Practical Issues on Using Weighted Logrank Tests with Interval-Censored Events in Clinical Trials - Interval-Censored-Time-to-Event Data: Methods and Applications - page 353

appears to be valid even under the CIA assumption with ATD. Section 13.3.7

describes other tests that are closely related to those found in the interval
package.

13.3.2 Practical Proposal: Ignoring Unscheduled Assess-

ments

For many types of interval-censored data, there may be intentional or uninten-

tional assessment-treatment dependence. Freidlin et al. (2007) discuss several

causes of ATD (which they call evaluation bias) that can occur in unblinded

trials when progression-free survival is used as an endpoint. For example, pa-

tients randomized to the control group may be more likely to drop out of the

study and it may be dicult to obtain cancer progression assessments from

those drop-outs. As another example, for patients who come into the clinic at

times other than scheduled assessment times, the patients on control may be

more likely to get an unscheduled assessment in an unblinded trial because

the clinicians may suspect progression more in those patients.
Freidlin et al. (2007) propose a modification of the definition of progression-

free survival to ensure there is no ATD; they defined the allowable assessments

for the cancer progression as only those assessments that occur at scheduled

assessment times, and they scheduled assessments at the same two times for

both treatment and control. The suggested assessment times are at the as-

sumed median and two times the median of the control arm. This idea can

be easily expanded to k scheduled assessments as long as the k assessments

are the same for both treatment groups and missingness of scheduled cancer

progression assessments is negligible (unless that missingness is due to death).

By ignoring all unscheduled assessments, Freidlin et al. (2007) ensure totally

independent assessment even if the unscheduled assessments are informative

and have ATD. To test for a treatment effect, Freidlin et al. (2007) simply

assumed that observed progressions occurred exactly at the right endpoint of

the scheduled interval (e.g., if the scheduled time is at 4 weeks with a 1-week