Rules for k-in-a-row, Biased-Coin UD, and Group UD, coded as functions compatible with
the generic dose-finding simulator dfsim()
These functions work on each virtual experimental run individually.
krow(doses, responses, k, lowTarget = NULL, fastStart = FALSE, ...)bcd(doses, responses, coin, lowTarget, fastStart = FALSE, ...)
groupUD(doses, responses, s, ll, ul, ...)
the next dose allocation
(mandatory arguments) vectors of the run's current sequence of doses (in ordinal/index scale) and responses
the number of consecutive identical responses required for dose transitions (k-in-a-row functions only).
(krow and bcd) logical: is the target below 0.5 (median threshold)?
(krow and bcd) logical: should the experiment begin with a classical-UD-like stage until the first "minority" response is observed (i.e., a 1 for below-median targets and vice versa)? Even though TRUE delivers better experimental performance and is recommended when allowed, default is FALSE because toxicity/safety studies are unlikely to allow it.
Technical pass-through argument, to allow for flexibility when constructing design-comparison simulation ensembles.
(bcd only) the biased-coin probability. Note that unlike bcdmat(), here the function does not figure out automatically the coin probability and upper/lower target location from the provided target.
(groupUD only) the group/cohort size, analogous to cohort in gudmat(). We use a different name here because cohort is already used in dfsim that calls these utilities.
(groupUD only) how many positive responses are allowed for a move upward, and how many are required for a move downward, respectively. Analogous to lower, upper in gudmat(). For example s=3, ll=0, ul=2 evaluates groups of 3 observations at a time, moves up if none are positive, down if \(>=2\) are positive, and repeats the same dose with 1 positive.
Rules for some popular or well-studied non-up-and-down