Paul Gallo 2019 Joint Statistical Meetings July 31, 2019 Implementing Effective DMC Decision-Making in Complex Clinical Trial Designs Paul Gallo 2019 Joint Statistical Meetings July 31, 2019

Expanding utilization of DMCs The usage of Data Monitoring Committees has increased in recent years So has the scope of potential recommendations, e.g.: stopping or modifying a trial for ethical reasons based on safety concerns stopping for demonstrated efficacy (usually based on a group sequential scheme) stopping for lack of effect, or futility implementing an adaptive design scheme To be implemented efficiently while maintaining integrity and interpretability of trial results

Some issues Understanding relevant benefit-risk considerations DMC role in complex / adaptive trials Level / details of unblinding in DMC reports Data quality Stated role of the DMC statistician

1) Benefit-risk Recently, often heard / seen: Do they really? “DMCs evaluate benefit-risk”, or “DMCs assess benefit-risk”, etc. Do they really? In the same manner as: FDA, when deciding whether a treatment should be available to patients, or what a label should describe? Me and my doctor, when deciding on a course of treatment for me?

Data access It’s increasingly accepted – correctly – that DMCs should have access to relevant safety and efficacy data regardless of the specific focus of the monitoring plan in order to ensure they can carry out their responsibilities to trial participants

Challenges Interim data often can not convey a clear picture of either risk or benefit, much less both misleading signals (especially early) are inevitable different schedules / time courses of different outcomes some important safety concerns are quite long-term Expert DMCs sensibly integrate the strength of numerical signals, relevant scientific knowledge, consistency of patterns, etc. e.g., what types of safety risks might be expected? but sometimes there are surprises . . .

So, what does the DMC do for B-R? Is this more correctly stated as “consider” rather than “assess” or “evaluate”? What is the question the DMC addresses as it considers whether a trial should continue? How about Is this study a rational, ethical medical experiment that can provide important information on therapies that may have favorable benefit-risk profiles, without exposing participants to undue risks, relative to the answers it may provide? Usually, not something that lends itself to simple algorithms

2) Adaptive designs Adaptive designs have received increasing focus during years Examples of aspects that might be changed: sample size, treatment arms, patient population, randomization allocation, etc. Validity of conclusions depends on adequate pre-specification of an adaptive plan

DMCs in adaptive designs A DMC seems a natural party to consider for a role in the adaptation decision process they’re already allowed access to interim results independence, objectivity Possible misunderstanding: what changes can a DMC implement pro-actively? apart from their safety responsibilities, which always take precedence

DMC role adaptive trials What the DMC can’t do Proactively initiate aspects of change based on their review – essentially re-design the trial in a more favorable direction pre-specification is a fundamental tenet of a valid design What the DMC could do Play a role in implementing a pre-planned adaptation scheme note: doesn’t mean that actions are fully algorithmic

DMC as implementer FDA 2010 draft AD guidance: FDA 2018: “a DMC . . . can help implement the adaptation decision according to the prospective adaptation algorithm, but it should not be in a position to otherwise change the study design except for serious safety-related concerns that are the usual responsibility of a DMC” FDA 2018: “. . . revisions based on non-prospectively planned analyses can create difficulty in controlling the Type I error probability and in interpreting the trial results”

One board, or two? Should adaptation decisions be made by a group separate from a more “familiar” DMC? Both single and separate board models seem to have been used FDA 2018 acknowledges both possibilities Adaptations are an aspect of trial design – decisions may not be completely independent of other DMC recommendations Personally, I’d usually recommend a single DMC (as per Antonijevic et al, TIRS 2013)

3) Coding / unblinding Fully unblinded outputs? Coded, e.g., A / B? Arguments in support of coding sometimes focus on the “accidental access” possibility I actually saw this once: A / B, then C / D, E / F . . . Different coding for efficacy vs safety (or other parameters with potential to de-code)? Or else, how can the DMC take into account critical benefit-risk considerations?

Fundamental principle We need to be NOT in the mindset that added clarity of information to a DMC in any way compromises a trial

Options Putting a de-coding envelope in the hands of the DMC usually satisfies most perspectives opened whenever they want (even on Day 1), no need to request, or to tell anyone In situations structurally more complex or where full labelling sensibly facilitates review (e.g., multiple doses), we’d do so up front If a DMC initially insists on explicit labelling, we wouldn’t object

4) Data quality / completeness Data of less than perfect quality is a source of uncertainty so is having less data! Timely decisions, based on current available information, have important ethical implications Inherent conflict between quality / completeness of data, and “recent-ness” can’t recent, not-fully-cleaned data be relevant? Overheard from an experienced DMC expert: If it’s clean, it’s too old!

Data quality A principle?: DMC should never make what turns out to be an incorrect decision because they didn’t understand limitations in the data they reviewed! I like the idea of making essentially everything available, clearly explaining any limitations Providing multiple versions? – e.g., clean / all? adjudicated / all? (+ adjudication confirmation rates) Allocate resources sensibly

5) DMC statistician – a 2nd class citizen? Proposals I’ve seen / heard: The DMC statistician is “a non-voting member” “No statistician is needed because the DMC is only reviewing safety” DMC composition reflects a variety of expertises and perspectives relevant to a decision Statistics is unquestionably one of these Is “voting” over-emphasized anyway? For major decisions, DMCs aim to arrive at a consensus that all members are comfortable with

Stated role of the DMC statistician Safety review These may be the setting where statistical perspective is most indispensable! Strength of signal vs scientific plausibility vs degree of exploration vs consistency of related outcomes – what is the data really saying? I would recommend to resist any specification of the DMC statistician as less than a full- fledged member

Thank you !!