Skip to contents

Survival outcomes

Source: vignettes/survival-outcomes.Rmd
survival-outcomes.Rmd

To illustrate Bayesian DCA on survival data, we load the example dataset dca_survival_data shown below.

library(bayesDCA)
head(dca_survival_data)
#>   outcomes.time outcomes.status model_predictions binary_test
#> 1     0.2948349       0.0000000        0.18806301           0
#> 2     1.8747960       0.0000000        0.25349061           1
#> 3     0.0355183       0.0000000        0.01025269           0
#> 4     1.2850960       0.0000000        0.04836840           0
#> 5     0.2398763       0.0000000        0.23502042           1
#> 6     2.0934150       0.0000000        0.04611183           0

The dataset contains the time-to-event outcomes (a survival::Surv object), the predictions from a prognostic model, and the results from a binary prognostic test. The time horizon for the event prediction is one time units (e.g., year), so we set prediction_time = 1. We also set chains = 1 to speed up MCMC sampling with Stan (in practice you should use at least chains = 4, maybe with cores = 4 for speed as well).

fit <- dca_surv(dca_survival_data, prediction_time = 1, iter = 1000, chains = 1)
plot(fit)

Interrogating the output

We can do all kinds of output interrogation just like with binary outcomes.

What is the best decision strategy?

Are the prediction model or the binary test useful at all? 

compare_dca(fit, type = "useful")

Is the model better than the test? 

compare_dca(fit, 
            strategies = c("model_predictions", "binary_test"), 
            type = "pairwise")

What is the price of uncertainty? 

compare_dca(fit, .evpi = TRUE)