Increments rules are used to define the set of doses that are eligible to be used in the next cohort. NextBest rules then select which of these eligible doses should actually be used. Currently, Increments rules exist to limit the set of eligible doses based on (relative) change from the current dose, the number of toxicities observed so far and other criteria. None, however, limit the set of eligible doses based on posterior probabilities of toxicity. (IncrementsHSRBeta uses a beta-binomial model with a fixed prior that is independent of the current fit of the CRM model.)
It would be useful to have such a rule. It should work both for the standard binary CRM as well as for the ordinal CRM. In the latter case, different probability limits for different toxicity grades should be permitted.
For example
[Binary case] The dose for the next cohort should be limited to those doses for which the current posterior mean probability of toxicity is less than 0.33.
Or
[Ordinal case] The dose for the next cohort should be limited to those doses for which the current posterior median probability of grade 1 toxicity is less than 0.4 and of grade 2 toxicity is less than 0.05.
This issue potentially supersedes #855.
Incrementsrules are used to define the set of doses that are eligible to be used in the next cohort.NextBestrules then select which of these eligible doses should actually be used. Currently,Incrementsrules exist to limit the set of eligible doses based on (relative) change from the current dose, the number of toxicities observed so far and other criteria. None, however, limit the set of eligible doses based on posterior probabilities of toxicity. (IncrementsHSRBetauses a beta-binomial model with a fixed prior that is independent of the current fit of the CRM model.)It would be useful to have such a rule. It should work both for the standard binary CRM as well as for the ordinal CRM. In the latter case, different probability limits for different toxicity grades should be permitted.
For example
Or