######################
#   Generic Inputs   #
######################

#Current Dataset
trials<-unique(dataset$trialnr)
n = length(dataset$trial[as.numeric(dataset$trial)==max(as.numeric(dataset$trial))]) #Sample size of current dataset
y = dataset$os[as.numeric(dataset$trial)==max(as.numeric(dataset$trial))] #Follow-up times, vector of length n
delta = as.numeric(dataset$osi[as.numeric(dataset$trial)==max(as.numeric(dataset$trial))])-1 #Event Indicators, vector of length n
p = 1 #Number of baseline covariates or # of columns in X below
X = matrix(cbind(dataset$age[as.numeric(dataset$trial)==max(as.numeric(dataset$trial))],as.numeric(dataset$sex[as.numeric(dataset$trial)==max(as.numeric(dataset$trial))])),ncol=p,nrow=n) #Baseline Covariates in both datasets, n x p matrix
X = matrix(dataset$age[as.numeric(dataset$trial)==max(as.numeric(dataset$trial))],ncol=p,nrow=n) #Baseline Covariates in both datasets, n x p matrix

q = 1 #Number of novel exposures or # of columns in Z below
Z = matrix(dataset$Intervention[as.numeric(dataset$trial)==max(as.numeric(dataset$trial))],ncol=q,nrow=n) #Exposures only in current dataset, n x q matrix

#Historical Dataset
n0 = length(dataset$trial[as.numeric(dataset$trial)<max(as.numeric(dataset$trial))]) #Sample size of historical dataset
y0 =  dataset$os[as.numeric(dataset$trial)<max(as.numeric(dataset$trial))] #Follow-up times, vector of length n0
delta0 =  as.numeric(dataset$osi[as.numeric(dataset$trial)<max(as.numeric(dataset$trial))])-1 #Event Indicators, vector of length n0
X0 = matrix(cbind(dataset$age[as.numeric(dataset$trial)<max(as.numeric(dataset$trial))],as.numeric(dataset$sex[as.numeric(dataset$trial)<max(as.numeric(dataset$trial))])),ncol=p,nrow=n0) #Baseline Covariates in both datasets, n0 x p matrix
X0 = matrix(dataset$age[as.numeric(dataset$trial)<max(as.numeric(dataset$trial))],ncol=p,nrow=n0)
y<-y/max(y0)
y0<-y0/max(y0)

colnames(X)<-"age"
##################################
#   Other Model Specifications   #
##################################
#"Spike and Slab" hyperparameters
l.slab=0.0001  #S_l (lower bound of slab) 
u.slab=2 #S_u (upper bound of slab)
p0=0.9 #p_0 (prior probability of being in the slab)
spike=c(200,200) #First Entry is Spike for Beta Regression Parameters, Second entry is Spike for baseline hazard (alpha parameter)

#Knot Placement/Location Calculations
K=max(5,min(floor(sum(delta)/8),20))
kappa=c(0,quantile(y[delta==1],seq(0,1,by=1/K))[-c(1,K+1)],1)

###################
#    JAGS Code    #
###################
model.data <- list("n0", "n", "p", "q", "K", "kappa", "y0", "delta0", "X0", "y", "delta", "X", "Z", "l.slab", "u.slab", "spike", "p0")
model.parameters <- c("beta","beta0", "gamma", "tau", "alpha","alpha0")
#Use for constant gamma
model.initial.values <- function() {list (beta0=rnorm(p,0,3), beta=rnorm(p,0,3), gamma=rnorm(q,0,3), sigma0=dunif(1,0.01,100), sigma=dunif(1,0.01,100), alpha=rnorm(K,0,3), alpha0=rnorm(K,0,3))}
#Use for time-dependent gamma
#model.initial.values <- function() {list (beta0=rnorm(p,0,3), beta=rnorm(p,0,3), gamma=matrix(rnorm(q*K,0,3),ncol=q,nrow=K), sigma0=dunif(1,0.01,100), sigma=dunif(1,0.01,100), alpha=rnorm(K,0,3), alpha0=rnorm(K,0,3))}

#Run JAGS
model.fit <- jags(model.file="OurModel.txt", data=model.data, inits=model.initial.values,
               parameters=model.parameters, n.chains = 1, n.iter=1.2*niter, n.burnin=0.2*niter, n.thin=5)
plot(model.fit)
#Compile Results and Compare to Cox PH
#Table 2 in Paper
#Hazard Ratio Estimates
est.comp = rbind(exp(c(model.fit$BUGSoutput$mean$beta,model.fit$BUGSoutput$mean$gamma)),
exp(coxph(Surv(y,delta)~X+Z)$coef),
exp(c(coxph(Surv(y0,delta0)~X0)$coef,NA)),
exp(coxph(Surv(c(y,y0),c(delta,delta0))~rbind(X,X0)+rbind(Z,cbind(rep(0,n0))))$coef))
rownames(est.comp) = c("Proposed Method","Current Cox PH","Historical Cox PH","Pooled Cox PH")
colnames(est.comp) = c(colnames(X),"Intervention")
print(round(t(est.comp),3))

#Confidence/Credible Interval Estimates
curr.ph = coxph(Surv(y,delta)~X+Z)
round(summary(curr.ph)$conf.int[,c(3,4)],2)
hist.ph = coxph(Surv(y0,delta0)~X0)
round(summary(hist.ph)$conf.int[,c(3,4)],2)
pool.ph = coxph(Surv(c(y,y0),c(delta,delta0))~rbind(X,X0)+rbind(Z,cbind(rep(0,n0))))
#round(summary(pool.ph)$conf.int[,c(3,4)],2)
#round(t(apply(exp(model.fit$BUGSoutput$sims.matrix[,c(13,14,16)]),2,function(x) quantile(x,c(.025,.975)))),2)

#If using time-dependent gamma
#exp(model.fit$BUGSoutput$mean$gamma)
#round(t(apply(exp(model.fit$BUGSoutput$sims.matrix[,16:27]),2,function(x) quantile(x,c(.025,.975)))),2)