Fits a Cox proportional hazards regression model. Time dependent variables, time dependent strata, multiple events per subject, and other extensions are incorporated using the counting process formulation of Andersen and Gill.
coxph(formula, data=, weights, subset, na.action, init, control, ties=c("efron","breslow","exact"), singular.ok=TRUE, robust=FALSE, model=FALSE, x=FALSE, y=TRUE, tt, method, ...)
- a formula object, with the response on the left of a
~operator, and the terms on the right. The response must be a survival object as returned by the
- a data.frame in which to interpret the variables named in the
formula, or in the
- vector of case weights. If
weightsis a vector of integers, then the estimated coefficients are equivalent to estimating the model from data with the individual cases replicated as many times as indicated by
- expression indicating which subset of the rows of data should be used in the fit. All observations are included by default.
- a missing-data filter function. This is applied to the model.frame after any subset argument has been used. Default is
- vector of initial values of the iteration. Default initial value is zero for all variables.
- Object of class
coxph.controlspecifying iteration limit and other control options. Default is
- a character string specifying the method for tie handling. If there are no tied death times all the methods are equivalent. Nearly all Cox regression programs use the Breslow method by default, but not this one. The Efron approximation is used as the default here, it is more accurate when dealing with tied death times, and is as efficient computationally. The “exact partial likelihood” is equivalent to a conditional logistic model, and is appropriate when the times are a small set of discrete values. If there are a large number of ties and (start, stop) style survival data the computational time will be excessive.
- logical value indicating how to handle collinearity in the model matrix. If
TRUE, the program will automatically skip over columns of the X matrix that are linear combinations of earlier columns. In this case the coefficients for such columns will be NA, and the variance matrix will contain zeros. For ancillary calculations, such as the linear predictor, the missing coefficients are treated as zeros.
- this argument has been deprecated, use a cluster term in the model instead.
- logical value: if
TRUE, the model frame is returned in component
- logical value: if
TRUE, the x matrix is returned in component
- logical value: if
TRUE, the response vector is returned in component
- optional list of time-transform functions.
- alternate name for the
- Other arguments will be passed to
The proportional hazards model is usually expressed in terms of a single survival time value for each person, with possible censoring. Andersen and Gill reformulated the same problem as a counting process; as time marches onward we observe the events for a subject, rather like watching a Geiger counter. The data for a subject is presented as multiple rows or "observations", each of which applies to an interval of observation (start, stop].
The routine internally scales and centers data to avoid overflow in the argument to the exponential function. These actions do not change the result, but lead to more numerical stability. However, arguments to offset are not scaled since there are situations where a large offset value is a purposefully used. Users should not use normally allow large numeric offset values.
an object of class
coxph representing the fit. See
coxph.object for details.
Depending on the call, the
survfit routines may need to reconstruct the x matrix created by
coxph. It is possible for this to fail, as in the example below in which the predict function is unable to find
tfun <- function(tform) coxph(tform, data=lung) fit <- tfun(Surv(time, status) ~ age) predict(fit)
In such a case add the
model=TRUE option to the
coxph call to obviate the need for reconstruction, at the expense of a larger
There are three special terms that may be used in the model equation. A
strata term identifies a stratified Cox model; separate baseline hazard functions are fit for each strata. The
cluster term is used to compute a robust variance for the model. The term
+ cluster(id) where each value of
id is unique is equivalent to specifying the
robust=T argument, and produces an approximate jackknife estimate of the variance. If the
id variable were not unique, but instead identifies clusters of correlated observations, then the variance estimate is based on a grouped jackknife.
A time-transform term allows variables to vary dynamically in time. In this case the
tt argument will be a function or a list of functions (if there are more than one tt() term in the model) giving the appropriate transform. See the examples below.
In certain data cases the actual MLE estimate of a coefficient is infinity, e.g., a dichotomous variable where one of the groups has no events. When this happens the associated coefficient grows at a steady pace and a race condition will exist in the fitting routine: either the log likelihood converges, the information matrix becomes effectively singular, an argument to exp becomes too large for the computer hardware, or the maximum number of interactions is exceeded. (Nearly always the first occurs.) The routine attempts to detect when this has happened, not always successfully. The primary consequence for he user is that the Wald statistic = coefficient/se(coefficient) is not valid in this case and should be ignored; the likelihood ratio and score tests remain valid however.
coxph can now maximise a penalised partial likelihood with arbitrary user-defined penalty. Supplied penalty functions include ridge regression (ridge), smoothing splines (pspline), and frailty models (frailty).
Andersen, P. and Gill, R. (1982). Cox's regression model for counting processes, a large sample study. Annals of Statistics 10, 1100-1120.
Therneau, T., Grambsch, P., Modeling Survival Data: Extending the Cox Model. Springer-Verlag, 2000.
# Create the simplest test data set test1 <- list(time=c(4,3,1,1,2,2,3), status=c(1,1,1,0,1,1,0), x=c(0,2,1,1,1,0,0), sex=c(0,0,0,0,1,1,1)) # Fit a stratified model coxph(Surv(time, status) ~ x + strata(sex), test1) # Create a simple data set for a time-dependent model test2 <- list(start=c(1,2,5,2,1,7,3,4,8,8), stop=c(2,3,6,7,8,9,9,9,14,17), event=c(1,1,1,1,1,1,1,0,0,0), x=c(1,0,0,1,0,1,1,1,0,0)) summary(coxph(Surv(start, stop, event) ~ x, test2)) # # Create a simple data set for a time-dependent model # test2 <- list(start=c(1, 2, 5, 2, 1, 7, 3, 4, 8, 8), stop =c(2, 3, 6, 7, 8, 9, 9, 9,14,17), event=c(1, 1, 1, 1, 1, 1, 1, 0, 0, 0), x =c(1, 0, 0, 1, 0, 1, 1, 1, 0, 0) ) summary( coxph( Surv(start, stop, event) ~ x, test2)) # Fit a stratified model, clustered on patients bladder1 <- bladder[bladder$enum < 5, ] coxph(Surv(stop, event) ~ (rx + size + number) * strata(enum) + cluster(id), bladder1) # Fit a time transform model using current age coxph(Surv(time, status) ~ ph.ecog + tt(age), data=lung, tt=function(x,t,...) pspline(x + t/365.25))
Documentation reproduced from package survival, version 2.37-4. License: LGPL (>= 2)