Does stress cause cancer?

In 1893 Snow presented what might be the first statistical summary of the psychological characteristics of patients with breast or uterine cancer.¹ Some 250 women with these cancers were described as having a “general liability to the buffets of ill-fortune.” Over 100 years later we still find researchers preoccupied with showing whether stressful life events are related to cancer—as in this week’s study by Protheroe et al (p 1027).² Many clearly believe that life is more stressful than ever before and that one consequence of this ubiquitous stress is disease, including cancer. Sontag describes this as a metaphorical view of disease as the “outward expression of character.”³ In more practical terms, patients with cancer may believe that their disease results from too much stress and relatives may feel guilty for contributing to the emotional ill health of the patient. Such beliefs may also have a bearing on what people do about seeking and sticking to treatment. It is important therefore to have a clear idea of what the evidence does show.

Two recent literature reviews have concluded that there is no good evidence for any relation between stressful life events and breast cancer,⁴,⁵ and both point out that the typical methods used in studies of the relation are problematic at best. What then should we make of this most recent study? The methods used are fairly well in line with previous research.² Women attending breast clinics in west Leeds after discovering a suspicious breast lump but before learning the outcome of biopsy were asked about life stresses in the previous five years. Biopsy outcome then identified those with malignancy (106) and those with benign disease (226). Women with malignancy were no more likely to experience one or more severe life events (adjusted odds ratio 0.91) or severe difficulties (odds ratio 0.86) in the previous five years than those with a benign lump.

While consistent with the recent literature reviews, these findings stand in contrast to an earlier report by Chen et al, in the BMJ, using much the same methods, which suggested that women with breast cancer were nearly 12 times more likely to experience severe life events over the same period before diagnosis.⁶ Why the discrepancy and what do these findings tell us about the relation between life events and breast cancer?

It is arguable whether the methods used in either of these studies could ever represent an adequate test of the hypothesis of a link between stress and cancer. Retrospective recall of life events in the five years before learning whether a breast lesion is malignant or benign constitutes a relatively weak test of the hypothesis, compared with good prospective studies. In Protheroe et al’s study, even this most basic safeguard against recall bias was ignored as 30% of the women with cancer knew their diagnosis by the time they were interviewed.²

Two other features of the two studies are worth comment. Both are described as case-control studies but might be better described as cross sectional. In a true case-control study the controls are drawn from the same population as the cases. However, the women with cancer are considerably older—an average of 10.6 years in the study of Protheroe et al and 7 years in that of Chen et al. Many of the other studies using these methods report similar age differences.⁵ It is unclear to what extent these studies can control for such large age differences in their analyses. This is important as age relates directly to risk of breast cancer and to experiencing particular life events.⁵

Both studies also use multivariate modelling with many predictors and relatively few outcome events—that is, cancers. A general rule of thumb is that there should be at least 10 outcome events for each predictor entered into the model,⁷ so the multivariate analyses in these studies are probably overfitted and the estimates unstable. This is particularly so in the study of Chen et al, where12 predictors were entered into a model based on 41 women with cancer. While adjustment for other factors should lead to more precise estimates of effects, the reverse is true in their analysis, with the unadjusted odds ratio increasing from about 3 to 12 in the adjusted model, with a correspondingly large increase in the confidence interval surrounding the estimate. This suggests that life events are so correlated with one or more of the other variables that it is difficult to disentangle their effect. Similar criticisms apply to Protheroe et al’s study, with 19 predictors entered into their model.

It is easy to go on picking holes in the methods of these types of studies—and perhaps unfair. One difficulty is that the hypothesis being tested is so vague. This is not the fault of the authors; the literature has not developed much beyond such vagueness. Any hypothesised relation does not seem to relate to cancer causation (causative factors may well be operating many years before detection) but may have something to do with stress accelerating the development of lesions or otherwise influencing the probability of diagnosis. The hypothesis needs to be stated in some more biologically plausible form to allow a stronger test of the association. Prospective longitudinal designs would be a good place to start.

Already some indications exist from prospective studies that there is no relation between stressful events and cancer. The results of a large scale study in the United Kingdom provide little evidence for an association between bereavement in men or women and later cancer.⁸ Other research has investigated the long term outcome for prisoners during the second world war and the Korean war.⁹ These men clearly suffered extremes of physical and mental hardship, and though they showed excess mortality due to accidental injury, suicide, and cirrhosis of the liver—suggesting continuing psychological distress—there was no excess mortality due to cancer. A second longitudinal study of Japanese men living in Hawaii showed no relation between stressful life situations and later cancer.¹⁰

Recriminations over real or imagined life stress may be counterproductive for individuals with cancer and their families. They should be reassured that the available scientific evidence does not support any direct role for stressful life events leading to a diagnosis of cancer.

Papers p 1027