Research Design: A Role for Experimental Trials of Psychological and Behavioral Interventions
We have previously reviewed the evidence for psychological and behavioral interventions with cancer patients (see Andersen, 1992
, for a discussion). Several controlled trials have demonstrated that such efforts can reduce distress, hasten resumption of routine activities, and improve social outcomes for groups at high risk for quality of life morbidity, such as patients with disseminated or recurrent disease, as well as patients at lower risk, such as those with localized disease and time-limited cancer therapy. We have suggested here that appropriately designed psychosocial interventions can reduce stress and enhance quality of life as well as improve behavioral responses, such as health behaviors and compliance. This research progress in behavioral oncology—the prospect of randomly assigning individuals to conditions that will result in differential psychological and behavioral outcomes— provides one of the necessary conditions for an experimental test of the model.
Therapy components for psychological interventions have included an emotionally supportive context to address fears and anxieties about the disease (e.g., Cain, Kohorn, Quinlan, Latimer, & Schwartz, 1986
; Capone, Good, Westie, & Jacobson, 1980
; Forester, Kornfeld, & Fleiss, 1985
; Maguire, Brooke, Tait, Thomas, & Sellwood, 1983
), information about the disease and treatment (e.g., Cain et al., 1986
; Fawzy et al., 1990
; Houts, Whitney, Mortel, & Bartholomew, 1986
; Jacobs, Ross, Walker, & Stockdale, 1983
; Maguire et al., 1983
), behavioral coping strategies (e.g., role playing difficult discussions with family or the medical staff; Fawzy et al., 1990
; Houts et al., 1986
), cognitive coping strategies (Cain et al., 1986
; Davis, 1986
; Houts et al., 1986
; Telch & Telch, 1986
), relaxation training to lower "arousal" or enhance one’s sense of control (Davis, 1986
; Fawzy et al., 1990
), and focused interventions for disease-specific problems (e.g., sexual functioning for gynecologic or breast cancer; Capone et al., 1980
It is more difficult to enumerate the intervention components that can affect health behaviors. Despite their importance, health behaviors have not been included as outcomes in cancer studies, even though many psychosocial interventions include educational components designed for them. Of the very few studies focusing on compliance per se, the data suggest similar interventions, including information about the disease and treatment (Richardson et al., 1987
; Robinson, 1990
), enlistment of help of significant others (i.e., social support; Richardson et al., 1987
), and practitioner counseling (i.e., prompts by a physician to be compliant; Robinson, 1990
). For reference, the extensive literature on compliance with antihypertensive regimens (e.g., Dunbar-Jacob et al., 1991
) suggests different techniques for different compliance targets. Specifically, mailed reminders and home visits by nursing personnel have been used to improve appointment keeping, whereas education, behavioral strategies (e.g., self-monitoring, contingency contracting), and enhanced social support have been used to improve medication compliance.
In considering an experimental trial to affect immunity or disease endpoints, a simple experimental design—treatment versus no treatment—would be the strategic next step. At present, there are insufficient data to choose among intervention components that would be expected to affect the immune system, but these and related findings would suggest an emphasis on relaxation, coping, social support, and disease-specific components (Andersen, 1992
; Kiecolt-Glaser & Glaser, 1992
). Such a design would not provide the basis for ruling out secondary hypotheses of therapist characteristics separate from treatment techniques, patient characteristics separate from psychological and behavioral difficulties, or specific therapeutic techniques as separate from nonspecific or placebo effects. What it can do, however, is establish cause–effect conclusions for the presence of intervention-producing enhanced psychological and behavioral outcomes, immune responses, and health effects. Once an effect is reliably demonstrated, it would then be relevant to study component questions. In the interim, investigators should document the content, the reliability of intervention delivery, and the involvement of the patients, for hypothesis generation in follow-up studies. Previous studies have omitted documentation and process measures (but see Gordon et al., 1980
, for an example of number of individual therapy sessions; or Telch & Telch, 1986
, for monitoring of homework assignments).
There already is suggestive evidence in support of the model, with data that indicate positive immune and health consequences for psychological interventions, and taken together they provide a basis for further scientific inquiry. One study included immune measures and disease endpoints, and two others reported disease endpoints. All of the studies used extensive psychological assessments, but none examined behavioral variables (i.e., health behaviors and compliance). For the first study, Fawzy et al. (1990)
studied newly treated Stage I or II melanoma patients randomly assigned to either no intervention or a structured short-term (10 sessions) group support intervention. At posttreatment and the six-month follow-up, significant psychological and coping outcomes for the intervention subjects were evident, as well as increases in the percentage of large granular lymphocytes, the NK cell phenotype, an increase in NK cell numbers (as determined by markers), and other positive findings, such as interferon alpha-augmented NK activity. These data are relevant to the findings of Kiecolt-Glaser et al. (1985)
, who found intervention differences in NK activity. Importantly, the magnitude of the NK changes Fawzy et al. found was frequently greater than 25% for the intervention subjects. Finally, the correlation data of immune and affective change provides additional support for the model in that interferon-augmented NK cytotoxic activity increased with concomitant reductions in anxiety (−.37) and depression (−.33). We believe the NK cell data are particularly important, because it has been shown that there is a reduction in NK cell activity with tumor progression (Akimoto et al., 1986
; Takasugi, Ramseyer, & Takasugi, 1977
). It is also known that the ability of NK cells to respond to interleukin-2 (IL-2) or gamma interferon is different in cancer patients who are managed with different types of therapy.
Six-year follow-up data on disease endpoints are also available (Fawzy et al., 1993
). Analyses of DFIs to death have indicated significant group differences, with 29% of controls and 9% of experimental subjects dying in the six-year interval. Analyses of DFIs to recurrence were in the same direction but only approached significance (p
= .09). Follow-up analyses suggested that the former effects were primarily due to the men in the control group dying and, in particular, men with the highest Breslow depth rating (a disease-related prognostic indicator; higher values indicate poorer prognosis). Considering the latter factor, 9 of 10 experimental subjects in the highest Breslow category were alive, versus only 1 of 9 control patients.
In the same follow-up, other analyses examined baseline and six-month psychological and immune parameters by survivor group and gender. The majority of the effects reside within the male group, so it is useful to consider the surviving versus the deceased males. These comparisons indicate that from baseline to the six-month assessment, the surviving males reported significant decreases in affective distress, increases in active behavioral coping, and increases in CD 16 NK cells and interferon alpha-augmented NK cell activity (i.e., immune up-regulation). In contrast, males who died showed no significant changes on any of these variables, that is, no quality of life improvement or immune enhancement. It also should be noted that the deceased males began the study reporting significantly lower overall levels of distress (i.e., 34 vs. 64 on the baseline Total Mood Disturbance of the POMS for the deceased vs. surviving males, respectively), but their immune responses were within the same range as those of the survivors.
Fawzy et al.’s (1990
) investigation was the first intervention study to combine psychological, immune, and disease endpoint data. The initial outcome data indicated that early, brief psychological efforts produced immediate (posttreatment) effects as well as long term (six-month) changes. We have suggested that the maintenance of gains may be a crucial foothold for immune effects to emerge (Andersen, 1992
); data from a relaxation intervention study suggest this as well. Gruber et al. (1993)
studied 13 Stage I, node-negative breast cancer patients who received electromyograph (EMG) biofeedback-assisted relaxation training. Weekly immune assessments during the nine-week intervention indicated significant differences between the treatment and control groups in the expected direction: WBC values were stable for the intervention group but declined for the control, and Con-A and mixed lymphocyte response values were higher for the intervention group. NK cell values significantly increased from pre- to posttreatment. Considering immune data taken each of the nine study weeks, there was suggestive evidence that the immune effects became evident only after several weeks into the intervention.
Other data relevant to the model come from samples very different from these good prognosis patients. Specifically, women with recurrent breast cancer and lung cancer patients have been studied. Spiegel, Bloom, and colleagues (Spiegel, Bloom & Yalom, 1981
; Spiegel & Bloom, 1983
) randomly assigned women with metastatic breast disease to a no-treatment condition or to a group treatment condition that met weekly for at least one year. The intervention subjects were also randomly assigned to a no additional treatment condition or to a self-hypnosis condition for pain problems. The intervention group subjects reported significantly lower emotional distress (POMS) and fewer maladaptive coping responses than did the control subjects, with the magnitude of the difference increasing during the intervention year. Data also suggested that the hypnosis component provided an additive analgesic effect to other group treatment components. A 10-year follow-up was conducted, at which time only three women, all of whom were intervention participants, remained alive (Spiegel et al., 1989
). A striking survival difference was found between the control subjects (18.9 months) and the intervention subjects (36.6 months) from study entry until death. Survival time differences between the groups began to emerge approximately 8 months after termination of the year-long intervention.
As might be expected, the publication of Spiegel et al.’s (1989)
survival follow-up study unleashed a torrent of interest in the role of psychological factors in cancer in both academic and popular circles (e.g., Moyers, 1993
; ten Have-de Labije & Balner, 1991
; Temoshok & Dreher, 1992
). One critic of the findings, however, was LeShan (1991
, who suggested that the survival effects might have been due to adverse effects on the control group (i.e., perceived rejection implied by being randomly selected out of the treatment group) rather than to positive effects on the intervention group. In fact, Fox’s (1992)
comparison of the study’s survival data with national survival rate data has supported the implication of LeShan’s concern: “The survival of the intervention group was somewhat worse than the survival of the national sample, while the survival of the control group was considerably worse (p. 83).” Despite this, Fox argued that it is unlikely that the control group died at a faster rate for “perceived rejection” reasons. He posited that this explanation is unlikely for several reasons, including the lack of national enthusiasm for group support interventions at the time of the study (late 1970s), no suggestion at all to the study participants that the study had any relevance to survival, and, more likely, that the differences reflect chance deviation of the entire study sample (consisting of only 84 women) from the population of women represented in national trends.
Finally, contradictory data come from Linn, Linn, and Harris (1982)
, who offered a death and dying intervention program to male cancer patients, 46% of whom had lung cancer. Despite favorable quality of life outcomes for the intervention subjects (e.g., lower distress [POMS], higher life satisfaction, lower alienation), there were no functional status or body system impairment differences between the control group and the intervention group. Survival analyses also revealed no significant differences. Aside from the many methodology differences between this and the Spiegel et al. (1989)
study, two disease factors may account for the discrepancy in survival outcomes. First, there is a shorter survival window for metastatic lung cancer in contrast to metastatic breast cancer. (Five-year survival rates are 13% and 73%, respectively, for initial Stage III disease and 1% and 19%, respectively, for initial Stage IV disease; Boring et al., 1993.) Second, hormonal and immune factors may be more important in breast cancer than in lung cancer (see subsequent discussion below).
Methodology: Maximizing the Signal to Noise Ratio
It is an understatement to characterize this as “a most difficult area of research” (Fox, 1978
, p. 117), and one that, like AIDS and HIV, is methodologically complex for behavioral immunology researchers (Kiecolt-Glaser & Glaser, 1988a
). We cannot adequately address all of the methodology challenges that tests of the biobehavioral model pose; however, we reference here some of the more difficult ones at the interface of behavioral oncology and immunology and we refer the reader to other methodology discussions of variables predicting risk for psychosocial morbidity, and individual-differences variables that may covary with psychosocial outcomes or intervention effectiveness (Andersen, 1994
, in press
First, the term cancer
refers to a heterogeneous group of diseases of multiple etiologies that vary in their tissue of origin, cell type, biologic behavior, anatomic site, and degree of differentiation (stage and degree of malignancy). Although we have used the generic term cancer
in this article, it is likely that quality of life and stress and health behavior factors would interact with immune function only in selected cases. Many hypothesized cancer-causing mechanisms are associated with immunological down-regulation. However, the likelihood of psychological or behavioral factors interacting with the immune system to influence disease progression would be expected to differ across sites. Cancers that are etiologically linked to hormonal stimuli (e.g., breast, ovarian, endometrial, and prostate) or to the immune system (e.g., leukemias, lymphomas) may be most susceptible to influence; Epstein-Barr virus (EBV)-associated tumors (i.e., EBV-associated B cell lymphomas; Levine, Ablashi, Nonoyama, Pearson, & Glaser, 1987
), viral (e.g., cervical; Goodkin, Antoni, Sevin, & Fox, 1993a
), and genetically linked forms (e.g., some types of breast and colon cancer) may be somewhere in the middle, and those cancers believed to be due to physical or chemical carcinogens (e.g., lung cancer linked to tobacco usage) may be the least susceptible to influence. Also, it is known that the risk of cancer increases with age, and as the immune system ages it becomes less efficient. It has been suggested that psychological and stress factors may be most relevant for middle-aged persons (ages 35–65) rather than for very young or very old persons, because of the disproportionate influence of hereditary factors on young cancer patients and the influence of aging factors on older persons (Fox, 1978
Thus, the proposed model may evidence the best fit for some sites (e.g., breast, ovary, prostate), and even for some forms within sites, as opposed to others. Application of the model to other chronic illnesses would require further refinement. Testing of the model might be additionally optimized when samples are as homogeneous as possible on other major dimensions, such as prognostic factors; such variables might be chosen for stratification or at least for documentation (e.g., the case with Breslow depth in the Fawzy et al., 1990
, study). Selection of variables would be based on their unique importance to the disease site being studied. This is the same tactic taken in moderate-sized clinical trials of medical therapies in which the effect of the prognostic variables is anticipated to be greater than the effect of the new cancer treatment. For most sites it is more feasible to stratify on disease or prognostic variables than to attempt to control for cancer treatment. There are diverse treatment regimens now available, and choices among them are made on several bases, including, but not limited to, current data on the treatment regimens’ relationship with prognostic variables; patient choice, if possible; physician preference, expertise, or specialty; and data from new trials (e.g., new uses of Taxol are occurring on a monthly basis). Because this is the scenario for most of the prevalent disease sites (e.g., breast, prostate, colon), careful selection of major prognostic indicators can result in de facto control of treatment, even when the available cancer therapies change during the course of the study. For example, the variables of nodal status, hormone receptor status, and menopausal status might be considered for a study of women with breast cancer (see Clark & McGuire, 1992
, for a discussion). These variables would, in various combinations, determine most of the treatment pathways, influencing, for example, the occurrence and extent of surgery, the type of chemotherapy for premenopausal (adjuvant chemotherapy) versus postmenopausal women (adjuvant Tamoxifen), or the likelihood of extreme treatments, such as bone marrow transplant.
Second, the model posits immune effects from psychological and behavioral factors beyond any immune downregulation that may accrue either from malignant disease processes or from cancer treatment effects. Some researchers believe intervention trials of this sort are misguided at best, because the immunosuppressive effects of the disease or treatments would override any positive effects from a psychological intervention. In fact, attempting to address this concern with evidence, such as determining the magnitude of immunosuppressive effects from disease or cancer treatments, is surprisingly difficult because of the dearth of basic immunology data. Also, data that is available is not always confirmatory. For example, Ludwig et al. (1985)
found unaltered immune function in patients with nondisseminated breast cancer (Stages I through III) at diagnosis, with the significant reductions (e.g., depressed PHA responsiveness) found only among women with metastatic (Stage IV) disease. Regarding cancer treatments—surgery, radiotherapy, chemotherapy, chemoradiation (chemotherapy that is radiosensitizing and given with radiation), and combination therapy—all have immunosuppressive effects, but much of the detail about the nature of the effects is unknown. It has been found that lymphocyte transformation is depressed during radiotherapy but may rebound within two months after therapy; there is also specificity due to site of treatment, as greater depression is found for pelvic–abdominal sites versus chest or head–neck sites (Slater, Ngo, & Lau, 1976
). Considering chemotherapies, some will suppress lymphocyte proliferation, yet others are designed to enhance lymphocyte proliferation; but more typically, the immune-moderating effects of most chemotherapies are unknown. One of the more well-studied drugs (from an immune standpoint) is cyclophosphamide (CY). It consistently causes a sharp reduction in circulating peripheral blood lymphocytes and lymphoproliferative responses to mitogens, although the effect on antibody production is more variable (Ehrke, Mihich, Berd, & Mastrangelo, 1989
). In contrast, CY can augment immunity to clinically relevant antigens; the leading hypothesis for how this occurs is that CY has selective toxicity for suppressor T cells or their precursors. Another drug, less well-studied from an immune standpoint, but one as widely used clinically, is adriamycin (ADM). The most prominent immunosuppressive effect of ADM is that it induces my elosuppression, which is likely due to the well-established antiproliferative effects of such anthracycline antibiotics. Although the immunopotentiation effects have not been studied as extensively, clinical studies with cancer patients have suggested that long-term ADM therapy does not appear to alter cell-mediated immunity, but data suggest that recovery is complete by one to three weeks after therapy (Kempf & Mitchell, 1984a
We acknowledge the complexity of these issues and suggest special care in the selection of stratification variables (see prior discussion) so that it might be possible to de facto equate groups on the heterogeneous treatment options that might be available for a single site of disease. Of course, full documentation of the nature of the regimens for subjects, including dosages and timing of delivery, is essential. In addition to following basic guidelines for behavioral immunology studies (Kiecolt-Glaser & Glaser, 1988a
), investigators will need to consider strategies for controlling variation in blood draws for cancer patients; for example, scheduling blood draws before chemotherapy administration may maximize the likelihood of tapping recovered responses at the end of cycles rather than any acute dysregulation with drug administration per se. Multiple blood draws to monitor immunity after patients are off therapy will also be important. Unfortunately, contextual factors relevant to blood draws may also be important for chemotherapy patients. Data from Bovbjerg et al. (1990)
suggest that psychological factors may operate for conditioned immune suppression following cytotoxic chemotherapy much as they do for other chemotherapy side effects, such as anticipatory nausea and vomiting.
Third, skeptics often note a related concern—the magnitude of change from psychological and behavioral factors that would be needed to affect immunity (or cancer progression) is unknown. We agree that such data are unavailable. However, the absence of this data is not unique to the PNI field but characterizes much of the basic research in immunology and cancer. For example, there are no such data linking changes in immune responses and disease progression for many of the current biological response modifiers (e.g., lymphokine-activated killer cells or interferons) being tested as cancer therapies. Immunotherapies are tested on the basis of their mechanisms at the level of the cell, and their effect on clinical outcomes (i.e., disease progression) is unknown but is seen as the relevant question to determine through clinical trials. The inability to specify the magnitude of change is not unique to this paradigm test for a psychological therapy, as the same criticism could be leveled against the testing of any new chemotherapeutic or chemopreventive agent. To illustrate, the current testing of Tomoxifen for breast cancer prevention is based on relevant but indirect lines of support. There is experimental support that Tomoxifen affects both the initiation and promotion of tumors in animal studies, and it has lengthened disease-free survival and reduced the incidence of contralateral disease in women with breast cancer (Fisher, 1991
; NSABP Protocol P-1). These data (and the fact that toxicity from Tomoxifen is low) were sufficiently encouraging to embark on a chemopreventive trial with 16,000 healthy women—even without two critical lines of evidence: (a) the precise mechanisms through which Tomoxifen achieves its effect are unknown and (b) there is only limited support for Tomoxifen’s ability to increase survival rates in women with breast cancer (Fisher, 1991
). Thus, it would seem unusual to hold tests of psychological therapies (which by their nature are nontoxic) to higher standards than the ones used for clinical trials of immunotherapies, new agents, or new uses of old agents.
Finally, a related methodologic concern is that the effects of stress on immune responses are small and are usually within normal ranges. We agree with this characterization, as it holds for the correlations with stress (Herbert & Cohen, 1993b
), the psychological intervention studies (Kiecolt-Glaser & Glaser, 1992
; Kiecolt-Glaser, Glaser, et al., 1985
), and experimental studies of self-disclosure (Pennebaker, Kiecolt-Glaser, & Glaser, 1988
) or relevant (personality) disclosure styles (Esterling, Antoni, Fletcher, Margulies, & Schneiderman, in press
). But in all of these lines of research—research programs that use different paradigms, different subject populations, and documentation of psychological changes—the findings are consistent for the direction of the effect (i.e., immune downregulation with heightened distress) and for the effects to covary with experimental manipulation via psychological interventions or methods (e.g., ranging from brief self-disclosure for healthy individuals to eight-week interventions for cancer patients). Thus, although the immune effects are small, they are robust across samples and manipulations. Furthermore, these supportive findings include studies with cancer patients (Fawzy et al., 1990
) and the same type of manipulations (Esterling et al., in press
; Kiecolt-Glaser, Glaser, et al., 1985
) that would be offered in psychological therapies.