Piantadosi Chapter 12:
KittelsonReading Notes
The chapter has a lot of excellent discussion of the key considerations in defining the study population (eligibility criteria). I suggest starting with the summary (section 12.5), which nicely states the major points.
Section12.5: Note differences between single-arm studies and randomized comparative trials (we should always be predisposed toward the latter).
Section 12.1: A trial’s results must be reproducible in that the same conclusions are reached if it were to be replicated in the same population (internal validity). A trial must also be generalizable to the intended target population (external validity).
Note impact of overly restrictive eligibility criteria on both study generalizability and on the ability to accrue a trial
Aside: Consider a recent trial of a new anti-platelet drug Ticagrelor (the PLATO trial). It caused controversy because it showed benefit in all parts of the world, but was harmful (not significantly so) in the US. The debate was about whether this is sub-group analysis run amuck or whether it is indication of general concern about approvability in the US.
Section 12.2: Single-arm (uncontrolled or historic-control) trials are difficult to interpret because it is always possible that the results are distorted by selection bias.
Section 12.2.1: Useful explanations of a random sample from the target population versus other approaches. Note that targeted therapies (those targeting a particular mechanism or subpopulation) must be co-developed with an accurate diagnostic test for identifying the appropriate patient population.
Section 12.2.2: Notice (third paragraph of 12.2.2) the need for a randomized control group in a setting where there might be a large “trial participant effect.”
[You can ignore the statistics on pages 312-314 – this is beyond the scope of this course.] The statistics are a formalization of the idea that a mixture of different patient groups (each with its own response rate) will result in a population response rate that is a weighted average of that from each of the different groups.
Section 12.2.3: The variability that is inherent in large simple trials generally produces better generalizability even though they might require a larger sample size. I liked the comment/observation on the difference between written eligibility criteria and their interpretation by study personnel.
Piantadosi makes an important point that the study population is often restricted in early studies until safety can be assured. Later studies sometimes have less restriction because safety is understood. [See also my class notes on reasons to restrict trial eligibility.]
Section 12.2.4: Excellent observation on heterogeneity on the definition of “normal” lab values, which leads to a false sense of precision. Note that sometimes these questions require central adjudication if the questions are important enough to the science of the trial (refer to CTEPH example).
Section 12.2.5: Notice that comparative trials will not be confounded by selection bias as long as treatment allocation (randomization) is adequately concealed. BUT, always remember that generalizability is affected by eligibility.
The argument about robustness of generalizability (last half of this section) is not accepted by many non-trialists (who often argue that carefully controlled trials do not generalize to routine practice).
Section 12.3: Many trials fail due to lack of accrual… the importance of assuring realistic accrual estimates is essential to good trial planning. The mathematical details on pages 319-322 can be skipped (they are beyond the scope of this course).
Section 12.4: Excellent discussion!
Section 12.4.2 merits emphasis: it is important to reduce barriers to participation in trials.
I always advocate for designs that favor effectiveness over efficacy. The discussion on pages 326-329 is interesting, but has not had much impact in my work in clinical trials (although that may say more about me than about the importance of the issue).
I liked the Benjamin Franklin quote on the bottom of page 324. I also liked the observation about Tamoxifen on page 327.