data.frame(data1, stringsAsFactors=FALSE)
names(data1) <- c("logAs","logRate", "weight", "smkgrp")
data1$logAs  <- as.numeric(as.character(data1$logAs))
data1$logRate <- as.numeric(as.character(data1$logRate))
data1$weight  <- as.numeric(as.character(data1$weight))
data1$smkgrp  <- as.numeric(as.character(data1$smkgrp))
data1$adjinc  <- c(as.numeric(coef(r1)[2])*r1$model[,2]*0, as.numeric(coef(r2)[2])*r2$model[,2]*0)
data1$adjrate <- data1$adjinc+data1$logRate

assmkp <- ggplot(data1, aes(x=logAs, y=adjrate, shape=factor(smkgrp), color=factor(smkgrp))) 
assmkp + stat_smooth(method = "glm", level=0.95, alpha=1, fill="grey80", color="black") +
		 #scale_color_manual(values=c("grey50","grey70")) +		
		 geom_point(aes(size=weight)) +
		 geom_point() +
		 opts(legend.position = "right") +
		 theme_bw()



## GLMS for smoking levels WITH SES
smkgrpf <- smkgrpfbak
## Bottom 50% vs Top 50%
smkgrpf <- ifelse(is.na(smkgrpf), NA, ifelse(smkgrpf==1 | smkgrpf==2, 1, 2))
r1 <- glm(full[smkgrpf==1,]$AdjRate ~ full[smkgrpf==1,]$lnAs + full[smkgrpf==1,]$MedIncome, family=poisson, weights=as.numeric(full[smkgrpf==1,]$Population))
summary(r1)
r2 <- glm(full[smkgrpf==2,]$AdjRate ~ full[smkgrpf==2,]$lnAs + full[smkgrpf==2,]$MedIncome, family=poisson, weights=as.numeric(full[smkgrpf==2,]$Population))
summary(r2)


## Plot the Interaction between Arsenic and Smoking with SES
data1 <- cbind( c(t(r1$model[2]),t(r2$model[2])), 
				c(log(r1$fitted.values),log(r2$fitted.values)),
				c(r1$weights, r2$weights), 
				c(rep("1",dim(r1$model[2])[1]), rep("2",dim(r2$model[2])[1])))
data1 <- as.data.frame(data1, stringsAsFactors=FALSE)
names(data1) <- c("logAs","logRate", "weight", "smkgrp")
data1$logAs  <- as.numeric(as.character(data1$logAs))
data1$logRate <- as.numeric(as.character(data1$logRate))
data1$weight  <- as.numeric(as.character(data1$weight))
data1$smkgrp  <- as.numeric(as.character(data1$smkgrp))
data1$adjinc  <- c(as.numeric(coef(r1)[2])*r1$model[,2], as.numeric(coef(r2)[2])*r2$model[,2])
data1$adjrate <- data1$adjinc+data1$logRate

assmkp <- ggplot(data1, aes(x=logAs, y=adjrate, shape=factor(smkgrp), color=factor(smkgrp))) 
assmkp + stat_smooth(method = "glm", level=0.95, alpha=1, fill="grey80", color="black") +
		 #scale_color_manual(values=c("grey50","grey70")) +		
		 geom_point(aes(size=weight)) +
		 geom_point() +
		 opts(legend.position = "right") +
		theme_bw()


#################################
## West Virginia / Kentucky Analyses ##
#################################

## Create WV/KY set via FIPS (WV=54, KY=21)
wvky <- full[full$SFIPS==54 | full$SFIPS==21,]

## Bivariate, National
glm1 <- glm(full$AdjRate ~ full$lnAs, family=poisson, weights=as.numeric(full$Population))
summary(glm1)
plot(full$AdjRate, full$lnAs, xlab="Cancer Incidence per 100,000", ylab="As [ppm]")
abline(a=glm1$coef[1], b=glm1$coef[2], col=1)

## Adjusted for Smoking, National
glm2 <- glm(full$AdjRate ~ full$smkrate + full$Ascounty, family=poisson, weights=as.numeric(full$Population))
summary(glm2)

## Adjusted for Smoking, Income, National
glm3 <- glm(full$AdjRate ~ full$smkrate + full$Ascounty + full$MedIncome, family=poisson, weights=as.numeric(full$Population))
summary(glm3)

## GGplot
full$wvkyid <- ifelse(full$SFIPS==54 | full$SFIPS==21, 1, 0)
cols <- c("0" = "red","1" = "blue")
p1 <- ggplot(full, aes(y=AdjRate, x=log(Ascounty), weight=Population, colour=full$wvkyid))
p1 + stat_smooth(method=glm, level=0.95, alpha=1, fill="grey80", color="black") + geom_point() + scale_colour_manual(values = cols, limits = c("0", "1")) + theme_bw()

## Compare residuals for predicting on WVKY using the National fully controlled model
pred1 <- predict(glm3, full)
hist(pred1)
diffresid <- pred1 - log(full$AdjRate)
resids <- data.frame(as.character(full$County), pred1, full$AdjRate, diffresid)
qplot(diffresid[full$wvkyid==1], data=resids, geom="histogram") + theme_bw()


## Determine the concentration of each heavy metal in KY|WV and !KY&!WV
## Average County averages for both states
## Arsenic
KYWVas 	  <- mean(na.omit(full[full$SFIPS==21 | full$SFIPS==54,]$Ascounty)) 
KYWVas
notKYWVas <- mean(na.omit(full[full$SFIPS!=21 & full$SFIPS!=54,]$Ascounty)) 
notKYWVas
t.test(na.omit(full[full$SFIPS==21 | full$SFIPS==54,]$Ascounty), 
	   na.omit(full[full$SFIPS!=21 & full$SFIPS!=54,]$Ascounty))

## Smoking Prevalence
KYWVsmk 	  <- mean(na.omit(full[full$SFIPS==21 | full$SFIPS==54,]$smkrate)) 
KYWVsmk
notKYWVsmk <- mean(na.omit(full[full$SFIPS!=21 & full$SFIPS!=54,]$smkrate)) 
notKYWVsmk
t.test(na.omit(full[full$SFIPS==21 | full$SFIPS==54,]$smkrate), 
	   na.omit(full[full$SFIPS!=21 & full$SFIPS!=54,]$smkrate))

## Median Income
KYWVmed	   <- mean(na.omit(full[full$SFIPS==21 | full$SFIPS==54,]$MedIncome)) 
KYWVmed
notKYWVmed <- mean(na.omit(full[full$SFIPS!=21 & full$SFIPS!=54,]$MedIncome)) 
notKYWVmed
t.test(na.omit(full[full$SFIPS==21 | full$SFIPS==54,]$MedIncome), 
	   na.omit(full[full$SFIPS!=21 & full$SFIPS!=54,]$MedIncome))

## Lung cancer incidence
KYWVrate	<- mean(na.omit(full[full$SFIPS==21 | full$SFIPS==54,]$AdjRate)) 
KYWVrate
notKYWVrate	<- mean(na.omit(full[full$SFIPS!=21 & full$SFIPS!=54,]$AdjRate)) 
notKYWVrate
t.test(na.omit(full[full$SFIPS==21 | full$SFIPS==54,]$AdjRate), 
	   na.omit(full[full$SFIPS!=21 & full$SFIPS!=54,]$AdjRate))


## Draw a bar chart of the values used in the T-tests
bardata <- c(notKYWVas, KYWVas,
			 notKYWVsmk*100, KYWVsmk*100, notKYWVmed/1000, KYWVmed/1000, notKYWVrate, KYWVrate)
barcats <- c("Arsenic", "Arsenic",
			 "Smoking", "Smoking", "Income", "Income", "Rate", "Rate")
barid <- c("Other", "WV/KY",
		   "Other", "WV/KY", "Other", "WV/KY", "Other", "WV/KY")
bardata1 <- data.frame(barid, barcats, bardata)
names(bardata1) <- c("ID", "Category", "Values")

bar1 <- ggplot(bardata1, aes(Category, fill=ID, y=Values))
bar1 +  geom_bar(position="dodge", stat="identity") + theme_bw()



## END of code