A broad literature supports the importance of lifestyle, stress, and other psychosocial variables as independent risk factors for cancer and other diseases, whereas positive psychological well being is associated with reduced mortality.1 The evidence is sufficiently strong to motivate clinical research to evaluate the potential that intervening on these factors would alter the trajectory of these diseases. In their provocatively titled article,2 Stefanek and colleagues criticize the design and interpretation of results from a recently published clinical trial3 of psychological interventions for breast cancer patients. Although they raise valid criticisms related to its design and analysis, their recommendation that clinical trials of this nature be suspended until putative mechanisms of action are understood better is inconsistent with the general history of clinical trials in medicine. Moreover, the trial conducted by Andersen et al stands as a considerable accomplishment that deserves a closer look in a broader context.
Two major trends have influenced the design and conduct of clinical trials of behavioral interventions. One is a widely held belief that behavioral interventions must be theory-based, reflecting psychological mechanisms of behavior change on 1 hand and addressing biologic mechanisms of disease causality on the other. The effect of these 2 principles on the design of clinical trials has been to include the collection of extensive secondary outcomes that might identify potential moderators and mediators of the intervention as well as biobehavioral mechanisms that might be affected if the intervention shows promise. The trial by Andersen et al is an excellent case in point.
The intervention was designed carefully to address all of the biobehavioral pathways that are believed to influence the course of illness4 and to monitor progress on each. The investigators collected extensive data on stress and depression, health habits, adherence to medical care and indicators of immune function, devising composite indices that included many of these variables. The various measures and methods were detailed in a series of articles1, 5-7 that reflect the enormous amount of careful effort expended to design and monitor numerous behavioral and biologic aspects of the trial. The fact that these aspects of the trial were executed successfully is reflected in the excellent availability of data, exceeding 90% on most variables. Conducting a trial of this complexity required remarkable attention to detail and consistency in performance. The analytic strategy used by Andersen et al also reflects intensive attention to secondary outcomes. No fewer than 3 articles describing secondary outcome results in extensive detail preceded the results article that is the subject of this discussion. The emphasis on seeking mechanisms, moderators, and mediators may be the only point on which Andersen et al and Stefanek et al agree: the former by virtue of the data-analytic strategy reflected in their series of publications on the trial participants and the latter by their articulated position that much more mechanistic research should be conducted before making “huge investments of time, money, and professional and patient resources” in clinical trials.
The crux of the matter is that the opposite strategy may be more appropriate. A preoccupation with mechanisms and secondary aims in behavioral clinical trials too often has detracted from attention to commonly accepted principles of trial design and has resulted in allocating resources that could have been spent better in increased sample size or quality of the intervention. When social support is an important component of the behavioral intervention or when repeated assessment of secondary variables results in increased self-monitoring of behavior in the control group, the treatment effect may be reduced. Moderators and mediators are a poor substitute for hard endpoints.
Understanding the mechanisms that may underlie pathology does not assure that treatments targeting them will be successful. Examples of treatment failures despite sound theoretical and mechanistic underpinnings abound in medicine. An often cited example is the Cardiac Arrhythmia Suppression Trial (CAST),8 in which investigators evaluated whether pharmacologic agents that suppressed ventricular tachycardia and ventricular fibrillation in patients after myocardial infarction reduced death because of arrhythmias. The treatment strategy was based on the rationale that abolishing ventricular arrhythmias would prevent sudden cardiac death. Eligible patients could be identified by monitoring their electrocardiograms during daily life or during an exercise stress test. Despite a very strong biologic rationale and the availability of methods with which to identify vulnerable patients, patients who received agents that suppressed complex ventricular ectopy had an approximately 3-fold increase in cardiac death compared with patients on placebo. It is difficult to envision a closer link between a biologic mechanism and a treatment modality than that evident in CAST. More recently, the Womens' Health Initiative trial of combined estrogen and progestin replacement therapy in healthy menopausal women was stopped early9 because of an increased risk of breast cancer and other conditions. The trial had been preceded by extensive data in support of hormone-replacement therapy (HRT) from observational studies and smaller trials in postmenopausal women and from a strong mechanistic justification for treatment. The perceived effectiveness of HRT in reducing cardiovascular events led to considerable controversy regarding the need for this large-scale trial, yet the results were unexpected even among its proponents. The lack of assurance that apparently well understood mechanisms might provide for the success of treatment has not been recognized in the behavioral clinical trials community because it has not been faced with such harsh surprises. Behavioral clinical trials typically result in either improvements or null effects, but not in harm. The detailed rationale for the design of the behavioral treatment arm of the Combined Pharmacotherapies and Behavioral Interventions for Alcohol Dependence trial (the COMBINE study)10 did not result in a successful combined behavioral intervention.11 The mechanistic rationale for the pharmacologic agents used in that trial, naltrexone and acamprosate, also failed to improve the outcome of treatment for alcohol dependence either alone or in combination. Thus, the main results of a well designed trial, in and of themselves, can be a test of the underlying mechanisms and an important criterion in the assessment of causality.12
In short, we should reconsider to what extent the mechanisms through which psychosocial factors might influence the course of disease or its treatment should precede randomized clinical trials or should be an essential component within them. Although this approach has its advantages, such as opportunity for discovery and improved treatments, excessive attention to these objectives can infringe on design and execution of the clinical trial as a valid experiment. The recommendation by Stefanek et al that clinical trials of psychological interventions for breast cancer patients should be delayed until the behavioral mechanisms of a treatment or causal biobehavioral pathways are understood well may have unintended consequences. Clinical trialists will continue to evaluate interventions, as they should, but may pursue unnecessary ancillary data collection that fosters a continued imbalance between attention afforded to the study of hypothetical moderators and mediators and the needed emphasis on the trial's primary objectives.
What, then, is the promise of psychological interventions for breast cancer patients? First, we must acknowledge that most if not all “psychosocial” or “behavioral” interventions cannot be regarded as equivalent. We cannot assume a class effect for drugs, and we should not assume a class effect for behavioral interventions. They differ widely in content, intensity, and the degree to which they address behavioral characteristics that actually have relevance to the patient. In considering the next steps, attention should be directed to all of these factors and to a diagnosis of the patient's needs.
The trial by Anderson et al lacked statistical power to be definitive, even under the best of circumstances. Nonetheless, the apparent improvement in disease-free survival of approximately 6 months in the intervention group is a clinically meaningful signal whether it was achieved through the hypothesized biobehavioral pathways or through adherence to the medical regimen. The 54 observed events and the sample of 227 women were far too small. Yusuf et al13 estimated that 350 to 650 events are needed to detect a convincingly significant reduction of approximately 25% at P < .01, consistent with the design of the trastuzumab trial.14 To their credit, Andersen and colleagues readily caution that, although their psychological interventions improved mental health and treatment-relevant behaviors, these results and the possibility that they improve the survivorship of cancer patients require replication. Much remains to be done, and the work must continue.