R/initial.surv.R

initial.surv <-
function (Time, d, W, WintF.vl, WintF.sl, id, times, method, 
        parameterization, long = NULL, long.deriv = NULL, extra = NULL, LongFormat) {
    old <- options(warn = (-1))
    on.exit(options(old))
    if (!is.null(long)) {
        long.id <- tapply(long, id, tail, 1)
        if (parameterization == "value") 
            longD.id <- NULL
    }
    if (!is.null(long.deriv)) {
        longD.id <- tapply(long.deriv, id, tail, 1)
        if (parameterization == "slope") 
            long.id <- NULL
    }
    idT <- extra$ii
    WW <- if (!LongFormat) {
        cbind(W, long.id, longD.id)
    } else {
        cbind(W, long.id[idT], longD.id[idT])
    }
    if (method %in% c("Cox-PH-GH", "weibull-PH-GH", "piecewise-PH-GH", 
            "spline-PH-GH", "spline-PH-Laplace")) {
        if (!LongFormat) {
            DD <- data.frame(id = id, Time = Time[id], d = d[id], times = times)
            if (!is.null(long)) {
                DD$long <- long * WintF.vl[id, , drop = FALSE]
                k <- ncol(DD$long)
            }
            if (!is.null(long.deriv)) {
                DD$longD <- long.deriv * WintF.sl[id, , drop = FALSE]
                l <- ncol(DD$longD)
            }
            dW <- as.data.frame(W[id, , drop = FALSE], row.names = row.names(DD))
            if (ncol(dW)) {
                names(dW) <- paste("W", seq_along(dW), sep = "")
                DD <- cbind(DD, dW)
            }
        } else {
            DD <- data.frame(Time = Time, d = d)
            if (!is.null(long)) {
                DD$long <- as.vector(long.id[idT]) * WintF.vl
                k <- ncol(DD$long)
            }
            if (!is.null(long.deriv)) {
                DD$longD <- as.vector(longD.id[idT]) * WintF.sl
                l <- ncol(DD$longD)
            }
            dW <- as.data.frame(W, row.names = row.names(DD))
            if (ncol(dW)) {
                names(dW) <- paste("W", seq_along(dW), sep = "")
                DD <- cbind(DD, dW)
            }
            DD$strata <- extra$strata
        }
        if (!LongFormat) {
            DD$start <- DD$times
            DD$stop <- unlist(lapply(split(DD[c("id", "start", "Time")], DD$id), 
                function (d) c(d$start[-1], d$Time[1])))
            DD$event <- ave(DD$d, DD$id, FUN = function(x) {
                if (length(x) == 1) {
                    x
                } else {
                    x[seq(length(x) - 1)] <- 0
                    x
                }
            })
        }
        baseCovs <- if (ncol(dW)) {
            paste("+", paste(names(dW), collapse = " + "))
        } else 
            NULL
        form <- if (!LongFormat) {
            switch(parameterization,
                "value" = paste("Surv(start, stop, event) ~", "long", baseCovs),
                "slope" = paste("Surv(start, stop, event) ~", "longD", baseCovs),
                "both" = paste("Surv(start, stop, event) ~", "long + longD", baseCovs))
        } else {
            switch(parameterization,
                "value" = paste("Surv(Time, d) ~", "long", baseCovs),
                "slope" = paste("Surv(Time, d) ~", "longD", baseCovs),
                "both" = paste("Surv(Time, d) ~", "long + longD", baseCovs))
        }
        if (!is.null(DD$strata))
            form <- paste(form, "+ strata(strata)")
        form <- as.formula(form)
        cph <- coxph(form, data = DD)
        coefs <- cph$coefficients
        out <- switch(parameterization,
            "value" = list(alpha = coefs[1:k], gammas = coefs[-(1:k)]), 
            "slope" = list(Dalpha = coefs[1:l], gammas = coefs[-(1:l)]),
            "both" = list(alpha = coefs[1:k], Dalpha = coefs[(k+1):(k+l)], 
                gammas = coefs[-(1:(k+l))])
        )
        if (method == "Cox-PH-GH") {
            out$lambda0 <- basehaz(cph, FALSE)$hazard
        }
        if (method == "weibull-PH-GH") {
            dat <- data.frame(Time = Time, d = d)
            init.fit <- survreg(Surv(Time, d) ~ WW, data = dat)
            coefs <- - init.fit$coefficients / init.fit$scale
            out$gammas <- c(coefs[1], out$gammas)
            out$sigma.t <- 1 / init.fit$scale
        }
        if (method == "piecewise-PH-GH") {
            dat <- data.frame(Time = Time, d = d)
            cph. <- coxph(Surv(Time, d) ~ WW, data = dat, x = TRUE)
            init.fit <- piecewiseExp.ph(cph., knots = extra$control$knots)
            coefs <- init.fit$coefficients
            out$xi <- exp(coefs[grep("xi", names(coefs))])
        }
        if (method == "spline-PH-GH" || method == "spline-PH-Laplace") {
            if (is.null(extra$strata)) {
                dat <- data.frame(Time = Time, d = d, as.data.frame(WW))
                rn <- tapply(row.names(dat), idT, tail, 1)
                ind <- row.names(dat) %in% rn
                dat <- dat[ind, ]
                init.fit <- survreg(Surv(Time, d) ~ ., data = dat)
                coefs <- init.fit$coefficients
                xi <- 1 / init.fit$scale
                phi <- exp(coefs[1])
                logh <- -log(phi * xi * dat$Time^(xi - 1))
                out$gammas.bs <- as.vector(lm.fit(extra$W2[ind, ], logh)$coefficients)
            } else {
                dat <- data.frame(Time = Time, d = d)
                dat <- cbind(dat, as.data.frame(WW))
                strata <- extra$strata
                split.dat <- split(dat, strata)
                gg <- NULL
                for (i in seq_along(split.dat)) {
                    ii <- strata == levels(strata)[i]
                    SpD.i <- split.dat[[i]]
                    idT.i <- idT[ii]
                    W2.i <- extra$W2[ii, ]
                    rn <- tapply(row.names(SpD.i), idT.i, tail, 1)
                    ind <- row.names(SpD.i) %in% rn
                    SpD.i <- SpD.i[ind, ]
                    init.fit <- survreg(Surv(Time, d) ~ ., data = SpD.i)
                    coefs <- init.fit$coefficients
                    xi <- 1 / init.fit$scale
                    phi <- exp(coefs[1])
                    logh <- -log(phi * xi * SpD.i$Time^(xi - 1))
                    gg <- c(gg, as.vector(lm.fit(W2.i[ind, ], logh)$coefficients))
                }
                out$gammas.bs <- gg[!is.na(gg)]
            }
            out
        }
    }
    if (method == "weibull-AFT-GH") {
        dat <- data.frame(Time = Time, d = d)
        if (!is.null(long.id)) {
            long.id <- c(long.id) * WintF.vl
            k <- ncol(WintF.vl)
        }
        if (!is.null(longD.id)) {
            longD.id <- c(longD.id) * WintF.sl
            l <- ncol(WintF.sl)
        }         
        WW <- cbind(W, long.id, longD.id)
        init.fit <- survreg(Surv(Time, d) ~ WW, data = dat)
        coefs <- - init.fit$coef
        nk <- if (is.null(W)) 1 else ncol(W) + 1
        out <- switch(parameterization, 
            "value" = list(gammas = coefs[1:nk], alpha = coefs[-(1:nk)], sigma.t = 1 / init.fit$scale),
            "slope" = list(gammas = coefs[1:nk], Dalpha = coefs[-(1:nk)], sigma.t = 1 / init.fit$scale),
            "both" = list(gammas = coefs[1:nk], alpha = coefs[seq(nk+1, nk+k)], 
                Dalpha = coefs[-seq(1, nk+k)], sigma.t = 1 / init.fit$scale)
        )
    }
    if (method == "ch-Laplace") {
        dat <- data.frame(Time = Time, d = d)
        init.fit <- survreg(Surv(Time, d) ~ WW, data = dat)
        coefs <- - coef(init.fit) / init.fit$scale
        min.x <- min(logT)
        max.x <- max(logT)
        kn <- if (is.null(extra$control$knots)) {
            kk <- seq(0, 1, length.out = extra$control$lng.in.kn + 2)[-c(1, extra$control$lng.in.kn + 2)]
            quantile(log(Time)[d == 1], kk, names = FALSE)
        } else {
            extra$control$knots
        }
        kn <- sort(c(rep(c(min.x, max.x), extra$control$ord), kn))
        W <- splineDesign(kn, log(Time), ord = extra$control$ord)
        nk <- ncol(W)
        nx <- NCOL(X)
        logH <- coefs[1] + logT / init.fit$scale
        coefs <- c(as.vector(lm.fit(W, logH)$coefficients), coefs[-1])
        out <- list(gammas = c(sort(coefs[1:nk]), if (nx > 1) coefs[seq(nk + 1, nk + nx - 1)] else NULL), 
            alpha = coefs[nk + nx])
    }
    out
}

Try the JM package in your browser

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

JM documentation built on Aug. 8, 2022, 5:09 p.m.