spsurv (version 1.1.0)

spbp.default: spbp: The BP Based Semiparametric Survival Analysis Function

Description

spbp: The BP Based Semiparametric Survival Analysis Function

Usage

# S3 method for default
spbp(
  formula,
  degree,
  data,
  approach = c("mle", "bayes"),
  model = c("ph", "po", "aft"),
  priors = list(beta = c("normal(0,2)"), gamma = c("lognormal(0,4)"), frailty =
    c("gamma(0.01,0.01)")),
  cores = .spbp_default_cores(),
  scale = TRUE,
  dist = NULL,
  baseline = NULL,
  verbose = FALSE,
  chains = 4,
  ...
)

Value

An object of class spbp. Component degree records the Bernstein polynomial degree used in the fit (also stored in call$degree).

Arguments

formula

a Surv object with time to event, status and explanatory terms

degree

Bernstein polynomial degree (integer). If omitted and neither dist nor baseline supplies bernstein(m), the default is ceiling(sqrt(n)) where n is the number of rows in data.

data

a data.frame object

approach

Bayesian or Maximum Likelihood estimation methods; default is "mle"

model

Bernstein PH ("ph"), PO ("po"), or AFT ("aft") model; default is "ph"

priors

prior settings for the Bayesian approach; `normal` or `cauchy` for beta; `lognormal` or `loglogistic` for gamma (BP coefficients). Defaults are normal(0,2) for standardised regression coefficients and lognormal(0,4) for Bernstein coefficients.

cores

number of core threads to use (Bayes sampling)

scale

logical; indicates whether to center and scale the data

dist

optional baseline specification; use bernstein(m) for the Bernstein polynomial degree

baseline

optional alias for dist

verbose

passed to Stan

chains

number of MCMC chains (Bayes)

...

further arguments passed to rstan::optimizing (MLE) or rstan::sampling (Bayes), e.g. iter, warmup, init.

Details

Right-censored survival data are modeled with a Bernstein-polynomial baseline and regression on covariates. With approach = "mle", parameters are estimated by Stan's optimizer and approximate inference uses the Hessian when available. With approach = "bayes", posterior samples are drawn with NUTS; use summary, tidy.spbp, and glance.spbp for output. Covariates are centered and scaled when scale = TRUE (default).

The returned object includes:

coefficients

Regression estimates on the original covariate scale.

bp.param

Bernstein baseline coefficients (gamma).

degree

Polynomial degree used (also in call$degree).

loglik

MLE: intercept-only and full-model log-likelihoods; Bayes: posterior mean pointwise log-likelihoods.

call

Matched call with approach, model, and degree.

See Also

bpph, bppo, bpaft, bernstein, summary.spbp