New package on the SSC archive: titebeta, a tool for monitoring late onset toxicity in clinical trials

[Paul Wileyto]

The command titebeta is used for continuous toxicity monitoring in clinical trials where censoring during the observation period is important. It does so by calculating the posterior probability that the toxicity rate (q) exceeds a target maximum value (q*), in indicting whether accrual should continue or stop based on a preset threshold for that posterior probability. This mirrors standard Bayesian toxicity monitoring based on the beta distribution and binomial toxicity data, but makes use of censored data using the likelihood described by Cheung and Thall (2002). The posterior probabilities must be calculated numerically because no conjugate exists.

The routine makes use of two data columns that together make Y_i(t). The first column (0/1) indicates whether an event has occurred inside the observation window, and the second gives elapsed time under observation up to the time limit for the observation window. The routine evaluates the posterior probability as follows:

where a and b are the parameters for the beta prior distribution. The quantities w_i are time to event based weights representing the proportion of information available for each patient (Cheung and Thall 2002), and values are constrained between 0 and 1. The default calculation of weights is t_elapsed/t_max,. Other representations of weights are accomplished by the use of options.

The command is r-class and returns a series of scalars, including the posterior probability, and macros containing text.

Cheung, Y.K. and Thall, P.F. (2002), Monitoring the Rates of Composite Events with Censored Data in Phase II Clinical Trials. Biometrics, 58: 89-97. https://doi.org/10.1111/j.0006-341X.2002.00089.x




E. Paul Wileyto
[email protected]

[Tiago Pereira]

Paul, thank you so much. It would be extremely useful to have a small paper published on Stata Journal explaining the methods and the command - with an example, if possible.