Flexible Event Chart for Time-to-Event Data

Description

Creates an event chart on the current graphics device. Also, allows user to plot legend on plot area or on separate page. Contains features useful for plotting data with time-to-event outcomes Which arise in a variety of studies including randomized clinical trials and non-randomized cohort studies. This function can use as input a matrix or a data frame, although greater utility and ease of use will be seen with a data frame.

Usage

event.chart(data, subset.r = 1:dim(data)[1], subset.c = 1:dim(data)[2],
        sort.by = NA, sort.ascending =TRUE,
        sort.na.last =TRUE, sort.after.subset =TRUE,
        y.var = NA, y.var.type = 'n',
        y.jitter =FALSE, y.jitter.factor = 1,
        y.renum =FALSE, NA.rm =FALSE, x.reference = NA,
        now = max(data[,subset.c], na.rm =TRUE),
        now.line =FALSE, now.line.lty = 2,
        now.line.lwd = 1, now.line.col = 1, pty='m',
        date.orig = c(1,1,1960), titl = 'Event Chart',

        y.idlabels = NA, y.axis = 'auto',
        y.axis.custom.at = NA, y.axis.custom.labels = NA,
        y.julian =FALSE, y.lim.extend = c(0,0),
        y.lab = ifelse(is.na(y.idlabels), '' , as.character(y.idlabels)),

        x.axis.all =TRUE, x.axis = 'auto',
        x.axis.custom.at = NA, x.axis.custom.labels = NA,
        x.julian =FALSE, x.lim.extend = c(0,0), x.scale = 1,
        x.lab = ifelse(x.julian, 'Follow-up Time', 'Study Date'),

        line.by = NA, line.lty = 1, line.lwd = 1, line.col = 1,
        line.add = NA, line.add.lty = NA,
        line.add.lwd = NA, line.add.col = NA,
        point.pch = 1:length(subset.c),
        point.cex = rep(0.6,length(subset.c)),
        point.col = rep(1,length(subset.c)),

        legend.plot =FALSE, legend.location = 'o', legend.titl = titl,
        legend.titl.cex = 3.0, legend.titl.line = 1.0,
        legend.point.at = list(x = c(5,95), y = c(95,30)),
        legend.point.pch = point.pch,
        legend.point.text = ifelse(rep(is.data.frame(data),
              length(subset.c)), names(data[,subset.c]), subset.c),
        legend.cex = 2.5, legend.bty = 'n',
        legend.line.at = list(x = c(5,95), y = c(20,5)),
        legend.line.text = names(table(as.character(data[,line.by]),
              exclude = c('','NA'))),
        legend.line.lwd = line.lwd, legend.loc.num = 1,

        ...)

event.convert(data2, event.time = 1, event.code = 2)

Arguments

Details

if you want to put, say, two eventcharts side-by-side, in a plot region, you should not set up par(mfrow=c(1,2)) before running the first plot. Instead, you should add the argument mfg=c(1,1,1,2) to the first plot call followed by the argument mfg=c(1,2,1,2) to the second plot call.

if dates in original data frame are in a specialized form (eg., mm/dd/yy) of mode CHARACTER, the user must convert those columns to become class dates or julian numeric mode (see ?dates for more information). For example, in a data frame called testdata, with specialized dates in columns 4 thru 10, the following code could be used: as.numeric(dates(testdata[,4:10]). This will convert the columns to numeric julian dates based on the function's default origin of January 1, 1960. If original dates are in class dates or julian form, no extra work is necessary.

In the survival analysis, the data typically come in two columns: one column containing survival time and the other containing censoring indicator or event code. The event.convert function converts this type of data into multiple columns of event times, one column of each event type, suitable for the event.chart function.

Side Effects

an event chart is created on the current graphics device. If legend.plot =TRUE and legend.location = 'o', a one-page legend will precede the event chart. Please note that par parameters on completion of function will be reset to par parameters existing prior to start of function.

Author(s)

J. Jack Lee and Kenneth R. Hess
Department of Biostatistics
University of Texas
M.D. Anderson Cancer Center
Houston, TX 77030
jjlee@mdanderson.org, khess@mdanderson.org

Joel A. Dubin
Division of Biostatistics
Department of Epidemiology and Public Health
Yale University
joel.dubin@yale.edu

References

Lee J.J., Hess, K.R., Dubin, J.A. (2000). Extensions and applications of event charts. The American Statistician, 54:1, 63–70.

Dubin, J.A., Lee, J.J., Hess, K.R. (1997). The Utility of Event Charts. Proceedings of the Biometrics Section, American Statistical Association.

Dubin, J.A., Muller H-G, Wang J-L (2001). Event history graphs for censored survival data. Statistics in Medicine, 20: 2951–2964.

Goldman, A.I. (1992). EVENTCHARTS: Visualizing Survival and Other Timed-Events Data. The American Statistician, 46:1, 13–18.

See Also

event.history

Examples

# The sample data set is an augmented CDC AIDS dataset (ASCII)
# which is used in the examples in the help file.  This dataset is 
# described in Kalbfleisch and Lawless (JASA, 1989).
# Here, we have included only children 4 years old and younger.
# We have also added a new field, dethdate, which
# represents a fictitious death date for each patient.  There was
# no recording of death date on the original dataset.
#   
# All dates are julian with julian=0 being 
# January 1, 1960, and julian=14000 being 14000 days beyond
# January 1, 1960 (i.e., May 1, 1998).

cdcaids <- data.frame(
age=c(4,2,1,1,2,2,2,4,2,1,1,3,2,1,3,2,1,2,4,2,2,1,4,2,4,1,4,2,1,1,3,3,1,3),
infedate=c(
7274,7727,7949,8037,7765,8096,8186,7520,8522,8609,8524,8213,8455,8739,
8034,8646,8886,8549,8068,8682,8612,9007,8461,8888,8096,9192,9107,9001,
9344,9155,8800,8519,9282,8673),
diagdate=c(
8100,8158,8251,8343,8463,8489,8554,8644,8713,8733,8854,8855,8863,8983,
9035,9037,9132,9164,9186,9221,9224,9252,9274,9404,9405,9433,9434,9470,
9470,9472,9489,9500,9585,9649),
diffdate=c(
826,431,302,306,698,393,368,1124,191,124,330,642,408,244,1001,391,246,
615,1118,539,612,245,813,516,1309,241,327,469,126,317,689,981,303,976),
dethdate=c(
8434,8304,NA,8414,8715,NA,8667,9142,8731,8750,8963,9120,9005,9028,9445,
9180,9189,9406,9711,9453,9465,9289,9640,9608,10010,9488,9523,9633,9667,
9547,9755,NA,9686,10084),
censdate=c(
NA,NA,8321,NA,NA,8519,NA,NA,NA,NA,NA,NA,NA,NA,NA,NA,NA,NA,NA,NA,NA,NA,
NA,NA,NA,NA,NA,NA,NA,NA,NA,10095,NA,NA))

cdcaids <- upData(cdcaids,
 labels=c(age     ='Age, y', infedate='Date of blood transfusion',
          diagdate='Date of AIDS diagnosis',
          diffdate='Incubation period (days from HIV to AIDS)',
          dethdate='Fictitious date of death',
          censdate='Fictitious censoring date'))

# Note that the style options listed with these
# examples are best suited for output to a postscript file (i.e., using
# the postscript function with horizontal=TRUE) as opposed to a graphical
# window (e.g., motif).

# To produce simple calendar event chart (with internal legend):
# postscript('example1.ps', horizontal=TRUE)
 event.chart(cdcaids,
  subset.c=c('infedate','diagdate','dethdate','censdate'),
  x.lab = 'observation dates',
  y.lab='patients (sorted by AIDS diagnosis date)',
  titl='AIDS data calendar event chart 1',
  point.pch=c(1,2,15,0), point.cex=c(1,1,0.8,0.8),
  legend.plot=TRUE, legend.location='i', legend.cex=1.0,
  legend.point.text=c('transfusion','AIDS diagnosis','death','censored'),
  legend.point.at = list(c(7210, 8100), c(35, 27)), legend.bty='o')

# To produce simple interval event chart (with internal legend):
# postscript('example2.ps', horizontal=TRUE)
 event.chart(cdcaids,
  subset.c=c('infedate','diagdate','dethdate','censdate'),
  x.lab = 'time since transfusion (in days)',
  y.lab='patients (sorted by AIDS diagnosis date)',
  titl='AIDS data interval event chart 1',
  point.pch=c(1,2,15,0), point.cex=c(1,1,0.8,0.8),
  legend.plot=TRUE, legend.location='i', legend.cex=1.0,
  legend.point.text=c('transfusion','AIDS diagnosis','death','censored'),
  x.reference='infedate', x.julian=TRUE,
  legend.bty='o', legend.point.at = list(c(1400, 1950), c(7, -1)))

# To produce more complicated interval chart which is
# referenced by infection date, and sorted by age and incubation period:
# postscript('example3.ps', horizontal=TRUE)
 event.chart(cdcaids,
  subset.c=c('infedate','diagdate','dethdate','censdate'),
  x.lab = 'time since diagnosis of AIDS (in days)',
  y.lab='patients (sorted by age and incubation length)',
  titl='AIDS data interval event chart 2 (sorted by age, incubation)',
  point.pch=c(1,2,15,0), point.cex=c(1,1,0.8,0.8),
  legend.plot=TRUE, legend.location='i',legend.cex=1.0,
  legend.point.text=c('transfusion','AIDS diagnosis','death','censored'),
  x.reference='diagdate', x.julian=TRUE, sort.by=c('age','diffdate'),
  line.by='age', line.lty=c(1,3,2,4), line.lwd=rep(1,4), line.col=rep(1,4),
  legend.bty='o', legend.point.at = list(c(-1350, -800), c(7, -1)),
  legend.line.at = list(c(-1350, -800), c(16, 8)),
  legend.line.text=c('age = 1', '       = 2', '       = 3', '       = 4'))

# To produce the Goldman chart:
# postscript('example4.ps', horizontal=TRUE)
 event.chart(cdcaids,
  subset.c=c('infedate','diagdate','dethdate','censdate'),
  x.lab = 'time since transfusion (in days)', y.lab='dates of observation',
  titl='AIDS data Goldman event chart 1',
  y.var = c('infedate'), y.var.type='d', now.line=TRUE, y.jitter=FALSE,
  point.pch=c(1,2,15,0), point.cex=c(1,1,0.8,0.8), mgp = c(3.1,1.6,0),
  legend.plot=TRUE, legend.location='i',legend.cex=1.0,
  legend.point.text=c('transfusion','AIDS diagnosis','death','censored'),
  x.reference='infedate', x.julian=TRUE,
  legend.bty='o', legend.point.at = list(c(1500, 2800), c(9300, 10000)))

# To convert coded time-to-event data, then, draw an event chart:
surv.time <- c(5,6,3,1,2)
cens.ind   <- c(1,0,1,1,0)
surv.data  <- cbind(surv.time,cens.ind)
event.data <- event.convert(surv.data)
event.chart(cbind(rep(0,5),event.data),x.julian=TRUE,x.reference=1)