get_survival_case_weigths_and_data: Static GLM fit for survival models

Description

Function used to get design matrix and weights for a static fit for survivals models where observations are binned into intervals

Usage

get_survival_case_weigths_and_data(formula, data, by, max_T, id, init_weights, risk_obj, use_weights = T, is_for_discrete_model = T, c_outcome = "Y", c_weights = "weights", c_end_t = "t")

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 co-variates

Length of each intervals that cases are binned into

max_T

The end time of the last bin

The id for each row in data. This is important when variables are time varying

init_weights

Weights for the rows data. Useful with skewed sampling and will be used when computing the final weights

risk_obj

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

use_weights

TRUE if weights should be used. See details

is_for_discrete_model

TRUE if the model is for a discrete hazard model like the logistic model. Affects how deaths are included when individuals have time varying coefficients

c_outcome, c_weights, c_end_t

Alternative names to use for the added columns described in the return section. Useful if you already have a column named Y, t or weights

Value

Returns a data frame with the design matrix from the formula where the following is added (column names will differ if you specified them): column Y for the binary outcome, column weights for weights of each row and additional rows if applicable. A column t is added for the stop time of the bin if use_weights = FALSE

Details

This function is used to get the data frame for e.g. a glm fit that is comparable to a ddhazard fit in the sense that it is a static version. For example, say that we bin our time periods into (0,1], (1,2] and (2,3]. Next, consider an individual who dies at time 2.5. He should be a control in the the first two bins and should be a case in the last bin. Thus the rows in the final data frame for this individual is c(Y = 1, ..., weights = 1) and c(Y = 0, ..., weights = 2) where Y is the outcome, ... is the co-variates and weights is the weights for the regression. Consider another individual who does not die and we observe him for all three periods. Thus, he will yield one row with c(Y = 0, ..., weights = 3)

This function use similar logic as the ddhazard for individuals with time varying co-variates (see the vignette "ddhazard" for details)

If use_weights = FALSE then the two individuals will yield three rows each. The first individual will have c(Y = 0, t = 1, ..., weights = 1), c(Y = 0, t = 2, ..., weights = 1), c(Y = 1, t = 3, ..., weights = 1) while the latter will have three rows c(Y = 0, t = 1, ..., weights = 1), c(Y = 0, t = 2, ..., weights = 1), c(Y = 0, t = 3, ..., weights = 1). This kind of data frame is useful if you want to make a fit with e.g. gam function in the mgcv package as described en Tutz et. al (2016) (see reference)

References