Estimate a survival function under current status sampling

Usage

currstatCIR(
  time,
  event,
  X,
  SL_control = list(SL.library = c("SL.mean", "SL.glm"), V = 3),
  HAL_control = list(n_bins = c(5), grid_type = c("equal_mass"), V = 3),
  deriv_method = "m-spline",
  eval_region,
  n_eval_pts = 101,
  alpha = 0.05
)

Arguments

time: n x 1 numeric vector of observed monitoring times. For individuals that were never monitored, this can be set to any arbitrary value, including NA, as long as the corresponding event variable is NA.
event: n x 1 numeric vector of status indicators of whether an event was observed prior to the monitoring time. This value must be NA for individuals that were never monitored.
X: n x p dataframe of observed covariate values.
SL_control: List of SuperLearner control parameters. This should be a named list; see SuperLearner documentation for further information.
HAL_control: List of haldensify control parameters. This should be a named list; see haldensify documentation for further information.
deriv_method: Method for computing derivative. Options are "m-spline" (the default, fit a smoothing spline to the estimated function and differentiate the smooth approximation), "linear" (linearly interpolate the estimated function and use the slope of that line), and "line" (use the slope of the line connecting the endpoints of the estimated function).
eval_region: Region over which to estimate the survival function.
n_eval_pts: Number of points in grid on which to evaluate survival function. The points will be evenly spaced, on the quantile scale, between the endpoints of eval_region.
alpha: The level at which to compute confidence intervals and hypothesis tests. Defaults to 0.05

Value

Data frame giving results, with columns:

t: Time at which survival function is estimated
S_hat_est: Survival function estimate
S_hat_cil: Lower bound of confidence interval
S_hat_ciu: Upper bound of confidence interval

Examples

if (FALSE) # This is a small simulation example
set.seed(123)
n <- 300
x <- cbind(2*rbinom(n, size = 1, prob = 0.5)-1,
           2*rbinom(n, size = 1, prob = 0.5)-1)
t <- rweibull(n,
              shape = 0.75,
              scale = exp(0.4*x[,1] - 0.2*x[,2]))
y <- rweibull(n,
              shape = 0.75,
              scale = exp(0.4*x[,1] - 0.2*x[,2]))

# round y to nearest quantile of y, just so there aren't so many unique values
quants <- quantile(y, probs = seq(0, 1, by = 0.05), type = 1)
for (i in 1:length(y)){
  y[i] <- quants[which.min(abs(y[i] - quants))]
}
delta <- as.numeric(t <= y)

dat <- data.frame(y = y, delta = delta, x1 = x[,1], x2 = x[,2])

dat$delta[dat$y > 1.8] <- NA
dat$y[dat$y > 1.8] <- NA
eval_region <- c(0.05, 1.5)
res <- survML::currstatCIR(time = dat$y,
                           event = dat$delta,
                           X = dat[,3:4],
                           SL_control = list(SL.library = c("SL.mean", "SL.glm"),
                                             V = 3),
                           HAL_control = list(n_bins = c(5),
                                              grid_type = c("equal_mass"),
                                              V = 3),
                           eval_region = eval_region)
#> Warning: Some fit_control arguments are neither default nor glmnet/cv.glmnet arguments: n_folds; 
#> They will be removed from fit_control

xvals = res$t
yvals = res$S_hat_est
fn=stepfun(xvals, c(yvals[1], yvals))
plot.function(fn, from=min(xvals), to=max(xvals)) # \dontrun{}