Adaptive Decision Making Based on Interval-Censored Data in a Clinical Trial to Optimize Rapid Treatment of Stroke - Interval-Censored-Time-to-Event Data: Methods and Applications - page 340

12.6 Discussion

A simplified version of this methodology has been applied to plan a trial of IA

tPA in pediatric stroke patients. Although pediatric AIS is rare, over 75% of

children with acute AIS die or suffer long-term neurological deficits (deVeber

et al., 2000). Because a bolus of size q = 0.20 or larger was considered too

risky for children, the bolus is xed at q 0.10 in this trial. Thus, the response
time hazard function is simplied by xing 2 1; f E;T (y j c;) depends
on c but not q; and the mean utility u(c;) is simplied accordingly. The goal

is to nd the optimal concentration among the four values f0.20, 0.30, 0.40,

0.50g, with c is chosen to maximize E fu(c;) jD n g:
The model and methodology proposed here may be extended to accom-

modate a variety of oncology settings. For example, many chemotherapeutic

anti-cancer agents are administered by ci, possibly with an initial bolus, the tu-

mor is imaged repeatedly, and therapy is stopped if and when tumor response

is achieved. Such treatment regimes encompass weeks or months, rather than

minutes. Ecacy would be tumor response observed as an interval-censored

event time, with toxicity evaluated as a binary variable. The schedule for eval-

uating response likely would be more complex than with simple ci, as a typical

multicycle chemotherapy regimen alternates periods of infusion with periods

of rest, so it would be necessary to modify the model for to reect this.

A nontrivial complication would be if toxicity also were scored as a time-to-

event variable, possibly occurring during infusion, and causing treatment to
be suspended, modified by decreasing c, or stopped permanently. This would

require substantive modifications of the model, including a joint distribution

for the two event times, allowing each to informatively right-censor the other,

as well as the utility function and decision rules.

A computer program, named \CiBolus," to implement this methodology is

available from the website