A package-local (as-yet, unexported) test harness adapted from dfcrm::crm().

for various performance tuning experiments. The impl arg allows selection of various alternative implementations:

'rusti' substitutes integrands crmh, crmht, crmht2 written in Rust
'dfcrm' is the original as implemented in package dfcrm.

crm(
  prior,
  target,
  tox,
  level,
  n = length(level),
  dosename = NULL,
  include = 1:n,
  pid = 1:n,
  conf.level = 0.9,
  method = "bayes",
  model = "empiric",
  intcpt = 3,
  scale = sqrt(1.34),
  model.detail = TRUE,
  patient.detail = TRUE,
  var.est = TRUE,
  impl = c("rusti", "ruste", "dfcrm")
)

Arguments

prior: The CRM skeleton: dose-wise prior probabilities of toxicity
target: Target toxicity rate
tox: A patient-wise vector of toxicity counts
level: A patient-wise vector of dose level assignments
n: The number of patients enrolled
dosename: Optional designators for the doses
include: Index of patients to include
pid: Vector of patient ID labels
conf.level: Used to assign upper and lower bounds on predicted probability of toxicity, which in turn may be referenced in escalation, deescalation and stopping decisions.
method: Estimation method:
model: Presently, only the ‘empiric’ (or ‘power’) model has a Rust likelihood implementation.
intcpt: Intercept for ‘logistic’ model
scale: Sigma parameter of prior on beta parameter
model.detail: If FALSE, the model content of an mtd object will not be displayed. Default is TRUE.
patient.detail: If FALSE, patient summary of an mtd object will not be displayed. Default is TRUE.
var.est: If TRUE, variance of the estimate of the model parameter and probability/confidence interval for the dose-toxicity curve will be computed
impl: Switch between 'rusti' and 'dfcrm' implementations. Currently the 'rusti' option is implemented only for the Bayes method of the empirical (‘power’) model. An experimental 'ruste' implementation is in the works.

Author

Adapted by David C. Norris, from Ken Cheung's dfcrm