DTR
Estimation and Comparison of Dynamic Treatment Regimes

Package index
Search the DTR package
Vignettes
Functions
35
Source code
21
Man pages
30
Home
/
CRAN
/
DTR
/
R/CHRestimate.R

R/CHRestimate.R
In DTR: Estimation and Comparison of Dynamic Treatment Regimes 

###################################################
### Reference:
### Tang X, Wahed AS: Cumulative hazard ratio estimation for treatment regimes in
### sequentially randomized clinical trials. Statistics in Biosciences, [Epub ahead of print]
###################################################

###################################################
### code chunk number 1: chunklibraries
###################################################

#Libraries required
require(survival)

###################################################
### code chunk number 2: chunkCHR
###################################################

updateBeta <- function(beta, # Current coefficient(s) for covariate(s)
                       V, # Covariate(s)
                       U, # Observed survival time
                       delta, # Censoring indicator
                       w11, # Weights for A1B1
                       w12, # Weights for A1B2
                       w21, # Weights for A2B1
                       w22 # Weights for A2B2
) {
  
  #Total number of subjects
  n <- length(U)
  
  #If only 1 covariate
  if(NCOL(V)==1) { 
    
    #Change V from matrix to numeric
    V <- as.numeric(V)
    
    #Calcualte exp(beta*V)
    e <- exp(beta*V)
    
    #Define ui and derivative of U: dui
    ui <- 0
    dui <- 0
    
    for(i in 1:length(U)) {
      
      ind <- as.numeric(U >= U[i])
      
      #Calculate s0j      
      s00 <- sum(ind * w11 * e) / n
      s01 <- sum(ind * w12 * e) / n
      s02 <- sum(ind * w21 * e) / n
      s03 <- sum(ind * w22 * e) / n
      
      #Calculate s1j   
      s10 <- sum(ind * w11 * e * V) / n
      s11 <- sum(ind * w12 * e * V) / n
      s12 <- sum(ind * w21 * e * V) / n
      s13 <- sum(ind * w22 * e * V) / n
      
      #Calculate s2j
      s20 <- sum(ind * w11 * e * V * V) / n
      s21 <- sum(ind * w12 * e * V * V) / n
      s22 <- sum(ind * w21 * e * V * V) / n
      s23 <- sum(ind * w22 * e * V * V) / n
      
      #Calculate v1?_bar
      if(s00 != 0) v10_bar <- s10/s00 else v10_bar <- 0
      if(s01 != 0) v11_bar <- s11/s01 else v11_bar <- 0
      if(s02 != 0) v12_bar <- s12/s02 else v12_bar <- 0
      if(s03 != 0) v13_bar <- s13/s03 else v13_bar <- 0
      
      #calculate v2?_bar
      if(s00 != 0) v20_bar <- s20/s00 else v20_bar <- 0
      if(s01 != 0) v21_bar <- s21/s01 else v21_bar <- 0
      if(s02 != 0) v22_bar <- s22/s02 else v22_bar <- 0
      if(s03 != 0) v23_bar <- s23/s03 else v23_bar <- 0     
      
      #Calculate each ui
      ui <- ui + delta[i]*w11[i]*(V[i] - v10_bar) + 
        delta[i]*w12[i]*(V[i] - v11_bar) +
        delta[i]*w21[i]*(V[i] - v12_bar) +
        delta[i]*w22[i]*(V[i] - v13_bar)
      
      #Calculate each dui
      dui <- dui + delta[i]*w11[i]*(v10_bar^2 - v20_bar) +
        delta[i]*w12[i]*(v11_bar^2 - v21_bar) +
        delta[i]*w21[i]*(v12_bar^2 - v22_bar) +
        delta[i]*w22[i]*(v13_bar^2 - v23_bar)
      
    }
    
  }
  
  #If more than 1 covariates
  if(NCOL(V)>1) { 
    
    #Calcualte exp(beta*V)
    e <- as.numeric(exp(matrix(beta, nrow=1, ncol=NCOL(V)) %*% t(V)))
    
    #Define ui and derivative of U: dui
    ui <- matrix(0, nrow=NCOL(V), ncol=1)
    dui <- matrix(0, nrow=NCOL(V), ncol=NCOL(V))
    
    for(i in 1:length(U)) {
      
      ind <- as.numeric(U >= U[i])
      
      #Calculate s0j      
      s00 <- sum(ind * w11 * e) / n
      s01 <- sum(ind * w12 * e) / n
      s02 <- sum(ind * w21 * e) / n
      s03 <- sum(ind * w22 * e) / n
      
      #Calculate s1j   
      s10 <- t(matrix(ind*w11*e,nrow=1, ncol=n) %*% V / n)
      s11 <- t(matrix(ind*w12*e,nrow=1, ncol=n) %*% V / n)
      s12 <- t(matrix(ind*w21*e,nrow=1, ncol=n) %*% V / n)
      s13 <- t(matrix(ind*w22*e,nrow=1, ncol=n) %*% V / n)
      
      #Calculate s2j
      s20 <- t(V) %*% diag(as.numeric(ind*w11*e)) %*% V / n
      s21 <- t(V) %*% diag(as.numeric(ind*w12*e)) %*% V / n
      s22 <- t(V) %*% diag(as.numeric(ind*w21*e)) %*% V / n
      s23 <- t(V) %*% diag(as.numeric(ind*w22*e)) %*% V / n
      
      #Calculate z1?_bar
      if(s00 != 0) v10_bar <- s10/s00 else v10_bar <- matrix(0, nrow=NCOL(V), ncol=1)
      if(s01 != 0) v11_bar <- s11/s01 else v11_bar <- matrix(0, nrow=NCOL(V), ncol=1)
      if(s02 != 0) v12_bar <- s12/s02 else v12_bar <- matrix(0, nrow=NCOL(V), ncol=1)
      if(s03 != 0) v13_bar <- s13/s03 else v13_bar <- matrix(0, nrow=NCOL(V), ncol=1)
      
      #calculate z2?_bar
      if(s00 != 0) v20_bar <- s20/s00 else v20_bar <- matrix(0, nrow=NCOL(V), ncol=NCOL(V))
      if(s01 != 0) v21_bar <- s21/s01 else v21_bar <- matrix(0, nrow=NCOL(V), ncol=NCOL(V))
      if(s02 != 0) v22_bar <- s22/s02 else v22_bar <- matrix(0, nrow=NCOL(V), ncol=NCOL(V))
      if(s03 != 0) v23_bar <- s23/s03 else v23_bar <- matrix(0, nrow=NCOL(V), ncol=NCOL(V))      
      
      #Calculate each ui
      ui <- ui + delta[i]*w11[i]*(matrix(V[i,], nrow=NCOL(V), ncol=1) - v10_bar) + 
        delta[i]*w12[i]*(matrix(V[i,], nrow=NCOL(V), ncol=1) - v11_bar) +
        delta[i]*w21[i]*(matrix(V[i,], nrow=NCOL(V), ncol=1) - v12_bar) +
        delta[i]*w22[i]*(matrix(V[i,], nrow=NCOL(V), ncol=1) - v13_bar)
      
      #Calculate each dui
      dui <- dui + delta[i]*w11[i]*(v10_bar %*% t(v10_bar) - v20_bar) +
        delta[i]*w12[i]*(v11_bar %*% t(v11_bar) - v21_bar) +
        delta[i]*w21[i]*(v12_bar %*% t(v12_bar) - v22_bar) +
        delta[i]*w22[i]*(v13_bar %*% t(v13_bar) - v23_bar)
      
    }
    
  }
  
  #Update beta
  
  beta_up <- beta - solve(dui) %*% ui    
  return(beta_up)
  
}


CHRestimate <- function(data, # A complete data frame representing the data for two-stage randomization designs
                        # data = data frame {X, R, Z, U, delta, V}
                        # V represents covariates
                        # There could be one covariate, or more than one covariates
                        # The function does not allow the absence of covariates
                        covar=names(data)[!names(data) %in% c("X", "R", "Z", "U", "delta")] # Covariate list
) {
   
  #Retrieve data
  n <- nrow(data)
  X <- data$X # X=0 for A1, X=1 for A2
  R <- data$R
  Z <- data$Z # Z=0 for B1, Z=1 for B2
  U <- data$U
  delta <- data$delta
  
  #Chek for errors
  if (is.null(X)) stop("R can not be empty")
  if (is.null(R)) stop("R can not be empty")
  if (is.null(Z)) stop("Z can not be empty")  
  if (is.null(U)) stop("V can not be empty")  
  if (is.null(delta)) stop("delta can not be empty") 
  
  #Times to be assessed
  t <- unique(U[which(delta==1)])
  #Order times
  t <- t[order(t)]

  #Number at risk
  n.risk <- apply(as.array(t), 1, function(x) sum(as.numeric(U >= x)))
  
  #Number event
  n.event <- apply(as.array(t), 1, function(x) length(which(U==x & delta==1)))
  
  #Estimate probability of being assigned to A2
  pi.x <- sum(X)/n
  
  #Estimate probability of being assigned to B2 (allowing probability to vary across A1 and A2)
  pi.z1 <- sum((1-X)*R*Z) / sum((1-X)*R)
  pi.z2 <- sum(X*R*Z) / sum(X*R)
  
  #Calculate weight for A1B1, A1B2, A2B1, and A2B2
  w11 <- (1-X)*(1-R)/(1-pi.x) + (1-X)*R*(1-Z)/((1-pi.x)*(1-pi.z1))
  w12 <- (1-X)*(1-R)/(1-pi.x) + (1-X)*R*Z/((1-pi.x)*pi.z1)
  w21 <- X*(1-R)/pi.x + X*R*(1-Z)/(pi.x*(1-pi.z1))
  w22 <- X*(1-R)/pi.x + X*R*Z/(pi.x*pi.z1)
  
  #################################################
  #################################################
  #If no covariates: ERROR
  #################################################
  #################################################
  
  if(length(covar)==0) { stop("Covariate(s) can not be empty") } else {
    
    if(FALSE %in% (covar %in% names(data))) { stop("Covariate(s) can not be found in the data") 
    } else { V <- as.matrix(data[, names(data) %in% covar]) }
 
  }
  
  #Define results
  est <- lest <- NULL
  
  #################################################
  #################################################
  #If only 1 covariates
  #################################################
  #################################################
  if(NCOL(V)==1) { 
    
    #Obtain the inital beta estimates
    beta <- as.numeric(coxph(Surv(U, delta)~., data=data[, names(data) %in% c("U", "delta", covar)])$coef)
    
    #Solve for beta using Newton-Raphson method
    cat("Calling for updateBeta() function to solve for coefficients... \n")

    for (p in 1:1000) {
 
      #Run updateBeta function
      ebeta <- updateBeta(beta, V, U, delta, w11, w12, w21, w22)
    
      #Calculate difference between updated ebeta and beta
      index <- max(abs(ebeta-beta))
    
      if (index <= 10^(-6)) break else { beta <- ebeta; p <- p + 1 }
    
    }
    
    cat("Calculating cumulative hazard ratios and variance/covariance... \n")
    
    #Change V from matrix to numeric
    V <- as.numeric(V)
    
    #Calcualte exp(beta*V)
    e <- exp(ebeta*V)
    
    for(j in 1:length(t)) {
      
      #Define s0 and v1_bar
      s00 <- s01 <- s02 <- s03 <- rep(0, n)
      v10_bar <- v11_bar <- v12_bar <- v13_bar <- rep(0, n)
      
      #Define lambda and omega
      lambda11 <- lambda12 <- lambda21 <- lambda22 <- 0
      omega <- 0
      
      #Define h
      h11 <- h12 <- h21 <- h22 <- 0
      
      for(i in 1:n) {
        
        ind <- as.numeric(U >= U[i])
        
        #Calculate s0j      
        s00[i] <- sum(ind * w11 * e) / n
        s01[i] <- sum(ind * w12 * e) / n
        s02[i] <- sum(ind * w21 * e) / n
        s03[i] <- sum(ind * w22 * e) / n
        
        #Calculate s1j   
        s10 <- sum(ind * w11 * e * V) / n
        s11 <- sum(ind * w12 * e * V) / n
        s12 <- sum(ind * w21 * e * V) / n
        s13 <- sum(ind * w22 * e * V) / n
        
        #Calculate s2j
        s20 <- sum(ind * w11 * e * V * V) / n
        s21 <- sum(ind * w12 * e * V * V) / n
        s22 <- sum(ind * w21 * e * V * V) / n
        s23 <- sum(ind * w22 * e * V * V) / n
        
        #Calculate z1?_bar
        if(s00[i] != 0) v10_bar[i] <- s10/s00[i] else v10_bar[i] <- 0
        if(s01[i] != 0) v11_bar[i] <- s11/s01[i] else v11_bar[i] <- 0
        if(s02[i] != 0) v12_bar[i] <- s12/s02[i] else v12_bar[i] <- 0
        if(s03[i] != 0) v13_bar[i] <- s13/s03[i] else v13_bar[i] <- 0
        
        #calculate z2?_bar
        if(s00[i] != 0) v20_bar <- s20/s00[i] else v20_bar <- 0
        if(s01[i] != 0) v21_bar <- s21/s01[i] else v21_bar <- 0
        if(s02[i] != 0) v22_bar <- s22/s02[i] else v22_bar <- 0
        if(s03[i] != 0) v23_bar <- s23/s03[i] else v23_bar <- 0     
        
        #Calculate each cumulative baseline hazards
        if(s00[i] != 0) lambda11 <- lambda11 + delta[i] * w11[i] * as.numeric(U[i] <= t[j]) / (n*s00[i])
        if(s01[i] != 0) lambda12 <- lambda12 + delta[i] * w12[i] * as.numeric(U[i] <= t[j]) / (n*s01[i])
        if(s02[i] != 0) lambda21 <- lambda21 + delta[i] * w21[i] * as.numeric(U[i] <= t[j]) / (n*s02[i])
        if(s03[i] != 0) lambda22 <- lambda22 + delta[i] * w22[i] * as.numeric(U[i] <= t[j]) / (n*s03[i])
        
        #Calculate each h
        if(s00[i] != 0) h11 <- h11 - delta[i]*w11[i]*as.numeric(U[i] <= t[j])*v10_bar[i] / (n*s00[i])
        if(s01[i] != 0) h12 <- h12 - delta[i]*w12[i]*as.numeric(U[i] <= t[j])*v11_bar[i] / (n*s01[i])
        if(s02[i] != 0) h21 <- h21 - delta[i]*w21[i]*as.numeric(U[i] <= t[j])*v12_bar[i] / (n*s02[i])
        if(s03[i] != 0) h22 <- h22 - delta[i]*w22[i]*as.numeric(U[i] <= t[j])*v13_bar[i] / (n*s03[i])
        
        #Calculate each tao = s2/s0 - v1^2
        tao11i <- v20_bar - v10_bar[i] * v10_bar[i]
        tao12i <- v21_bar - v11_bar[i] * v11_bar[i]
        tao21i <- v22_bar - v12_bar[i] * v12_bar[i]
        tao22i <- v23_bar - v13_bar[i] * v13_bar[i]
        
        #Calculate each omega
        omega <- omega + (delta[i]*w11[i]*tao11i + delta[i]*w12[i]*tao12i + delta[i]*w21[i]*tao21i + delta[i]*w22[i]*tao22i) / n
        
      }    
      
      #Calculate cumulative hazards ratio
      CHR1211 <- lambda12 / lambda11; LogCHR1211 <- log(CHR1211)
      CHR2111 <- lambda21 / lambda11; LogCHR2111 <- log(CHR2111)
      CHR2211 <- lambda22 / lambda11; LogCHR2211 <- log(CHR2211)
      CHR2112 <- lambda21 / lambda12; LogCHR2112 <- log(CHR2112)
      CHR2212 <- lambda22 / lambda12; LogCHR2212 <- log(CHR2212)
      CHR2221 <- lambda22 / lambda21; LogCHR2221 <- log(CHR2221)
      
      # Define xi 
      xi1211 <- xi2111 <- xi2211 <- xi2112 <- xi2212 <- xi2221 <- rep(0, n)
      
      for(k in 1:n) {
        
        #Defind I(Uk >= U)
        uind <- as.numeric(U[k] >= U)
        
        #Calculate each latter part of psi
        psi11i <- w11[k]*uind*e[k]*delta*w11*(V[k]-v10_bar) / (n*s00); psi11i[is.na(psi11i)] <- 0
        psi12i <- w12[k]*uind*e[k]*delta*w12*(V[k]-v11_bar) / (n*s01); psi12i[is.na(psi12i)] <- 0
        psi21i <- w21[k]*uind*e[k]*delta*w21*(V[k]-v12_bar) / (n*s02); psi21i[is.na(psi21i)] <- 0
        psi22i <- w22[k]*uind*e[k]*delta*w22*(V[k]-v13_bar) / (n*s03); psi22i[is.na(psi22i)] <- 0
        
        #Calculate psi
        psi <- delta[k]*w11[k]*(V[k] - v10_bar[k]) + 
          delta[k]*w12[k]*(V[k] - v11_bar[k]) +
          delta[k]*w21[k]*(V[k] - v12_bar[k]) +
          delta[k]*w22[k]*(V[k] - v13_bar[k]) -
          sum(psi11i) - sum(psi12i) - sum(psi21i) - sum(psi22i)
        
        #Calculate each latter part of phi integral         
        intl11i <- w11[k]*uind*e[k]*delta*w11*as.numeric(U<=t[j]) / (n*s00*s00); intl11i[is.na(intl11i)] <- 0
        intl12i <- w12[k]*uind*e[k]*delta*w12*as.numeric(U<=t[j]) / (n*s01*s01); intl12i[is.na(intl12i)] <- 0
        intl21i <- w21[k]*uind*e[k]*delta*w21*as.numeric(U<=t[j]) / (n*s02*s02); intl21i[is.na(intl21i)] <- 0
        intl22i <- w22[k]*uind*e[k]*delta*w22*as.numeric(U<=t[j]) / (n*s03*s03); intl22i[is.na(intl22i)] <- 0
        
        #Calculate each phi_L      
        if(s00[k] != 0) phiL11i <- delta[k]*w11[k]*as.numeric(U[k]<=t[j])/s00[k] - sum(intl11i) else phiL11i <- 0
        if(s01[k] != 0) phiL12i <- delta[k]*w12[k]*as.numeric(U[k]<=t[j])/s01[k] - sum(intl12i) else phiL12i <- 0
        if(s02[k] != 0) phiL21i <- delta[k]*w21[k]*as.numeric(U[k]<=t[j])/s02[k] - sum(intl21i) else phiL21i <- 0
        if(s03[k] != 0) phiL22i <- delta[k]*w22[k]*as.numeric(U[k]<=t[j])/s03[k] - sum(intl22i) else phiL22i <- 0
        
        #Calculate each phi      
        phi11 <- h11 * psi / omega + phiL11i
        phi12 <- h12 * psi / omega + phiL12i
        phi21 <- h21 * psi / omega + phiL21i
        phi22 <- h22 * psi / omega + phiL22i
        
        #Calculate individual xi
        
        xi1211[k] <- phi12 / lambda11 - lambda12 * phi11 / (lambda11)^2
        xi2111[k] <- phi21 / lambda11 - lambda21 * phi11 / (lambda11)^2
        xi2211[k] <- phi22 / lambda11 - lambda22 * phi11 / (lambda11)^2
        xi2112[k] <- phi21 / lambda12 - lambda21 * phi12 / (lambda12)^2
        xi2212[k] <- phi22 / lambda12 - lambda22 * phi12 / (lambda12)^2
        xi2221[k] <- phi22 / lambda21 - lambda22 * phi21 / (lambda21)^2
        
      }
      
      #Save the results
      temp <- c(CHR1211, CHR2111, CHR2211, CHR2112, CHR2212, CHR2221,
                sqrt(mean(xi1211*xi1211)/n), sqrt(mean(xi2111*xi2111)/n),
                sqrt(mean(xi2211*xi2211)/n), sqrt(mean(xi2112*xi2112)/n),
                sqrt(mean(xi2212*xi2212)/n), sqrt(mean(xi2221*xi2221)/n),
                mean(xi1211*xi2111)/n, mean(xi1211*xi2211)/n, mean(xi1211*xi2112)/n,
                mean(xi1211*xi2212)/n, mean(xi1211*xi2221)/n, mean(xi2111*xi2211)/n,
                mean(xi2111*xi2112)/n, mean(xi2111*xi2212)/n, mean(xi2111*xi2221)/n,
                mean(xi2211*xi2112)/n, mean(xi2211*xi2212)/n, mean(xi2211*xi2221)/n,
                mean(xi2112*xi2212)/n, mean(xi2112*xi2221)/n, mean(xi2212*xi2221)/n)
      #Change 0 and NaN to NA
      temp[which(temp==0 | is.na(temp)==TRUE | temp==Inf | temp==-Inf)] <- NA
      est <- rbind(est, temp)
      rownames(est) <- NULL
      ltemp <- c(LogCHR1211, LogCHR2111, LogCHR2211, LogCHR2112, LogCHR2212, LogCHR2221,
                 sqrt(mean(xi1211*xi1211)/(n*CHR1211*CHR1211)), sqrt(mean(xi2111*xi2111)/(n*CHR2111*CHR2111)),
                 sqrt(mean(xi2211*xi2211)/(n*CHR2211*CHR2211)), sqrt(mean(xi2112*xi2112)/(n*CHR2112*CHR2112)),
                 sqrt(mean(xi2212*xi2212)/(n*CHR2212*CHR2212)), sqrt(mean(xi2221*xi2221)/(n*CHR2221*CHR2221)),
                 mean(xi1211*xi2111)/(n*CHR1211*CHR2111), mean(xi1211*xi2211)/(n*CHR1211*CHR2211), mean(xi1211*xi2112)/(n*CHR1211*CHR2112),
                 mean(xi1211*xi2212)/(n*CHR1211*CHR2212), mean(xi1211*xi2221)/(n*CHR1211*CHR2221), mean(xi2111*xi2211)/(n*CHR2111*CHR2211),
                 mean(xi2111*xi2112)/(n*CHR2111*CHR2112), mean(xi2111*xi2212)/(n*CHR2111*CHR2212), mean(xi2111*xi2221)/(n*CHR2111*CHR2221),
                 mean(xi2211*xi2112)/(n*CHR2211*CHR2112), mean(xi2211*xi2212)/(n*CHR2211*CHR2212), mean(xi2211*xi2221)/(n*CHR2211*CHR2221),
                 mean(xi2112*xi2212)/(n*CHR2112*CHR2212), mean(xi2112*xi2221)/(n*CHR2112*CHR2221), mean(xi2212*xi2221)/(n*CHR2212*CHR2221))
      ltemp[which(ltemp==0 | is.na(ltemp)==TRUE | ltemp==Inf | ltemp==-Inf)] <- NA
      lest <- rbind(lest, ltemp)
      rownames(lest) <- NULL

    }
    
  }
  
  #################################################
  #################################################
  #If more than 1 covariates
  #################################################
  #################################################
  if(NCOL(V)>1) { 
    
    #Obtain the inital beta estimates
    beta <- as.numeric(coxph(Surv(U, delta)~., data=data[, names(data) %in% c("U", "delta", covar)])$coef)
    
    #Solve for beta using Newton-Raphson method
    cat("Calling for updateBeta() function to solve for coefficient(s)...\n")

    for (p in 1:1000) {
      
      #Run updateBeta function
      ebeta <- updateBeta(beta, V, U, delta, w11, w12, w21, w22)
      
      #Calculate difference between updated ebeta and beta
      index <- max(abs(ebeta-beta))
      
      if (index <= 10^(-6)) break else { beta <- ebeta; p <- p + 1 }
      
    }
    
    cat("Calculating cumulative hazard ratio and variance/covariance...\n")
    
    #Calcualte exp(beta*V)
    e <- as.numeric(exp(matrix(ebeta, nrow=1, ncol=NCOL(V)) %*% t(V)))
    
    for(j in 1:length(t)) {
    
      #Define s0 and v1_bar
      s00 <- s01 <- s02 <- s03 <- rep(0, n)
      v10_bar <- v11_bar <- v12_bar <- v13_bar <- matrix(0, nrow=NCOL(V), ncol=n)
    
      #Define lambda and omega
      lambda11 <- lambda12 <- lambda21 <- lambda22 <- 0
      omega <- matrix(0, nrow=NCOL(V), ncol=NCOL(V))
    
      #Define h
      h11 <- h12 <- h21 <- h22 <- matrix(0, nrow=NCOL(V), ncol=1)
    
      for(i in 1:n) {
      
        ind <- as.numeric(U >= U[i])
      
        #Calculate s0j      
        s00[i] <- sum(ind * w11 * e) / n
        s01[i] <- sum(ind * w12 * e) / n
        s02[i] <- sum(ind * w21 * e) / n
        s03[i] <- sum(ind * w22 * e) / n
      
        #Calculate s1j   
        s10 <- t(matrix(ind*w11*e,nrow=1, ncol=n) %*% V / n)
        s11 <- t(matrix(ind*w12*e,nrow=1, ncol=n) %*% V / n)
        s12 <- t(matrix(ind*w21*e,nrow=1, ncol=n) %*% V / n)
        s13 <- t(matrix(ind*w22*e,nrow=1, ncol=n) %*% V / n)
      
        #Calculate s2j
        s20 <- t(V) %*% diag(as.numeric(ind*w11*e)) %*% V / n
        s21 <- t(V) %*% diag(as.numeric(ind*w12*e)) %*% V / n
        s22 <- t(V) %*% diag(as.numeric(ind*w21*e)) %*% V / n
        s23 <- t(V) %*% diag(as.numeric(ind*w22*e)) %*% V / n
      
        #Calculate z1?_bar
        if(s00[i] != 0) v10_bar[,i] <- s10/s00[i] else v10_bar[,i] <- matrix(0, nrow=NCOL(V), ncol=1)
        if(s01[i] != 0) v11_bar[,i] <- s11/s01[i] else v11_bar[,i] <- matrix(0, nrow=NCOL(V), ncol=1)
        if(s02[i] != 0) v12_bar[,i] <- s12/s02[i] else v12_bar[,i] <- matrix(0, nrow=NCOL(V), ncol=1)
        if(s03[i] != 0) v13_bar[,i] <- s13/s03[i] else v13_bar[,i] <- matrix(0, nrow=NCOL(V), ncol=1)
      
        #calculate z2?_bar
        if(s00[i] != 0) v20_bar <- s20/s00[i] else v20_bar <- matrix(0, nrow=NCOL(V), ncol=NCOL(V))
        if(s01[i] != 0) v21_bar <- s21/s01[i] else v21_bar <- matrix(0, nrow=NCOL(V), ncol=NCOL(V))
        if(s02[i] != 0) v22_bar <- s22/s02[i] else v22_bar <- matrix(0, nrow=NCOL(V), ncol=NCOL(V))
        if(s03[i] != 0) v23_bar <- s23/s03[i] else v23_bar <- matrix(0, nrow=NCOL(V), ncol=NCOL(V))      
      
        #Calculate each cumulative baseline hazards
        if(s00[i] != 0) lambda11 <- lambda11 + delta[i] * w11[i] * as.numeric(U[i] <= t[j]) / (n*s00[i])
        if(s01[i] != 0) lambda12 <- lambda12 + delta[i] * w12[i] * as.numeric(U[i] <= t[j]) / (n*s01[i])
        if(s02[i] != 0) lambda21 <- lambda21 + delta[i] * w21[i] * as.numeric(U[i] <= t[j]) / (n*s02[i])
        if(s03[i] != 0) lambda22 <- lambda22 + delta[i] * w22[i] * as.numeric(U[i] <= t[j]) / (n*s03[i])
      
        #Calculate each h
        if(s00[i] != 0) h11 <- h11 - delta[i]*w11[i]*as.numeric(U[i] <= t[j])*v10_bar[,i] / (n*s00[i])
        if(s01[i] != 0) h12 <- h12 - delta[i]*w12[i]*as.numeric(U[i] <= t[j])*v11_bar[,i] / (n*s01[i])
        if(s02[i] != 0) h21 <- h21 - delta[i]*w21[i]*as.numeric(U[i] <= t[j])*v12_bar[,i] / (n*s02[i])
        if(s03[i] != 0) h22 <- h22 - delta[i]*w22[i]*as.numeric(U[i] <= t[j])*v13_bar[,i] / (n*s03[i])
      
        #Calculate each tao = s2/s0 - v1^2
        tao11i <- v20_bar - v10_bar[,i] %*% t(v10_bar[,i])
        tao12i <- v21_bar - v11_bar[,i] %*% t(v11_bar[,i])
        tao21i <- v22_bar - v12_bar[,i] %*% t(v12_bar[,i])
        tao22i <- v23_bar - v13_bar[,i] %*% t(v13_bar[,i])
        
        #Calculate each omega
        omega <- omega + (delta[i]*w11[i]*tao11i + delta[i]*w12[i]*tao12i + delta[i]*w21[i]*tao21i + delta[i]*w22[i]*tao22i) / n
      
      }    
  
      #Calculate cumulative hazards ratio
      CHR1211 <- lambda12 / lambda11; LogCHR1211 <- log(CHR1211)
      CHR2111 <- lambda21 / lambda11; LogCHR2111 <- log(CHR2111)
      CHR2211 <- lambda22 / lambda11; LogCHR2211 <- log(CHR2211)
      CHR2112 <- lambda21 / lambda12; LogCHR2112 <- log(CHR2112)
      CHR2212 <- lambda22 / lambda12; LogCHR2212 <- log(CHR2212)
      CHR2221 <- lambda22 / lambda21; LogCHR2221 <- log(CHR2221)
    
      # Define xi
      xi1211 <- xi2111 <- xi2211 <- xi2112 <- xi2212 <- xi2221 <- rep(0, n)
    
      for(k in 1:n) {
      
        #Defind I(Uk >= U)
        uind <- as.numeric(U[k] >= U)
      
        #Calculate each latter part of psi
        psi11i <- w11[k]*uind*e[k]*delta*w11*(V[k,]-v10_bar) / (n*s00); psi11i[is.na(psi11i)] <- 0
        psi12i <- w12[k]*uind*e[k]*delta*w12*(V[k,]-v11_bar) / (n*s01); psi12i[is.na(psi12i)] <- 0
        psi21i <- w21[k]*uind*e[k]*delta*w21*(V[k,]-v12_bar) / (n*s02); psi21i[is.na(psi21i)] <- 0
        psi22i <- w22[k]*uind*e[k]*delta*w22*(V[k,]-v13_bar) / (n*s03); psi22i[is.na(psi22i)] <- 0

        #Calculate psi
        psi <- delta[k]*w11[k]*(matrix(V[k,], nrow=NCOL(V), ncol=1) - v10_bar[,k]) + 
          delta[k]*w12[k]*(matrix(V[k,], nrow=NCOL(V), ncol=1) - v11_bar[,k]) +
          delta[k]*w21[k]*(matrix(V[k,], nrow=NCOL(V), ncol=1) - v12_bar[,k]) +
          delta[k]*w22[k]*(matrix(V[k,], nrow=NCOL(V), ncol=1) - v13_bar[,k]) -
          sum(psi11i) - sum(psi12i) - sum(psi21i) - sum(psi22i)
            
        #Calculate each latter part of phi integral         
        intl11i <- w11[k]*uind*e[k]*delta*w11*as.numeric(U<=t[j]) / (n*s00*s00); intl11i[is.na(intl11i)] <- 0
        intl12i <- w12[k]*uind*e[k]*delta*w12*as.numeric(U<=t[j]) / (n*s01*s01); intl12i[is.na(intl12i)] <- 0
        intl21i <- w21[k]*uind*e[k]*delta*w21*as.numeric(U<=t[j]) / (n*s02*s02); intl21i[is.na(intl21i)] <- 0
        intl22i <- w22[k]*uind*e[k]*delta*w22*as.numeric(U<=t[j]) / (n*s03*s03); intl22i[is.na(intl22i)] <- 0

        #Calculate each phi_L      
        if(s00[k] != 0) phiL11i <- delta[k]*w11[k]*as.numeric(U[k]<=t[j])/s00[k] - sum(intl11i) else phiL11i <- 0
        if(s01[k] != 0) phiL12i <- delta[k]*w12[k]*as.numeric(U[k]<=t[j])/s01[k] - sum(intl12i) else phiL12i <- 0
        if(s02[k] != 0) phiL21i <- delta[k]*w21[k]*as.numeric(U[k]<=t[j])/s02[k] - sum(intl21i) else phiL21i <- 0
        if(s03[k] != 0) phiL22i <- delta[k]*w22[k]*as.numeric(U[k]<=t[j])/s03[k] - sum(intl22i) else phiL22i <- 0
      
        #Calculate each phi      
        phi11 <- t(h11) %*% solve(omega) %*% psi + phiL11i
        phi12 <- t(h12) %*% solve(omega) %*% psi + phiL12i
        phi21 <- t(h21) %*% solve(omega) %*% psi + phiL21i
        phi22 <- t(h22) %*% solve(omega) %*% psi + phiL22i
      
        #Calculate individual xi
      
        xi1211[k] <- phi12 / lambda11 - lambda12 * phi11 / (lambda11)^2
        xi2111[k] <- phi21 / lambda11 - lambda21 * phi11 / (lambda11)^2
        xi2211[k] <- phi22 / lambda11 - lambda22 * phi11 / (lambda11)^2
        xi2112[k] <- phi21 / lambda12 - lambda21 * phi12 / (lambda12)^2
        xi2212[k] <- phi22 / lambda12 - lambda22 * phi12 / (lambda12)^2
        xi2221[k] <- phi22 / lambda21 - lambda22 * phi21 / (lambda21)^2
      
      }

      #Save the results
      #Save the results
      temp <- c(CHR1211, CHR2111, CHR2211, CHR2112, CHR2212, CHR2221,
                sqrt(mean(xi1211*xi1211)/n), sqrt(mean(xi2111*xi2111)/n),
                sqrt(mean(xi2211*xi2211)/n), sqrt(mean(xi2112*xi2112)/n),
                sqrt(mean(xi2212*xi2212)/n), sqrt(mean(xi2221*xi2221)/n),
                mean(xi1211*xi2111)/n, mean(xi1211*xi2211)/n, mean(xi1211*xi2112)/n,
                mean(xi1211*xi2212)/n, mean(xi1211*xi2221)/n, mean(xi2111*xi2211)/n,
                mean(xi2111*xi2112)/n, mean(xi2111*xi2212)/n, mean(xi2111*xi2221)/n,
                mean(xi2211*xi2112)/n, mean(xi2211*xi2212)/n, mean(xi2211*xi2221)/n,
                mean(xi2112*xi2212)/n, mean(xi2112*xi2221)/n, mean(xi2212*xi2221)/n)
      #Change 0 and NaN to NA
      temp[which(temp==0 | is.na(temp)==TRUE | temp==Inf | temp==-Inf)] <- NA
      est <- rbind(est, temp)
      rownames(est) <- NULL
      ltemp <- c(LogCHR1211, LogCHR2111, LogCHR2211, LogCHR2112, LogCHR2212, LogCHR2221,
                 sqrt(mean(xi1211*xi1211)/(n*CHR1211*CHR1211)), sqrt(mean(xi2111*xi2111)/(n*CHR2111*CHR2111)),
                 sqrt(mean(xi2211*xi2211)/(n*CHR2211*CHR2211)), sqrt(mean(xi2112*xi2112)/(n*CHR2112*CHR2112)),
                 sqrt(mean(xi2212*xi2212)/(n*CHR2212*CHR2212)), sqrt(mean(xi2221*xi2221)/(n*CHR2221*CHR2221)),
                 mean(xi1211*xi2111)/(n*CHR1211*CHR2111), mean(xi1211*xi2211)/(n*CHR1211*CHR2211), mean(xi1211*xi2112)/(n*CHR1211*CHR2112),
                 mean(xi1211*xi2212)/(n*CHR1211*CHR2212), mean(xi1211*xi2221)/(n*CHR1211*CHR2221), mean(xi2111*xi2211)/(n*CHR2111*CHR2211),
                 mean(xi2111*xi2112)/(n*CHR2111*CHR2112), mean(xi2111*xi2212)/(n*CHR2111*CHR2212), mean(xi2111*xi2221)/(n*CHR2111*CHR2221),
                 mean(xi2211*xi2112)/(n*CHR2211*CHR2112), mean(xi2211*xi2212)/(n*CHR2211*CHR2212), mean(xi2211*xi2221)/(n*CHR2211*CHR2221),
                 mean(xi2112*xi2212)/(n*CHR2112*CHR2212), mean(xi2112*xi2221)/(n*CHR2112*CHR2221), mean(xi2212*xi2221)/(n*CHR2212*CHR2221))
      ltemp[which(ltemp==0 | is.na(ltemp)==TRUE | ltemp==Inf | ltemp==-Inf)] <- NA
      lest <- rbind(lest, ltemp)
      rownames(lest) <- NULL
     
    }
    
  }
  
  #Return class
  results <- list(Call=match.call(), 
                  coefficients=ebeta, 
                  comparison=c("A1B2 vs. A1B1", "A2B1 vs. A1B1", "A2B2 vs. A1B1",
                               "A2B1 vs. A1B2", "A2B2 vs. A1B2", "A2B2 vs. A2B1"),
                  time75P=as.numeric(round(quantile(data$U, probs=0.75),2)),
                  time=t, n.risk=n.risk, n.event=n.event,
                  CHR1211=est[,1], CHR2111=est[,2], CHR2211=est[,3], 
                  CHR2112=est[,4], CHR2212=est[,5], CHR2221=est[,6],
                  SE1211=est[,7], SE2111=est[,8], SE2211=est[,9],
                  SE2112=est[,10], SE2212=est[,11], SE2221=est[,12],
                  COV1211_2111=est[,13], COV1211_2211=est[,14], COV1211_2112=est[,15],
                  COV1211_2212=est[,16], COV1211_2221=est[,17], COV2111_2211=est[,18],
                  COV2111_2112=est[,19], COV2111_2212=est[,20], COV2111_2221=est[,21],
                  COV2211_2112=est[,22], COV2211_2212=est[,23], COV2211_2221=est[,24],
                  COV2112_2212=est[,25], COV2112_2221=est[,26], COV2212_2221=est[,27],
                  CHR1211.LOG=lest[,1], CHR2111.LOG=lest[,2], CHR2211.LOG=lest[,3], 
                  CHR2112.LOG=lest[,4], CHR2212.LOG=lest[,5], CHR2221.LOG=lest[,6],
                  SE1211.LOG=lest[,7], SE2111.LOG=lest[,8], SE2211.LOG=lest[,9],
                  SE2112.LOG=lest[,10], SE2212.LOG=lest[,11], SE2221.LOG=lest[,12],
                  COV1211_2111.LOG=lest[,13], COV1211_2211.LOG=lest[,14], COV1211_2112.LOG=lest[,15],
                  COV1211_2212.LOG=lest[,16], COV1211_2221.LOG=lest[,17], COV2111_2211.LOG=lest[,18],
                  COV2111_2112.LOG=lest[,19], COV2111_2212.LOG=lest[,20], COV2111_2221.LOG=lest[,21],
                  COV2211_2112.LOG=lest[,22], COV2211_2212.LOG=lest[,23], COV2211_2221.LOG=lest[,24],
                  COV2112_2212.LOG=lest[,25], COV2112_2221.LOG=lest[,26], COV2212_2221.LOG=lest[,27])
  class(results) <- "CHR"
  return(results)
  
}

Try the DTR package in your browser

Any scripts or data that you put into this service are public.

DTR documentation built on May 2, 2019, 3:26 p.m.

R Package Documentation

rdrr.io home R language documentation Run R code online

Browse R Packages

CRAN packages Bioconductor packages R-Forge packages GitHub packages

We want your feedback!

Note that we can't provide technical support on individual packages. You should contact the package authors for that.
GitHub issue tracker
ian@mutexlabs.com
Personal blog

 
Embedding an R snippet on your website

Add the following code to your website.

For more information on customizing the embed code, read Embedding Snippets.

Close