Last week we released an infographic on why Phase 3 trials fail. The numbers, while eye-opening, did not capture a related and equally important issue: Why are so many late stage clinical trials discontinued?
Nearly 50% of all Phase 3 trials that are submitted to the FDA fail upon first submission . However, 25% of all trials that begin are never even submitted for review . According to a 2014 JAMA study, nearly 40% of these discontinuations cite poor enrollment as the primary reason for stopping a trial. Needless to say, the costs of discontinuity are significant.The complexities of trial enrollment are well-known. The clinical operations team receives two sets of instructions: the first from the sponsor who tells them the trial deadline and the second from statisticians who specify population numbers and distribution. They then turn to one or more historical databases outlining how well trial sites have performed for various therapeutic areas in the past. Enrollment requires optimizing across these three domains, the most flexible of which is the time it takes to complete a trial.
As a result, trials become longer, more expensive, and finally, suitable candidates for discontinuation.
Trial forecasting is an interesting subfield within clinical development, currently evolving along several parameters. Undoubtedly, a part of the challenge of forecasting is the number of variables that could potentially deviate from expectation. A conventional strategy has been to make a list of what could potentially go wrong, estimate the likelihood that it will, wait until something actually happens, and handle it as quickly as possible. It’s a pragmatic solution, but one with several disadvantages:
- First of all, knowing that something has gone wrong or even the probability that something will go wrong, does not tell you how much your asset depends on it for success. For example, knowing that there is a 90% chance that some event is likely to interfere with success may often mean avoiding dependency on that event. However, in other cases it might mean aggressively pursuing the 10% chance of success, if a lot of benefit is expected to come of it. In order to make fast executive decisions about whether or not to pursue a certain eventuality, it helps to have a model of each possible event, its likelihood, and a plan of how respond to its eventuality, based on the effects of such an event on asset value.
- Secondly, having a single point estimate of how likely an event is going to occur is typically less useful than having a range of probabilities. Moreover, more information about this range is likely to unfold as a trial progresses. As a result, planning for a possible event might require access to information about how its probability will be shaped over time by other factors
In general, a move towards model-based forecasting is creating exceptional benefits to both trial planning (feasibility studies) and implementation (check out our East® PREDICT). It anticipates a move towards the use of new statistical methods and the use of simulations, at all stages of clinical operations.
Related Items of Interest
 Sacks LV, Shamsuddin HH, et al., 'Scientific and Regulatory Reasons for Delay and Denial of FDA Approval of Initial Applications for New Drugs,’ 2000-2012,' JAMA. 2014 311(4):378-384.
 Kasenda B, von Elm E, You J, et al. "Prevalence, Characteristics, and Publication of Discontinued Randomized Trials." JAMA. 2014;311(10):1045-1052. (This article is available to all).
From Cytel's Leadership Team
Nitin Patel, Pralay Senchaudhuri and Suresh Ankolekar (2012). "Better Planning through Design: Forecasting Enrollment in Clinical Trials When Site-Level Accural Rates Vary with Time." (Slides Available to All.)
Yannis Jemiai (2012). "Better Planning through Design: Modeling and Forecast Enrollment and Event Arrivals to Optimize Trial Execution." (Slides Avialable to All.)