The idea that by adopting a fighting spirit, cancer patients improve their chances for survival predates its endorsement by positive psychology. Yet, if true, the idea epitomizes the triumph of character and attitude over biology that is so key to a positive psychology of cancer. Fighting spirit is characterized by patients optimistically viewing cancer as a challenge and having a determination to fight the cancer and not to allow it to disrupt their lives. It was first conceptualized as an attitude toward cancer, but later as a positive coping style. An early study [11
] was widely interpreted as demonstrating that patients having a fighting spirit were more likely to be disease-free and to survive to 5 and 10 years post-assessment. However, these claims were based on a small investigation that used a brief open-ended question assessment in a sample of 57 early-stage breast cancer patients and which lacked statistical control for node status. A later, larger study (n
578) that used a self-report measure failed to find a prognostic value for fighting spirit [12
]. The investigators expressed relief: “Our findings suggest that women can be relieved of the burden of guilt that occurs when they find it difficult to maintain a fighting spirit” ([12
A systematic review [13
] of effects of coping on cancer identified 12 studies examining the prognostic value of fighting spirit for cancer progression and survival, with most of them having negative findings. The larger studies were uniformly negative, with the two positive studies being smaller and methodologically flawed and with inadequate statistical control of biomedical and treatment confounds. Overall, Petticrew et al. [13
] found little evidence of the influence of psychological factors on cancer progression and survival, leading the authors to conclude: “People with cancer should not feel pressured into adopting particular coping styles to improve survival or reduce the risk of recurrence” (p. 1066). Since the Petticrew et al. review, a 10-year follow-up of the Watson et al. [12
] study findings again found that fighting spirit conferred no advantage for survival [14
], consistent with another large study of women with early-stage breast cancer [15
Claims about the physical health benefits of having a fighting spirit persist in the literature concerning a positive psychology of cancer [14
], but most commentators without such an allegiance agree that the accumulation of studies have demonstrated the fighting spirit lacks value as a prognostic factor in cancer, much less as a causal factor. Yet, there is still little appreciation of how unrealistic the idea of being able to demonstrate such an effect was from the beginning. As Watson et al. [12
] noted, their study of 559 early breast cancer patients generated so few cases of progression and death in 5 years that, in order for a valid effect of fighting spirit to be demonstrated, adopting such an attitude would have had to be more potent than adjuvant therapy. For a valid effect to have been demonstrated in Greer et al.’s [11
] more modest sample of 57 women with early breast cancer, effects of a fighting spirit would have to be even much stronger than most known biological factors or treatments.
On an a priori basis, one might assume that it should be easy to demonstrate that positive psychological characteristics predict subsequent physical health, but upon reflection, it is also clear that it might be difficult to establish a causal connection. People possessing material and social resources are more likely to report positive psychological states, and having these external resources can be related directly to better subsequent health, with any association observed between positive psychological states and health potentially being spurious. Similarly, people in better health may report better positive psychological states and better initial health predicts better later health. They report better psychological states because they feel better physically and because they may also possess other information about their health status, as when cancer patients are informed of the treatment they will receive and their prognosis. Thus, in attempting to demonstrate that positive psychological characteristics cause better health, it is important that baseline differences in resources and biomedical and treatment variables be specified, adequately assessed, and statistically controlled. Otherwise, there is the likelihood that positive psychological states will only appear to predict subsequent health because of residual confounding, i.e., the failure of statistical controls. Of course, demonstration that direct manipulation of psychological states produces changes in health would be persuasive, but there are few such demonstrations, and so, we must more typically rely on statistical manipulation of what are essentially observational correlational data.
An analogy can be made with studies examining self-perceived health (SPH) or physical health quality of life (PHQoL) as predictors of subsequent health. Even with statistical control of available disease and treatment variables, these self-reports very often remain significant predictors of mortality among chronically ill populations [16
]. This is because available control variables do not completely capture disease severity, treatment intensity, or medical history. For instance, pretreatment clinical tumor staging is often used to control for severity of disease, but provides an incomplete summary of overall disease burden. Thus, introducing stage as a control variable typically leaves considerable residual variance in disease severity unexplained.
Variables such as SPH or PHQoL are confounded with disease and treatment variables, particularly when patient perceptions are informed by medical opinion or awareness of the extent of illness, allowing these self-reports to account for residual variance in mortality. Yet, no one would advocate attempting to influence patient self-report, independent of medical treatment in order to improve survival because these variables are not in themselves causal—confounding with variance resulting from incomplete control is what provides the prognostic value of SPH and PHQoL. Extending this analogy to positive psychological states, some relationship to mortality through health and behavior history or current disease and treatment status is to be expected and does not necessarily indicate anything theoretically interesting or clinically useful. Pressman and Cohen [17
] make similar arguments in expressing “serious conceptual and methodological reservations” (p. 960) about the literature relating positive affect to physical health.
Claims are nonetheless widespread that positive psychological factors slow progression, increase survival time, and reduce mortality in cancer. However, Chida and Steptoe’s [18
] systematic review of positive psychological factors and survival identified 35 studies of mortality in samples of patients having particular chronic illnesses. Their meta-analysis then examined 19 studies of which five were cardiovascular disease and six were cancer, and null results were obtained for both conditions. Yet, in seeming contrast, Chida, Hamer, Wardle, and Steptoe [19
] investigated the contribution of positive (i.e., fighting spirit and social support), ambiguous (locus of control and extroversion), and negative (stressful life events and negative coping style) stress-related factors to cancer incidence and mortality. They indicated in their abstract that they had found highly statistically significant associations for incidence of cancer in initially healthy populations, (p
0.005), shorter survival time for persons diagnosed with cancer (p
0.001), and higher rates of cancer mortality (p
0.001). Chida et al. qualified their conclusion, providing statistical evidence of a publication bias, and so, these results should be interpreted with caution.
Nonetheless, these claims warrant scrutiny because they are provocative, fuel existing beliefs that psychological variables influence progression and outcome of cancer, and are already being cited uncritically as the justification for preventive and therapeutic intervention [20
]. These meta-analyses aggregate a considerable number studies, 70% of which individually yielded null findings. The meta-analysis for incidence of cancer involved integrating 142 studies with an average sample size of 87,062; for survival time, the integration of 157 studies with an average sample size of 418; and for cancer mortality, the integration of 50 studies with an average size of 93,059. With such huge aggregate samples, levels of statistical significance are much less informative than hazard ratios (HRs) and confidence intervals (CIs). For cancer incidence, the results were HR
1.02–1.11); for survival, HR
1.02–1.04); and for mortality, 1.29 (CI
1.16–1.44). The CIs for incidence and survival barely exceed the 1.0 that would have rendered them nonsignificant. Certainly, finding such an HR would not typically generate much excitement in the larger epidemiological literature. Recall too that Chida et al. acknowledge finding statistical indications of a publication bias, i.e., that would be “likely to create a positive result bias if authors are more likely to submit, or editors to accept positive rather than positive than negative or inconclusive results” (p. 473). A positive publication bias could easily explain such modest effects, especially those for incidence and survival.
Yet, further scrutiny casts even greater doubt upon the validity of Chida et al.’s [19
] conclusions, in part because of decisions made in integrating these studies in meta-analyses and in part because of the poor quality of many of the studies on which the meta-analyses draw. Chida et al.’s meta-analyses were conducted incorrectly because they counted a number of samples multiple times, effectively treating each as an independent cohort, rather than more appropriately entering one effect size per cohort [22
]. Thus, one study of the effects of the stressor death of a child on cancer mortality [24
] was counted eight times because two ways of dying, accidental or in war, and four different cancer sites were examined. Another study [25
] was counted 22 times because the effects of death of a spouse and death of a child were each considered for different cancer sites. The “stress-related variables” that were considered equivalent for the purposes of meta-analysis (such as fighting spirit, shift work, death of child, neuroticism, and the MMPI Lie scale) were so highly heterogeneous as to defy any integrative theoretical interpretation of the omnibus effect size produced by a meta-analysis. Few of these studies provided even minimal statistical control over potential medical confounds: only 28% of the incidence studies, 1% of the survival studies, and 40% of the mortality studies. Many of the outlier highest estimates of an association between a stress-related variable and cancer came from underpowered studies with inadequate statistical control.
Chida et al.’s [19
] analyses were seriously flawed, but the methodologically limited and highly heterogeneous set of studies on which their analyses drew would pose challenges to making any integrative summary statement. Moreover, consistent with Pressman and Cohen [17
], the problems that our cursory review of the problems in the available studies that Chida et al. encountered do not bode well for making any substantive interpretation of the literature relating psychological variables to cancer outcomes.
Recently, there were hopes that newly emerging data from the large Women’s Health Initiative might shed light on the role of positive and negative factors in incidence of breast cancer. Michael et al. [27
] followed a sample of 84,334 women for an average of 7.6 years during which 2,481 invasive breast cancers were diagnosed. The unanticipated results were that, in this sample, low social support and high life stress seemed to protect against the development of breast cancer.