Title
landemets -- Boundaries for group sequential clinical trials using alpha spending functions
landemets , [options]
options Description ------------------------------------------------------------------------- t(numlist) set monitoring times; default is t(1) alpha(numlist) set type I error probability(ies); default is alpha(.05) method(method) set the alpha spending function(s); default is method(obf) rho(numlist) set the parameter of alpha spending functions in the power family onesided control if one-sided boundaries are to be computed plot display a graph with the boundaries -------------------------------------------------------------------------
landemets computes boundaries for group sequential clinical trials using the method of Lan and DeMets (1983).
landemets requires that the Stata module -moremata- (Jann, 2005) be installed.
Compute boundaries with the default options: . landemets
Ten equally spaced monitoring times between 0.1 and 1: . landemets, t(0.1(0.1)1)
The same, but plotting the boundaries: . landemets, t(0.1(0.1)1) plot
Type I error probability set to 0.01, with a different choice of four, not equally spaced times: . landemets, t(0.2 0.3 0.5 1) alpha(0.01)
The same, but using a Pocock type alpha spending function: . landemets, t(0.2 0.3 0.5 1) alpha(0.01) method(poc)
One-sided boundaries using an alpha spending function in the power family, with parameter set to 1.5: . landemets, t(0.2(0.2)1) method(pow) rho(1.5) onesided
Two-sided asymmetric boundaries; the total type I error probability, 0.05, is divided in 0.04 and 0.01 for the lower and upper boundaries, respectively; an O'Brien-Fleming type alpha spending function is used for both boundaries: . landemets, t(0.2(0.2)1) alpha(0.04, 0.01) method(obf obf)
Two-sided asymmetric boundaries; the total type I error probability, 0.05, is equally divided between the lower and upper boundaries, but different alpha spending functions are used for each boundary (lower: power family, with rho = 2, upper: O'Brien-Fleming type): . landemets, t(0.2(0.2)1) alpha(0.025, 0.025) method(pow obf) rho(2)
t(numlist) specifies the times at which analyses are to be performed. They must be contained in (0,1], and the maximum value should be 1. The default is t(1), meaning no interim analyses.
alpha(numlist) specifies the type I error probability. The numlist may be of length one or two. If it has only one value, two-sided (symmetric) or one-sided stopping boundaries for a type I error probability equal to that value are computed (whether two-sided or one-sided boundaries are computed is controlled by the onesided option; see below). If the length of numlist is two, its components give the type I error probability for, respectively, the lower and upper boundaries. Thus, if the two alpha values or the alpha spending function used for each boundary are different (see the method option below), asymmetric two-sided boundaries are obtained. The default is alpha(0.05).
method(method) specifies the alpha spending function(s) to be used. The accepted arguments are: a) one of obf (O'Brien-Fleming type), poc(Pocock type), or pow (power family). This is the way to specify the alpha spending function for symmetric two-sided or one-sided boundaries. When the method is thus specified, the argument of the alpha option must be of length one. b) any two (possibly repeated) values in a) separated by one or more spaces. This is the way to specify the alpha spending functions used for each boundary (the first word for the lower boundary, the second word for the upper one) for asymmetric boundaries. When the method is thus specified, the argument of the alpha option must be of length two. The default is method(obf). For details on the types of alpha spending functions see, e.g., Cook and DeMets, 2008.
rho(numlist) specifies the parameter(s) of the power family alpha spending function(s). The numlist may be of length one or two, depending on how many alpha spending functions of this type are specified by the method option.
onesided specifies if one-sided, instead of two-sided boundaries, are to be computed.
plot specifies if a plot of stopping boundaries vs. monitoring times is to be displayed.
landemets saves the following in r():
Scalars
r(alpha) overall probability of type I error (omitted when asymmetric boundaries are computed) r(alpha_lower) overall probability of type I error for the lower boundary (omitted when symmetric or one-sided boundaries are computed) r(alpha_upper) overall probability of type I error for the upper boundary(omitted when symmetric or one-sided boundaries are computed) r(K) number of interim analyses Macros
r(bound_type) type of boundaries computed: two-sided (asymmetric or not) or one-sided r(method) alpha spending function used (omitted when asymmetric boundaries are computed) r(method_lower) alpha spending function used for the lower boundary (omitted when symmetric or one-sided boundaries are computed) r(method_upper) alpha spending function used for the upper boundary (omitted when symmetric or one-sided boundaries are computed)
Matrices
r(bound_alpha) a matrix with the following named columns: time, monitoring times lower, lower boundaries (omitted if one-sided boundaries are computed) upper, upper boundaries cumalpha, cumulative alpha values diffalpha, first difference of the cumulative alpha values
Ignacio López de Ullibarri Department of Mathematics University of A Coruña, Spain E-mail: ilu@udc.es
Cook TD and DeMets DL (2008), Introduction to Statistical Methods for Clinical Trials, Boca Raton: Chapman & Hall/CRC
Jann B (2005), moremata: Stata module (Mata) to provide various functions, available from http://ideas.repec.org/c/boc/bocode/s455001.html
Lan K and DeMets DL (1983), Discrete sequential boundaries for clinical