Blend Parametric Survival Curves Using Model-Averaging Weights

Produces a single, weighted-average survival curve by combining predicted survival probabilities from the top-$k$ parametric models selected by compute_weights. The blend is a convex combination: $$\hat{S}(t) = \sum_{i=1}^{k} w_i \, S_i(t),$$ where $S_i(t)$ is the predicted survival from the $i$-th model and $w_i$ are the relative-likelihood weights.

Usage

blend_survival(models, weights_tbl, times)

Arguments

models: A named list of fitted flexsurvreg objects, typically from fit_models.
weights_tbl: A tibble from compute_weights with columns model (matching names in models) and weight.
times: Numeric vector of time points at which to evaluate the blended survival curve.

Value

A tibble with two columns:

time: Time points (same as times).
surv: Blended survival probability.

Examples

# \donttest{
t <- 0:5
s <- c(1, 0.549, 0.440, 0.394, 0.363, 0.341)
ipd  <- prep_ipd(t, s)
mods <- fit_models(ipd[-1, ])
ic   <- extract_ic(mods, "AIC")
wts  <- compute_weights(ic, top_k = 3, criterion = "AIC")
grid <- seq(0, 10, by = 0.25)
S_blend <- blend_survival(mods, wts, grid)
#> New names:
#> • `surv` -> `surv...1`
#> • `surv` -> `surv...2`
#> • `surv` -> `surv...3`
head(S_blend)
#> # A tibble: 6 × 2
#>    time  surv
#>   <dbl> <dbl>
#> 1  0    1    
#> 2  0.25 0.976
#> 3  0.5  0.951
#> 4  0.75 0.925
#> 5  1    0.901
#> 6  1.25 0.876
# }