plan: Plan

In precautionary: Safety Diagnostics for Dose-Escalation Trial Designs

Description

Plan

Version 0.2.6-1

• Improve feedback during long calculations

  • Set J = '...' pending first progress report

  • Analyze where Cpe$path_array() spends time

  • Give suitable feedback on path_array progress

Version 0.2.7

• Demote escalation to a 'Suggests'

• Thoroughly rewrite 'Intro' and 'FDA Clinical Hold' vignettes

Version 0.2.8

• BOIN recs via isotonic regression

• Test-that CPE matches BOIN::get.oc() sim

• Implement CRM logistic model

Version 0.2.9

• Implement TITE CRM?

  • Is TITE even amenable to path enumeration?

  • If not, there may be little sense in implementing it here!

  • If TITE is truly off the table, can existing numerics be sped up further?

  • OTOH, implementing TITE would help place the numerics for enumerable designs in context.

Version 0.3.0

• Stop exposing the impl parameter

Version 0.3.1

• Native Rust CPE

Dependencies

• Eliminate dependence on/adherence to dtpcrm layout

  • Package dtpcrm makes many design decisions unsuited to comprehensive enumeration of whole trials.

  • Allowing the Crm class to evolve along now-'natural' lines probably makes a suitable 'plan' for now.

  • Expunging superseded code/dependencies ASAP will facilitate this evolution.

  • Even the term 'DTP' seems not quite right anymore for a complete path enumeration (CPE?), and should be abandoned.

• Eliminate dependence on escalation and dfcrm

  • With the special emphasis on speed created by CPE, I now must implement all underlying trial designs in Rust.

• What may be retained in each case are 'Suggests:'-type relations, wherever regression tests are helpful – esp. wrt dfcrm.

Fast CRM

• Benchmark; try mul_add()s

Document

• Add examples to the documented Rust functions

• Add a vignette applying DTP to mTPI and BOIN

• Expose some visual numerics checks, via vignette or documented function

Refactor

• Remove the $safety component of exact trials?

  • Perhaps this ought to be calculated 'on the fly' by the summary method.

  • On-the-fly calculation would postpone use of ordinalizer, in keeping with the pattern established already for (non-exact) simulations.

• Should summary(EXACT)$safety bear class 'safetytab'?

• Example for 'as.data.table.exact'

• What is role of G function in exact.R?

Extend

• Implement exact 3+3 variant with allow_deescalation=FALSE

• Implement rolling 6

• Allow an accelerated titration phase

  • Note that this requires access to graded toxicities at simulation time, and therefore constitutes a substantial challenge to the generalizability of this software design.

• Index sims$fits to exact outcomes in A[[D]] where appropriate

  • See the haystack function in exact.R

Robustness

• Tests comparing results from multiple CRAN packages

precautionary documentation built on Aug. 9, 2021, 9:14 a.m.