At the recent JSM meeting in Chicago, Cytel's Jim Bolognese presented the results of work he has conducted evaluating the T-Statistic ( or T-Stat) method for adaptive dose finding of MTD. In this blog we'll provide a brief summary of Jim's findings, and share his slides with our blog readers.
Motivation
In previous blogs we have discussed the mTPI method which is popular for adaptive dose-finding of MTD. Part of its appeal is the fact that it is easy to implement ( with a fixed pre-stated algorithm and Dose Ranging model independent). It may be preferable to traditional rule based 3+3 designs and is efficient as a competitor to model based approaches such as the CRM/ BLRM ( which are incorporated in East Escalate)
The T-statistic ( T-stat) method has a similar appeal and popularity for adaptive dose finding of target dose due to its easy implementation and competitive efficiency with model based methods such as Emax, Normal Dynamic Linear Model (NDLM), and Bayesian 4 parameter logistic model( 4PL) .
In his presentation, Bolognese conducts an evaluation of how the T-stat stacks up against the mTPI method, CRM and BLRM approaches for toxicity dose finding trials. For the T-stat simulations Bolognese used Cytel's Compass software and East for the CRM, BLRM and mTPI.
Comparision
For the comparision, Bolognese assumed 7 doses and a total of 30 subjects, with 10 sequential cohorts of 3 subjects each with the 1st cohort at dose 1 and each subsequent cohort assigned a single dose per adaptive design. The target toxicity level was 0.3 and 10k simulations were created of each selected DR curve scenario.
Bolognese compares the methods using the following criteria:
- Probability of identifying correct target ID,
- Probability of estimating MTD at or adjacent to correct MTD
- Probability of assigning subjects to doses > target (ODs),
- Number of dose-limiting toxicities (DLT’s) observed
He finds that the T-stat method is an efficient competitor in consideration of the spectrum of TRUE underlying DR curves simulated when the 4 performance criteria are combined with equal weights.
To review Bolognese's findings in full click below to access the slide deck.
