Cox proportional hazards regression. Equivalent to R's survival::coxph(). As
coxph(Surv(time, status) ~ x [+ strata(strata)], weights, ties, init, robust, cluster, control) for right-censored
data, with the values of summary(fit).
With robust (by default when cluster is given or a case weight is not an integer) the variance is the robust
(Lin-Wei sandwich) one, summed over the clusters (each observation its own cluster without cluster), as R:
var is the robust variance and naiveVar the model-based one; se (and stdError) stay the model-based standard
errors (R's se(coef)), while z, p, the intervals lower/upper, wald and waldtest use the robust ones
(robustSe); robscore is the robust score test at init. The likelihood ratio and score tests and the R^2 ignore
the clustering.
Rows with a missing time, status, covariate, weight or stratum are dropped. Times within sqrt(eps) of each other are
treated as tied. Covariates whose values all lie in nocenter are not centred; the others are centred by their
weighted means, which only shifts the linear predictors. R's warnings (non-convergence, possibly infinite
coefficients) are returned in warnings.
Parameters
time: readonlynumber[]
Follow-up times.
status: readonlynumber[]
1 for an event, 0 for a censored time.
x: readonly (readonlynumber[])[]
The covariate matrix, one row per observation (at least one column).
TypeError if robust is not a boolean. RangeError for invalid arguments, among them an infinite weight
and a robust variance with fewer than two clusters.
Cox proportional hazards regression. Equivalent to R's
survival::coxph(). Ascoxph(Surv(time, status) ~ x [+ strata(strata)], weights, ties, init, robust, cluster, control)for right-censored data, with the values ofsummary(fit).With
robust(by default whenclusteris given or a case weight is not an integer) the variance is the robust (Lin-Wei sandwich) one, summed over the clusters (each observation its own cluster withoutcluster), as R:varis the robust variance andnaiveVarthe model-based one;se(andstdError) stay the model-based standard errors (R'sse(coef)), whilez,p, the intervalslower/upper,waldandwaldtestuse the robust ones (robustSe);robscoreis the robust score test atinit. The likelihood ratio and score tests and the R^2 ignore the clustering.Rows with a missing time, status, covariate, weight or stratum are dropped. Times within sqrt(eps) of each other are treated as tied. Covariates whose values all lie in
nocenterare not centred; the others are centred by their weighted means, which only shifts the linear predictors. R's warnings (non-convergence, possibly infinite coefficients) are returned inwarnings.