![](/img/trans.png)
[英]R :Fitting survival trees with time-varying covariates in RandomForestSRC
[英]Fitting a fully parametric proportional hazard model with time-varying covariates in R
我需要使用隨時間變化的協變量來擬合參數化PH模型(因此,不是Cox模型)。 我們可以在R中做到嗎? 我聽說survreg函數無法處理時變協變量。 我徒勞地尋找可以解決這個問題的軟件包。
正如@adibender所寫,您可以輕松地估計poisson
族具有恆定基線且對數時間offset
為glm
。 這是一個例子
> # Input parameters
> n <- 100 # Number of individuals
> t_max <- 5 # max number of period per individual
> beta <- c(-1, 1) # true coefficient
>
> # Simulate data
> set.seed(47261114)
> sim_dat <- replicate(
+ n,
+ {
+ out <- data.frame(
+ tstart = rep(NA_integer_, t_max),
+ tstop = rep(NA_integer_, t_max),
+ event = rep(NA, t_max),
+ x = rnorm(t_max))
+
+ for(i in 1:t_max){
+ rate <- exp(beta %*% c(1, out$x[i]))
+ tstop <- min(rexp(1, rate), 1)
+ out[i, ] <- list(i - 1, i - (1 - tstop), tstop < 1, out$x[i])
+ if(out$event[i])
+ break
+ }
+ out[!is.na(out$tstart), ]
+ }, simplify = FALSE)
>
> sim_dat <- do.call(rbind, sim_dat)
> head(sim_dat) # show final data
tstart tstop event x
1 0 0.3018182 TRUE 0.7095841
2 0 0.6724803 TRUE 1.5152877
3 0 1.0000000 FALSE 0.1036868
4 1 2.0000000 FALSE -0.5214508
5 2 2.4831577 TRUE 1.0101403
6 0 1.0000000 FALSE 0.1437594
>
> # Fit with glm
> glm(event ~ x + offset(log(tstop - tstart)), sim_dat, family = poisson())
Call: glm(formula = event ~ x + offset(log(tstop - tstart)), family = poisson(),
data = sim_dat)
Coefficients:
(Intercept) x
-0.9053 0.9714
Degrees of Freedom: 248 Total (i.e. Null); 247 Residual
Null Deviance: 382.5
Residual Deviance: 306.4 AIC: 498.4
聲明:本站的技術帖子網頁,遵循CC BY-SA 4.0協議,如果您需要轉載,請注明本站網址或者原文地址。任何問題請咨詢:yoyou2525@163.com.