Fit a Panel of Parametric Survival Models to Pseudo-IPD

Fits each distribution named in dists to a pseudo-individual patient data (pseudo-IPD) data frame using flexsurvreg with observation weights. The resulting named list is designed to be passed directly to extract_ic and blend_survival.

Usage

fit_models(
  ipd,
  dists = c("exp", "weibull", "gompertz", "llogis", "lnorm", "gamma")
)

Arguments

ipd

A tibble / data frame with at least three columns:

time: Event or censoring time (numeric, > 0).
event: Event indicator (1 = event, 0 = censored).
weight: Observation weight (non-negative numeric).

Typically the output of prep_ipd with the first row (time = 0) removed.

dists

Character vector of distribution names accepted by flexsurvreg. Defaults to the six-distribution set used throughout seerSurv: c("exp", "weibull", "gompertz", "llogis", "lnorm", "gamma").

Value

A named list of flexsurvreg objects, one per distribution. Names match the corresponding element of dists.

Details

The ipd data frame should have the first row (time = 0) removed before calling this function because flexsurvreg requires strictly positive event times. A common idiom is:


  ipd  <- prep_ipd(time_vec, surv_vec)
  mods <- fit_models(ipd[-1, ])

Examples

# \donttest{
t <- 0:5
s <- c(1, 0.549, 0.440, 0.394, 0.363, 0.341)   # melanoma DR (SEER-17)
ipd   <- prep_ipd(t, s)
mods  <- fit_models(ipd[-1, ])
names(mods)
#> [1] "exp"      "weibull"  "gompertz" "llogis"   "lnorm"    "gamma"   
# }