{smcl}
{* 23dec2014}{...}
{title:Title}
{p2colset 5 19 21 2}{...}
{p2col :{hi:artsurvdlg} {hline 2}}ART (Survival Outcomes) - Sample Size and Power dialog{p_end}
{p2colreset}{...}
{title:General remarks}
{p 4 4}
This dialog acts as an easy-to-use "front end" to the Stata program {cmd:artsurv}.
You should be aware,
however, that {cmd:artsurv} can do more than the dialog offers. For example, there is
a facility to save survival probabilities and hazard ratios to a new file for plotting,
and there are some additional options that are not accessible from the dialog. All of
these are described in the help file on {help artsurv}. It is easy to use the dialog
to create the equivalent {cmd:artsurv} command and then edit the command and re-run
it. This allows the creation of command files (through logging of output), and also
gives access to the extra options just mentioned.
{title:Panel 1: Basic set-up and options}
{p 0 4}
{cmd:Number of periods} (default 1) The number of notional time periods of equal
length over which the trial is to be run, i.e. the duration of
the trial in unspecified units. The default is 1.
Typically the length of the trial may exceed the
duration of recruitment. Patients may be followed up
after recruitment is complete and before definitive
analysis of the data. Then more than one period is
needed, e.g. recuitment for one year and follow-up for
two further years would require you to specify the
number of periods to be at least 3. This option also
allows specification of details such as hazard ratios
which vary with time and other advanced features.
The choice of how long in real time one period lasts is
up to you and will be chosen in line with the
anticipated characteristics of the study.
{p 4 4}
It may be wise to select a finer time-scale than the one that
appears natural at first sight. For example, most cancer trials
are planned on a scale of years, but a scale of quarters
or even months may have advantages; for example, it allows
greater detail on the projected survival
curves in each group (given in the output from {cmd:artsurv}).
It may be useful later in projections of patients, power
and events in "what if?" calculations (see for example
help on {help artpep}). See {cmd:Time unit} for further
details of available time-scales.
{p 4 4}
Note that once you have selected a time-scale and number of
periods, several items in the design depend on the choices, and
changing the time-scale must be done with care since it is easy
inadvertently to introduce errors. It is better to plan
with a finer time-scale in the first place.
{p_end}
{p 0 4}
{cmd:Number of groups} (default 2, max. 6) Number of arms in the clinical trial.
{p_end}
{p 0 4}
{cmd:Time unit} (default 1) The unit of time representing one period. The options
are given in the list box (default one year). The time unit
selected does not alter the calculations but is used
to label the output.
{p_end}
{p 0 4}
{cmd:Alpha} (default 0.05, two-sided) The Type 1 error probability for the trial. One-sided
alpha values may be imposed by checking the "One-sided alpha" box.
With one-sided alpha the significance level
used by the program is doubled, resulting in a larger power or
smaller sample size. This option should be used with caution.
{p_end}
{p 0 4}
{cmd:Median survival time} Checking this box allows you to enter the time for 50%
of patients in group 1 to have experienced an event. Time
is in the same units as the periods (see "Time unit" above).
Non-integer times are allowed. Checking this box
prevents data on cumulative survival or failure probabilities
being entered. These probabilities are calculated from the
supplied median survival time, assuming an exponential survival
distribution in group 1.{break}
{p 4 4}
Note that if you know the median survival time in group 2 (or 3, or 4, ...), you
can calculate the hazard ratio (HR) as HR = (median in group 1)/
(median in group 2) and enter this value in panel 2.
The calculation assumes exponential survival distributions in each
group and must be done manually. There is currently no facility
for entering median survival times for groups other than group 1.
{p_end}
{p 0 4}
{cmd:Power or N} (default power 0.8) The program can be used to calculate the sample size
for a given power (the default), or the power for a given sample size
(specified by checking the radio button "Specify sample size" in the "Options" area).
{p_end}
{p 0 4}
{cmd:Baseline survival or failure probabilities} The baseline cumulative probability
of survival or of failure at the end of each period,
with "period" as just
defined. Typically, this will be the distribution in
the control arm of the trial. In the simplest case,
you give just one value, which is taken as the
cumulative survival or failure probability at the end of the
trial. To specify values e.g. for periods 1 and 2, say
of 0.8 and 0.5 respectively, enter as {cmd:1 2} in
{cmd:At the end of period(s)}. Values for any subset of periods may be specified.
The radio buttons in the "Options" area allow you to specify
whether the values you have entered are survival
probabilities (the default) or failure probabilities.
{p_end}
{p 0 4}
{cmd:Non-inferiority design} In a non-inferiority design one wishes to test whether the effects of the
experimental treatment is not inferior to the control treatment by more than a
a prespecified amount. In the calculations the roles of the null
and alternative (alternate) hypotheses are reversed. That is,
the sample size is calculated with signficance level equal to
1-power and power equal to 1-alpha. By default, a two-sided alpha
is used. In many cases the preferred approach is to set a one-sided
alpha level, and then the {cmd:One-sided alpha} box should be ticked.
A side-effect of the reversal of power and alpha is that
the program is not able to compute the power of a non-inferiority
design for a given sample size. However, the power can still be determined
by trial and error, by repeatedly entering alpha and power until the desired
sample size is achieved.
{p_end}
{p 0 4}
{cmd:One-sided alpha} (default two-sided) See {cmd:Alpha} above.
{p_end}
{title:Panel 2:Hazard ratios and allocation ratios}
{p 0 4}
{cmd:Choose treatment group} This listbox selects the group for entry of Hazard ratios, Allocation
ratios and (where the Trend option is selected) Doses.
Defaults are provided for group 1 and for allocation ratios.
Values of hazard ratios must be entered for all groups other
than 1.
{p_end}
{p 0 4}
{cmd:Hazard ratios} In the simplest case, you specify a single hazard
ratio (HR) for failure for each group, with the default HR=1 for
group 1 (the control group). If desired, you may
specify as many hazard ratios as there are periods;
this allows you to design a trial in which non-proportional
hazards are expected. If for a given group
you enter fewer HRs than the number of periods, the
remaining HRs are taken as the last specified HR.
If you do not specify an HR for a particular group,
its value in a given period is
taken to be the geometric mean of the HRs specified for the
same period across all the groups for which you have
entered a value.
{p_end}
{p 0 4}
{cmd:Allocation ratios} By default, all groups are assumed of equal size, so
the allocation ratios (more precisely, weights) are
all equal to 1. You can vary this, e.g. assigning
allocation ratio 1 to group 1 and 2 to group 2 would
specify that group 2 should have twice as many
patients allocated as group 1.
{p_end}
{p 0 4}
{cmd:Trend} Implements a design assuming a linear trend test across the groups, with
scores 1, 2, 3,... attached to the groups. A trend
test may be more powerful than a general comparison
between the groups. See also {cmd:Dose}, which allows you to change
the scores or doses.
{p_end}
{p 0 4}
{cmd:Dose} Dose is a quantity assigned to each group which represents the
dose of some medication or other intervention received
by the subjects in that group. If you specify a dose
level for any group, you must specify a level for
every group. If you ask for a trend design (see {cmd:Trend})
and do not specify dose levels, the latter are taken
to be the numbers 1, 2, 3, ... and represent the doses for
groups 1, 2, 3, ... respectively. The assigned doses depend
on the specific design and need to be chosen
carefully.
{p_end}
{title:Panel 3:Patient recruitment and Model options}
{p 0 4}
{cmd:Duration} (default 0 periods) specifies the duration of recruitment.
The maximum duration of recruitment is the number of
periods specified in {cmd:Number of periods} (panel 1). The minimum
duration is 0, in which case recruitment
is assumed to be complete at the start of the trial.
When the duration>0 is specified, recruitment is assumed
to occur at a uniform rate for the number of periods
specified, and then stop.
{p_end}
{p 0 4}
{cmd:Proportion recruited at start} (default 0) Sometimes you may have
patients already available for randomization
at the start of the trial. The proportion of the total
sample size represented by this group of patients may be specified here.
The default of 0 assumes that all patients are recruited
in a "staggered entry" pattern - the usual situation.
{p_end}
{p 0 4}
{cmd:Unequal weights} (default Equal weights over periods) If you check this radio button,
you may then enter values which represent the relative numbers of
patients recruited in each period (the so-called "period weights").
This is a powerful option important for prospective and
retrospective calculations. For example, you may expect to recruit
say half as many patients in the first year (period) than in subsequent
years; you would specify {cmd:Unequal weights} as {cmd:0.5 1}
(or equivalently, as {cmd:1 2}, and so on). Or, you may want to
enter the actual number of patients recruited so far, to perform
"what if?" calculations. If you put fewer values than there are
periods, the remainder are assumed equal to the last value you
entered.
{p_end}
{p 0 4}
{cmd:Exponential accrual} (default uniform accrual) The shape of the recruitment
distribution can be altered to negative exponential by checking this
button. You then enter the rate in each period.
{p_end}
{p 0 4}
{cmd:Local and distant alternatives} (default local). This is a rarely used setting which
determines the way the program calculates under the alternative hypothesis.
It usually affects the resulting power or sample size very little.
You are only likely to wish to specify distant alternatives
if the target hazard ratio(s) are very far from 1, e.g. < 0.4.
{p_end}
{p 0 4}
{cmd:Method of sample size calculation} (default unweighted logrank test) This refers to the
statistical model used in the computations. It is unusual that
you would depart from the default logrank test.
There are four additional options. Alternatives to the standard
logrank test are the Tarone-Ware test, which is logrank with weights
proportional to the square root of the total number at risk at event times,
Harrington-Fleming, which is logrank with weights proportional to S^I, where
S is the estimated pooled survival function at event times and I is
the index for Harrington-Fleming weights (see option index()),
a binomial test conditional on the proportion of failures at the end of
the study, using Peto's approximation to the log odds ratio, and an
unconditional binomial test.
Note that values other than 1 of the index I for the Harrington-Fleming test
are available only through the {cmd:index()} option of {help artsurv}
and must be invoked by issuing a {cmd:artsurv} command.
{p_end}
{p 0 4}
{cmd:Additional details in output} provides event-rate and other information calculated by the program.
{p 0 4}
{cmd:Save using filename} allows certain results to be saved to a Stata file for plotting etc.
Details are given under {it:Remarks} in {help artsurv}.
{p_end}
{title:Advanced options}
{p 0 4}
{cmd:Loss to follow-up} The cumulative distribution function of time to loss
to follow-up. You may enter the cumulative proportion
of patients lost to to follow-up by the end of each
period. In the simplest case, you give just one value,
which is taken as the cumulative probability of loss
to follow-up at the end of the last period. To specify
values e.g. for periods 1 and 2, say of 0.05 and 0.1
respectively, enter as {cmd:1 2} in {cmd:At the end of period(s)}.
Values for any subset of periods may be specified.
{p_end}
{p 0 4}
{cmd:Withdrawal from allocated treatment} The cumulative distribution function of time to
withdrawal from allocated treatment (cross-over). You
enter values in the same way as for loss to follow-up.
The failure time distribution after cross-over is
specified by either the post-withdrawal hazard ratio
function or the target group upon cross-over. The
radio buttons allow you to choose which you wish to
enter.
{p_end}
{p 0 4}
{cmd:Hazard ratios post-withdrawal} For each arm subject to cross-over, you enter the
post-withdrawal hazard ratio function (relative to the
hazard of the baseline (control) failure time
distribution). You may enter as many values as there
are periods. If the number of values entered is less
than the number of periods, then the last HR value
applies to the remaining periods.
{p_end}
{p 0 4}
{cmd:Target group on cross-over} For each arm subject to cross-over, enter the target
group number. By default, group 1 crosses over to
group 2 and all other groups cross over to group 1.
{p_end}
{title:Authors}
{pstd}Abdel Babiker, MRC Clinical Trials Unit at UCL{break}
{browse "mailto:a.babiker@ucl.ac.uk":Ab Babiker}
{pstd}Friederike Maria-Sophie Barthel, formerly MRC Clinical Trials Unit{break}
{browse "mailto:sophie@fm-sbarthel.de":Sophie Barthel}
{pstd}Babak Choodari-Oskooei, MRC Clinical Trials Unit at UCL{break}
{browse "mailto:b.choodari-oskooei@ucl.ac.uk":Babak Oskooei}
{pstd}Patrick Royston, MRC Clinical Trials Unit at UCL{break}
{browse "mailto:j.royston@ucl.ac.uk":Patrick Royston}
{title:Also see}
Manual: {hi:[R] sampsi}, {hi:[R] stpower}
{p 4 13 2}
Online: help for {help artbin}, {help artsurv}, {help artmenu}, {help artpep}