AIC.curesurv {curesurv} | R Documentation |
Akaike's An Information Criterion for cure models
Description
Calculates the Akaike's "An Information Criterion" for fitted models from curesurv
Usage
## S3 method for class 'curesurv'
AIC(object, ..., k = 2)
Arguments
object |
a fitted model object obtained from |
... |
optionally more fitted model objects obtained from |
k |
numeric, the penalty per parameter to be used; the default k = 2 is the classical AIC. |
Details
When comparing models fitted by maximum likelihood to the same data, the smaller the AIC, the better the fit.
However in our case, one should be careful when comparing the AIC. Specifically, when one implements a mixture cure model with curesurv
without correcting the rate table (pophaz.alpha=FALSE
), one is not obligated to specify cumpophaz
. However, you cannot compare a model where cumpophaz
is not specified with a model where cumpophaz
is specified. If one wants to compare different models using AIC, one should always specify cumpophaz
when
using the curesurv
function.
Value
the value corresponds to the AIC calculated from the log-likelihood of the fitted model if just one object is provided. If multiple objects are provided, a data.frame with columns corresponding to the objects and row representing the AIC
Examples
library("curesurv")
library("survival")
testiscancer$age_crmin <- (testiscancer$age- min(testiscancer$age)) /
sd(testiscancer$age)
fit_m1_ad_tneh <- curesurv(Surv(time_obs, event) ~ z_tau(age_crmin) +
z_alpha(age_crmin),
pophaz = "ehazard",
cumpophaz = "cumehazard",
model = "nmixture", dist = "tneh",
link_tau = "linear",
data = testiscancer,
method_opt = "L-BFGS-B")
AIC(fit_m1_ad_tneh)