CombiModel=function(nIter=1e5,paUncPref=T,paSpPeV=T,paUncPerf=T,included=c(T,T,T,T,F,T,T,F)){
  
  ### 1 LOADING PREREQUISITE PACKAGES ###
  # required package for multivariate normal distribution
  require("MASS")
  # required package for simultaneously drawing from multiple normal distributions 
  require("plyr")
  # required package for converting from correlation matrix to covariance matrix
  require("MBESS")
  
  # required function for rank probabilities
  getRankMat <- function(values, doExclude=FALSE, excludeA=NA){
    require("plyr")
    if (doExclude==TRUE){values[, excludeA] <- min(values) - 1e6}
    ranks <- aaply(-values,1,rank,.progress = "text")
    return(ranks)
  }
  
  ### 2 DATA LOADING ###
  # load preference data for case
  load("data")
  # load performance data for case
  load("pdata")
  drugNames <- c("Abacavir/Lamivudine", 
                 "Tenofovir/Emtricitabine", 
                 "Dolutegravir+TE/AL", 
                 "Efavirenz+AL", 
                 "Raltegravir (EXCLUDED)", 
                 "Atazanavir/Ritonavir+EL", 
                 "Elvitegravir/Cobicistat+EL", 
                 "Darunavir/Ritonavir (EXCLUDED)"
  )
  
  # show plots of preference data
  barplot(-mdata$mean / (min(mdata$mean) - max(mdata$mean)), main="Preference weights from Hauber et al study (blue=included)", names.arg=mdata$par, col=rainbow(2)[c(2,2,2,1,1,1,1,2)])
  
  ### 3 PREFERENCSE ###
  # 3.1 Include parameter uncertainty in preferences or not?
  if (paUncPref){
    popweights <- mvrnorm(nIter, mdata$mean, mdata$meancov)
  } else {
    popweights <- matrix(rep(mdata$mean, 8 * nIter), nrow=nIter, ncol=8, byrow = T)
  }
  
  # 3.2 include patient-specific preference variation or not?
  ptspecificvariation <- matrix(0, nrow=nIter, ncol=8)
  if (paSpPeV){
    parUncVarEstimator <- mvrnorm(nIter, mdata$sd, mdata$sdcov)
    for(r in 1:nIter){ 
      ptspecificvariation[r, ]=mvrnorm(1, rep(0,8), diag(parUncVarEstimator[r, ] ^ 2))
    }
  }
  
  # 3.3 combine preference samples according to Formula 3
  prefsamples <- popweights + ptspecificvariation
  
  colnames(prefsamples) <- mdata$par
  
  ### 4 CLINICAL PERFORMANCES ###
  # 4.1 get samples
  perfsamples <- matrix(0, nrow=nIter, ncol=length(pdata$p1))
  colnames(perfsamples) <- paste(rep(pdata$par,(length(pdata$p1)/length(pdata$par))),"_",rep(1:(length(pdata$p1)/length(pdata$par)),each=length(pdata$par)),sep="")
  if (paUncPerf){
    for(i in 1:ncol(perfsamples)){
      if(pdata$p2[i]>pdata$p1[i]){
        perfsamples[,i] <- rbeta(nIter,pdata$p1[i],pdata$p2[i]) # assumption : all distributions are beta
      }
    }
  } else {
    for(i in 1:ncol(perfsamples)){
      if (pdata$p1[i]>0 & pdata$p2[i]>0){
        perfsamples[,i] <- pdata$p1[i] / (pdata$p1[i]+pdata$p2[i])
      }
    }
  }
  
  # 4.2 rescale because preferences are over percentages [1;100], not over probabilities [0;1]
  perfsamples <- perfsamples * 100
  
  ### 5 VALUE PREDICTION ###
  # 5.1 predicted values per attribute
  predPrefWeights <- matrix(nrow=nIter, ncol=length(pdata$p1))
  colnames(predPrefWeights) <- paste(colnames(perfsamples),"_pw",sep="")
  n <- length(pdata$par)
  for(i in 1:(length(pdata$p1) / length(pdata$par))){
    predPrefWeights[, ((i-1)*n+1):(i * n)] <- prefsamples * perfsamples[, ((i-1)*n+1):(i*n)]
  }
  
  ### 6 OUTCOME MEASURES
  # 6.1 combine value per attribute to get overall values
  overallValues <- matrix(nrow=nIter, ncol=length(pdata$p1) / length(pdata$par))
  n <- length(pdata$par)
  for(i in 1:(length(pdata$p1) / length(pdata$par))){
    overallValues[,i] <- rowSums(predPrefWeights[, ((i-1)*n+1):(i*n)])
  }
  
  # 6.2 calculate ranks
  r <- getRankMat(overallValues, T, !included)
  
  ### RETURN STATEMENT ###
  return(list(
    prefsamples=prefsamples,
    perfsamples=perfsamples,
    meanValues=colMeans(overallValues[,included]),
    values=overallValues[,included],
    ranks=r[,included]
  ))
}

nIter=1e4
dat=list(scen1=NA,scen2=NA,scen3=NA,scen4=NA)
dat[[1]]=CombiModel(nIter,T,T,T) # all types of uncertainty
dat[[2]]=CombiModel(nIter,T,F,F) # only parameter uncertainty in preferences
dat[[3]]=CombiModel(nIter,F,T,F) # only patient-specific preference variation
dat[[4]]=CombiModel(nIter,F,F,T) # only parameter uncertainty in performances