OR WAIT 15 SECS
Acceptance of data ambiguity by the industry is the key to adaptive design adoption and use.
From a clinical trial context, this article will explore how our own desire for regularity in our data (e.g., clean data) undermines the utilization of adaptive clinical trial design. Only through the acceptance of an intermediate level of data variance, with the capture of the focused range of patient parameters, will clinical scientists and trial managers be able to fully exploit the information for each patient and make the necessary midtrial changes.
PHOTOGRAPHY: JIM SHIVE ILLUSTRATION: PAUL A. BELCI
Biologic systems exist in a state balanced between complete crystalline order and random disorder. Nobel Laureate Murray Gell-Mann, a pre-eminent theoretical physicist, characterizes such a state as having maximum "effective information complexity." His definition of effective complexity is a measure of degree by which information presented is random (i.e., incidental) verses regular (i.e., significant). Intuitively, however, it is the individual perception of data that determines whether we regard the information as incidental or significant.1
This broad theoretical principle around data interpretation and significance plays a crucial role within our execution of clinical trials today and is actually preventing us from enabling a trial design that would successfully deploy an adaptive model.
Physicians embrace a high degree of effective information complexity. They are taught to express their understanding of a patient as a differential diagnosis. "I have no clue" is as equally unacceptable as "I have complete certainty." With most conditions, the reality is actually a range of possible explanations in between these two extremes.
Each possible diagnostic explanation is judged to carry a different relative probability. A well-constructed differential diagnosis reflects the maximum effective information available on a patient at any given time. It also is most consistent with how biology is manifest at the individual patient level. However, in clinical research, this inherent complexity is discounted as unacceptable ambiguity.
Investigators are generally physicians and as such have been taught to think diagnostically in terms of differential diagnoses. The picture of an individual patient is often inconsistent and contradictory. This usually means that the signs and symptoms could be indicative of a wide range of alternatives, each with a plausibility/probability based on the investigator's interpretation of the overall patient picture. However, in clinical research, investigators are forced to forgo this differential diagnosis and select a diagnosis that appears definitive when, in fact, it may just be the most likely of several plausible scenarios.
For instance, it is not unusual with a patient presenting with a condition such as shortness of breath to have several plausible explanations with relatively similar presumptive probability, such as: anxiety, pulmonary embolism (PE), cardiac ischemis, respiratory inflammation, etc. Despite the range of possibilities and the correlating probability of each diagnosis, clinical trial data collection forces an investigator to select a diagnosis. [Trial data collection forms do not provide the opportunity to record the symptoms—valuable information that is ultimately lost.] The investigator essentially has two choices: Select a general term describing the patient condition or enter a precise diagnosis.
If the investigator selects the general term, this term will later be arbitrarily coded to a body system group. On the other hand, if the investigator selects one precise diagnosis, based on the range and probability of potential diagnosis, the degree of accuracy that is suitable to make this decision is arbitrarily chosen by the investigator. Information on other plausible solutions is once again lost. Either way, the effective complexity of the information readily available about the event has been reduced.
So, without the ability to capture the differential diagnosis from the investigator, critical patient information has been lost and thus can never be analyzed or measured. If one applies a mathematical analysis of the information contained by a differential diagnosis verses the information contained by a specific diagnosis, as much as 75% of the information value is lost.
For example, because PE are so difficult to diagnose they are almost never listed as the primary cause of death in a patient with chronic obstructive pulmonary disease (COPD). However, if a mortality classification committee were asked, PE could consistently be the second or third most plausible differential cause of death. As a result of pressing for diagnosis certitude within the confines of clinical trial forms, we fail to capture or fully contemplate the ramifications that as many as 25% of patients presenting with an acute exacerbation of COPD have undiagnosed PEs.2 Many of these inadequately treated patients then die as a result of long-term complications.
However, a different picture is likely to emerge if that same classification committee provides a differential set of potential causes of death with a consensus on probability level. Just as overcooking vegetables degrades their nutritional value, disregarding the range of possible patient variables degrades the value of the critical clinical trial data.
Insistence on a singular if not always precise diagnosis of medical conditions and the treatment of a large portion of subject information as incidental are the largest impediments to the industry's utilization of adaptive trial designs.
A major concern that has been routinely cited with adaptive trial designs is that the interim analyses that are used to drive significant trial modifications are based on very little data. To the extent that as much as two-thirds to three-fourths of the effective information has been lost through overzealous clarification of patient information, this is a valid objection.
However, it might be possible to double or triple the information yields from current sample sizes with minimal capital investment. Limited change in data collection methods and statistical methods would make a significant difference. Study teams must recognize the full range of applicable subject information and adapt their trial and protocol to capture all critical points of variability. Rather than a singular diagnosis for adverse events, differentials and probability weights are called for.
As the industry learns to embrace ambiguity, the use of adaptive design is bound to proliferate. Then, the practical limitations associated with the execution of adaptive trial designs will become more apparent. Immediate understanding of patient exposure levels and the flexibility to implement new drug allocation schemes will be required. The interactive voice response system/interactive Web response system (IVRS/IWRS) will be the key to overcoming these practical barriers to adaptive trials.
Unlike other data capture solutions, IVR/IWR data represents a modality of action, meaning it prompts an event to occur rather than recording the event's occurrence at a later time. This is essential in managing an adaptive trial.
There are two critical aspects in IVRS/IWRS designs that must be considered to ensure success in implementing an adaptive trial. First, an appropriate amount of flexibility must be built into the IVRS/IWRS design before the trial begins. This flexibility will ensure that the randomization scheme can be changed or treatment arms can be added or deleted in real-time without slowing or stopping the trial. Second, there must be the ability to retrieve cumulative, actionable, real-time data from a robust database into the necessary format for immediate evaluation.
To ensure the successful implementation of an adaptive trial design, theoretical and practical limitations must be addressed and overcome by the industry. It is only by embracing and enabling the more effective use of critical patient data that we can derive enough information to make comprehensive decisions over the course of a trial that will ensure patient safety both during the trial and after the drug is launched on the market.
Lawrence A. Meinert, MD, MPH, is senior vice president of medical and scientific affairs at Covance, Inc., 210 Carnegie Center, Princeton, NJ 08540, email: email@example.com
1. M. Gell-Mann and S. Lloyd, "Effective Complexity," Santa Fe Institute (2003), www.santafe.edu/research/publications/workingpapers/03-12-068.pdf.
2. M. Ambrosetti, W. Ageno, A. Spanevello et al., "Prevalence and Prevention of Venous Thromboembolism in Patients with Acute Exacerbations of COPD," Thrombosis Research, 112 (4). 203-207 (2003).
Related Content:Trial Design