survfit.formula: Compute a Survival Curve for Censored Data

Description

Computes an estimate of a survival curve for censored data. More multi-state data the Andersen-Johansen estimate is use, for ordinary survival either the Kaplan-Meier or Fleming-Harrington estimate is produced.

Usage

# S3 method for formula
survfit(formula, data, weights, subset, na.action,  
        etype, id, istate, timefix=TRUE, ...)

Arguments

formula

a formula object, which must have a Surv object as the response on the left of the ~ operator and, if desired, terms separated by + operators on the right. One of the terms may be a strata object. For a single survival curve the right hand side should be ~ 1.

data

a data frame in which to interpret the variables named in the formula, subset and weights arguments.

weights

The weights must be nonnegative and it is strongly recommended that they be strictly positive, since zero weights are ambiguous, compared to use of the subset argument.

subset

expression saying that only a subset of the rows of the data should be used in the fit.

na.action

a missing-data filter function, applied to the model frame, after any subset argument has been used. Default is options()$na.action.

etype

a variable giving the type of event. This has been superseded by multi-state Surv objects; see example below.

identifies individual subjects, when a given person can have multiple lines of data.

istate

for multi-state models, identifies the initial state of each subject

timefix

process times through the aeqSurv function to eliminate potential roundoff issues.

…

The following additional arguments are passed to internal functions called by survfit.

type: a character string specifying the type of survival curve. Possible values are "kaplan-meier", "fleming-harrington" or "fh2" if a formula is given. This is ignored for competing risks or when the Turnbull estimator is used.
error: a character string specifying the error. Possible values are "greenwood" for the Greenwood formula or "tsiatis" or "aalen" for the Tsiatis/Aalen formula, or "robust" for a robust variance. The last of these is assumed if non-integer case weights are provided.
conf.type: One of "none", "plain", "log" (the default), "log-log" or "logit". Only enough of the string to uniquely identify it is necessary. The first option causes confidence intervals not to be generated. The second causes the standard intervals curve +- k *se(curve), where k is determined from conf.int. The log option calculates intervals based on the cumulative hazard or log(survival). The log-log option bases the intervals on the log hazard or log(-log(survival)), and the logit option on log(survival/(1-survival)).
conf.lower: a character string to specify modified lower limits to the curve, the upper limit remains unchanged. Possible values are "usual" (unmodified), "peto", and "modified". The modified lower limit is based on an "effective n" argument. The confidence bands will agree with the usual calculation at each death time, but unlike the usual bands the confidence interval becomes wider at each censored observation. The extra width is obtained by multiplying the usual variance by a factor m/n, where n is the number currently at risk and m is the number at risk at the last death time. (The bands thus agree with the un-modified bands at each death time.) This is especially useful for survival curves with a long flat tail. The Peto lower limit is based on the same "effective n" argument as the modified limit, but also replaces the usual Greenwood variance term with a simple approximation. It is known to be conservative.
start.time: numeric value specifying a time to start calculating survival information. The resulting curve is the survival conditional on surviving to start.time.
conf.int: the level for a two-sided confidence interval on the survival curve(s). Default is 0.95.
se.fit: a logical value indicating whether standard errors should be computed. Default is TRUE.
influence: a logical value indicating whether to return the infinitesimal jackknife (influence) values for each subject. These contain the values of the derivative of each value with respect to the case weights of each subject i: $\partial p/\partial w_i$, evaluated at the vector of weights $w=1$. The array will have dimensions (number of subjects, 1+ number of unique times, number of states); be forewarned that this can be huge. If the total number of elements is larger then the maximum integer the underlying C program can not create it.

Value

an object of class "survfit". See survfit.object for details. Methods defined for survfit objects are print, plot, lines, and points.

Details

The estimates used are the Kalbfleisch-Prentice (Kalbfleisch and Prentice, 1980, p.86) and the Tsiatis/Link/Breslow, which reduce to the Kaplan-Meier and Fleming-Harrington estimates, respectively, when the weights are unity.

The Greenwood formula for the variance is a sum of terms d/(n*(n-m)), where d is the number of deaths at a given time point, n is the sum of weights for all individuals still at risk at that time, and m is the sum of weights for the deaths at that time. The justification is based on a binomial argument when weights are all equal to one; extension to the weighted case is ad hoc. Tsiatis (1981) proposes a sum of terms d/(n*n), based on a counting process argument which includes the weighted case.

The two variants of the F-H estimate have to do with how ties are handled. If there were 3 deaths out of 10 at risk, then the first increments the hazard by 3/10 and the second by 1/10 + 1/9 + 1/8. For the first method S(t) = exp(H), where H is the Nelson-Aalen cumulative hazard estimate, whereas the fh2 method will give results S(t) results closer to the Kaplan-Meier.

When the data set includes left censored or interval censored data (or both), then the EM approach of Turnbull is used to compute the overall curve. When the baseline method is the Kaplan-Meier, this is known to converge to the maximum likelihood estimate.

The cumulative incidence curve is an alternative to the Kaplan-Meier for competing risks data. For instance, in patients with MGUS, conversion to an overt plasma cell malignancy occurs at a nearly constant rate among those still alive. A Kaplan-Meier estimate, treating death due to other causes as censored, gives a 20 year cumulate rate of 33% for the 241 early patients of Kyle. This estimates the incidence of conversion if all other causes of death were removed, which is an unrealistic assumption given the mean starting age of 63 and a median follow up of over 21 years.

The CI estimate, on the other hand, estimates the total number of conversions that will actually occur. Because the population is older, this is much smaller than the KM, 22% at 20 years for Kyle's data. If there were no censoring, then CI(t) could very simply be computed as total number of patients with progression by time t divided by the sample size n.

References

Dorey, F. J. and Korn, E. L. (1987). Effective sample sizes for confidence intervals for survival probabilities. Statistics in Medicine 6, 679-87.

Fleming, T. H. and Harrington, D. P. (1984). Nonparametric estimation of the survival distribution in censored data. Comm. in Statistics 13, 2469-86.

Kalbfleisch, J. D. and Prentice, R. L. (1980). The Statistical Analysis of Failure Time Data. New York:Wiley.

Kyle, R. A. (1997). Moncolonal gammopathy of undetermined significance and solitary plasmacytoma. Implications for progression to overt multiple myeloma}, Hematology/Oncology Clinics N. Amer. 11, 71-87.

Link, C. L. (1984). Confidence intervals for the survival function using Cox's proportional hazards model with covariates. Biometrics 40, 601-610.

Turnbull, B. W. (1974). Nonparametric estimation of a survivorship function with doubly censored data. J Am Stat Assoc, 69, 169-173.

Examples

Run this code

# NOT RUN {
#fit a Kaplan-Meier and plot it 
fit <- survfit(Surv(time, status) ~ x, data = aml) 
plot(fit, lty = 2:3) 
legend(100, .8, c("Maintained", "Nonmaintained"), lty = 2:3) 

#fit a Cox proportional hazards model and plot the  
#predicted survival for a 60 year old 
fit <- coxph(Surv(futime, fustat) ~ age, data = ovarian) 
plot(survfit(fit, newdata=data.frame(age=60)),
     xscale=365.25, xlab = "Years", ylab="Survival") 

# Here is the data set from Turnbull
#  There are no interval censored subjects, only left-censored (status=3),
#  right-censored (status 0) and observed events (status 1)
#
#                             Time
#                         1    2   3   4
# Type of observation
#           death        12    6   2   3
#          losses         3    2   0   3
#      late entry         2    4   2   5
#
tdata <- data.frame(time  =c(1,1,1,2,2,2,3,3,3,4,4,4),
                    status=rep(c(1,0,2),4),
                    n     =c(12,3,2,6,2,4,2,0,2,3,3,5))
fit  <- survfit(Surv(time, time, status, type='interval') ~1, 
              data=tdata, weight=n)

#
# Time to progression/death for patients with monoclonal gammopathy
#  Competing risk curves (cumulative incidence)
fitKM <- survfit(Surv(stop, event=='pcm') ~1, data=mgus1,
                    subset=(start==0))

fitCI <- survfit(Surv(stop, status*as.numeric(event), type="mstate") ~1,
                    data=mgus1, subset=(start==0))

# CI curves are always plotted from 0 upwards, rather than 1 down
plot(fitCI, xscale=365.25, xmax=7300, mark.time=FALSE,
            col=2:3, xlab="Years post diagnosis of MGUS")
lines(fitKM, fun='event', xmax=7300, mark.time=FALSE,
            conf.int=FALSE)
text(10, .4, "Competing risk: death", col=3)
text(16, .15,"Competing risk: progression", col=2)
text(15, .30,"KM:prog")
# }

Run the code above in your browser using DataLab