rm(list=ls())
setwd("C:\\Users\\julia\\Documents\\Thesis\\SocClass\\Scandinavian journal of public health\\Response to reviewers")

library(metafor)
library(ggplot2)

#get data
soc_cl <- read.csv2("SocCl.csv")
str(soc_cl)

#effect size
es<-escalc(measure="RR", ai=ai, ci=ci, n1i=n1i, n2i=n2i, data=soc_cl)
es$AV2 <- log(es$AV)
es$SD  <- (log(es$IC_U) - log(es$IC_L))/(2*1.96)
es$V <- es$SD^2
es$yi[is.na(es$yi)] <- es$AV2[is.na(es$yi)]
es$vi[is.na(es$vi)] <- es$V[is.na(es$vi)]
es<-es[order(es$E, es$yi) ,]
str(es)
View(es)

#pooled effect size  all indicators together, fixed and random effect
a<-rma(yi=yi, vi=vi, data=es, measure="RR", method="FE", slab=Author)
predict(a,  transf=exp, digits=2)

a<-rma(yi=yi, vi=vi, data=es, measure="RR", slab=paste(Author, E, sep=", "))
predict(a,  transf=exp, digits=2)


#pooled effect size  in indicators, fixed and random effect
a <- rma(yi=yi, vi=vi, data=es, method="FE")
tbl1 <-function(a) {c(paste(round(exp(a$beta[1,1]),2), " (", round(exp(a$ci.lb[1]),2), ", ", 
                            round(exp(a$ci.ub[1]),2), ")", "; I2 = ", round(a$I2,2), "%", sep=""))}
tbl1(a)

t1 <- rep(NA)
for (i in (unique(es$E))) {
  for (j in (c("FE", "REML"))) {
    a<-rma(yi=yi, vi=vi, data=es[es$E==i,], method=j)
    b<-c(paste(i,j), tbl1(a))
    t1<-c(t1,b)
  }
}

t0<-t1[-1]
t1d<- as.data.frame(matrix(t0, byrow=TRUE, ncol=4, nrow=5))
write.csv2(t1d, "table1.csv")

#separate models in social class indicators
tbl2 <-function(a) {c(paste(round(exp(a$beta[1,1]),2), " (", round(exp(a$ci.lb[1]),2), ", ", round(exp(a$ci.ub[1]),2), ")", sep=""),
paste(round(exp(a$beta[2,1]),2), " (", round(exp(a$ci.lb[2]),2), ", ", round(exp(a$ci.ub[2]),2), ")", sep=""),
paste(round(exp(a$beta[3,1]),2), " (", round(exp(a$ci.lb[3]),2), ", ", round(exp(a$ci.ub[3]),2), ")", sep=""),
paste(round(exp(a$beta[4,1]),2), " (", round(exp(a$ci.lb[4]),2), ", ", round(exp(a$ci.ub[4]),2), ")", sep=""),
round(a$QMp,3))}
a<-rma.mv(yi=yi, V=vi,  data=es[es$E=="E",],  mods = ~ Year+Age+Women)


t1 <- rep(NA,5)
t2 <- rep(NA,5)
t<- data.frame(t1, t2)
for (i in (c("E", "I", "O"))){
  a<-rma.mv(yi=yi, V=vi,  data=es[es$E==i,],  mods = ~ Year+Age+Women)
  b<-tbl2(a)
  t<-cbind(t,b)
}

rownames(t)<-c("Intercept", "Year", "Age", "Sex", "Moderators' p-value")
colnames(t)<-c("1", "2", "Education", "Income", "Occupation")
t<-t[,3:5]
write.csv2(t, "table2.csv")

#effect of country and setting

a<-rma.mv(yi=yi, V=vi, data=es[es$E=="O",],  mods = ~ Year+Age+Women,  random = ~ 1 | Country/Author)
b<-rma.mv(yi=yi, V=vi, data=es[es$E=="O",],  mods = ~ Year+Age+Women,  random = ~ 1 | Country/Author, sigma2 =  c(0, NA))
print(anova(b,a))


#cohort effect

a<-rma.mv(yi=yi, V=vi, data=es[es$E=="O",],  mods = ~ Age+Women+Country,  random = ~ 1 | Year/Author)
b<-rma.mv(yi=yi, V=vi, data=es[es$E=="O",],  mods = ~ Age+Women+Country,  random = ~ 1 | Year/Author, sigma2 =  c(0, NA))
print(anova(b,a))

rma.mv(yi=yi, V=vi, data=es,  mods = ~ Age+Women+Beginning+E) 
# race
rma(yi=yi, vi=vi, data=es,  mods = ~ Ethn)


#pulication bias
res <- rma(yi,vi, method = "ML",  data=es)
taf <- trimfill(res)
funnel(taf, main="Funnel plot, social class indicators and the risk of dementia") 
a<-trimfill(rma.uni(yi,vi, method = "ML",  data=es), side="left")
b<-trimfill(rma.uni(yi,vi, method = "ML",  data=es), side="right")

predict(a,  transf=exp, digits=2)
predict(b,  transf=exp, digits=2)


# Forest plot dans les susgroupes  3
str(es)
res <- rma(yi,vi,data=es, slab=paste(Author, ", ", E, " " ,Neib, sep=""))

funtxt <-function(res) {paste ("RR: ", round(exp(res$b),2), "; p-value: ", round(res$pval, 3),
                               "; tau2: ", round(res$tau2,2) , "; I2: ", round(res$I2,1),"%",  sep="")}

forest(res, order=order(es$E, es$yi), 
       xlim=c(-4.5, 4), at=log(c(0.25, 0.5, 1, 3.5, 7)), ylim=c(-1, 63), atransf=exp, 
       ilab.xpos=c(-9.5,-8,-6,-4.5), cex=0.75, 
       rows=c(4:15, 20:28, 33:38, 43:52, 57:58),
       mlab=funtxt(res), psize=1, header="Author(s), year", 
       main="Social class indicators and the risk of dementia")

par(font=4)

### add text for the subgroups
text(-4.55,c(16, 29, 39, 53,59), pos=4, c("Education", "Income", 
                                          "Neighbourhood, aggregated", "Occupation", "Wealth"), cex=0.75)

### fit random-effects model in the subgroups
res.e <- rma(yi,vi,data=es, subset=(E=="E"), slab=Author)
res.i <- rma(yi,vi,data=es, subset=(E=="I"), slab=Author)
res.o <- rma(yi,vi,data=es, subset=(E=="O"), slab=Author)
res.n <- rma(yi,vi,data=es, subset=(E=="N"), slab=Author)
res.w <- rma(yi,vi,data=es, subset=(E=="W"), slab=Author)

### add diamnats
addpoly(res.e, row=2.5,  cex=0.75, atransf=exp, mlab=funtxt(res.e), col="white")
addpoly(res.i, row=18.5, cex=0.75, atransf=exp, mlab=funtxt(res.i), col="white")
addpoly(res.o, row=31.5, cex=0.75, atransf=exp, mlab=funtxt(res.o), col="white")
addpoly(res.n, row=41.5, cex=0.75, atransf=exp, mlab=funtxt(res.n), col="white")
addpoly(res.w, row=55.5, cex=0.75, atransf=exp, mlab=funtxt(res.w), col="white")

text(-4.5, -2, pos=4, cex=0.75, c("Risk ratios, reference group: the 'highest' class"))