dynamichazard (version 1.0.2)

static_glm: Static glm Fit

Description

Method to fit a static model corresponding to a ddhazard fit. The method uses weights to ease the memory requirements. See get_survival_case_weights_and_data for details on weights.

The parallelglm_quick and parallelglm_QR methods are similar to two methods used in bam function in the mgcv package (see the `use.chol` argument or Wood et al. 2015). parallelglm_QR is more stable but slower. See Golub (2013) section 5.3 for a comparison of the Cholesky decomposition method and the QR method.

Usage

static_glm(
  formula,
  data,
  by,
  max_T,
  ...,
  id,
  family = "logit",
  model = FALSE,
  weights,
  risk_obj = NULL,
  speedglm = FALSE,
  only_coef = FALSE,
  mf,
  method_use = c("glm", "speedglm", "parallelglm_quick", "parallelglm_QR"),
  n_threads = getOption("ddhazard_max_threads")
)

Value

The returned list from the glm call or just coefficients depending on the value of only_coef.

Arguments

formula

coxph like formula with Surv(tstart, tstop, event) on the left hand site of ~.

data

data.frame or environment containing the outcome and covariates.

by

interval length of the bins in which parameters are fixed.

max_T

end of the last interval interval.

...

arguments passed to glm or speedglm. If only_coef = TRUE then the arguments are passed to glm.control if glm is used.

id

vector of ids for each row of the in the design matrix.

family

"logit", "cloglog", or "exponential" for a static equivalent model of ddhazard.

model

TRUE if you want to save the design matrix used in glm.

weights

weights to use if e.g. a skewed sample is used.

risk_obj

a pre-computed result from a get_risk_obj. Will be used to skip some computations.

speedglm

depreciated.

only_coef

TRUE if only coefficients should be returned. This will only call the speedglm::speedglm.wfit or glm.fit which will be faster.

mf

model matrix for regression. Needed when only_coef = TRUE

method_use

method to use for estimation. glm uses glm.fit, speedglm::speedglm uses speedglm::speedglm.wfit and parallelglm_quick and parallelglm_QR uses a parallel C++ estimation method.

n_threads

number of threads to use when method_use is "parallelglm".

References

Wood, S.N., Goude, Y. & Shaw S. (2015) Generalized additive models for large datasets. Journal of the Royal Statistical Society, Series C 64(1): 139-155.

Golub, G. H., & Van Loan, C. F. (2013). Matrix computations (4th ed.). JHU Press.

Examples

library(dynamichazard)
fit <- static_glm(
 Surv(time, status == 2) ~ log(bili), pbc, id = pbc$id, max_T = 3600,
 by = 50)
fit$coefficients