The Role of Psychological Factors in Cancer Incidence and Prognosis
The Role of Psychological Factors in Cancer Incidence and Prognosis
The relationships between psychological variables and the presence of cancer, its prediction, and the prediction of cancer mortality and course of disease have been studied extensively. From a limited list of about 50 such variables, the following have been the focus of the most intensive research and are discussed in this report: human and animal stress; bereavement; depressed mood; psychosis, especially schizophrenia; suppression of emotions, especially anger; helplessness and hopelessness; social support; and psychotherapeutic intervention. For all of these variables, studies have shown both positive relationships and absence of relationships. The evidence against any such relationship is strongest for human stress, depressed mood, psychosis, and bereavement. Studies of animal stress show that it stimulates the development of cancers of viral origin and exacerbates their growth, while inhibiting the development and progression of chemically induced cancers. For the other factors, the literature remains contradictory.
This article reviews the literature regarding the possible effects of various psychological factors on cancer incidence and prognosis. For a clear understanding of the findings in the literature, a number of terms need to be defined. The term "psychological factors" refers to personality or behavior traits or reactions. In studies of psychological factors and cancer, these are the independent variables, with one restriction: If a psychological factor is associated with a physical carcinogen, it will not be considered an acceptable independent variable, although it may be regarded as a possible confounder. For example, it is claimed that certain traits predispose a person to smoking. Those traits will not be of interest insofar as they affect smoking and its carcinogenic consequences, but only insofar as they may affect cancer independently of smoking.
The dependent variables are cancer incidence, mortality, and prognosis (disease-free interval and survival time). Possible confounders, cofactors, or primary etiologic agents that may act in conjunction with psychological factors include familiar factors such as radiation, genetic attributes, viruses, smoking, and exposure to chemicals, as well as some less familiar factors.
Incidence of cancer is determined in two ways, either by clinical diagnosis or at autopsy. In both cases, the malignancy has had to grow to the point of detection. That is, if it is a blood or lymphatic cancer, enough cells have been transformed to be detected, and if a solid tumor, it has grown large enough to be detected by x-ray, palpation, sonogram, visualization, or some other method. The first actual mutation of a cell to malignancy, however, has taken place long before-in rare cases months before, and in most cases years before [1,2]. For example, the median time to detection of leukemia incidence after the atomic bomb was dropped on Hiroshima was between 6 and 7 years . The estimated median growth time to detection for breast cancer is 7 to 11 years, depending on the researchers [1,4]. According to Steel , in a mix of several cancers, with lung heavily represented, the median was about 5.5 years. Thus, all incidence statistics are a combination of the first mutation of a cell to cancer and the progression of that cancer to detectable size.
Another issue is type of study. The two classic types of epidemiologic studies are case-control, sometimes called retrospective, and cohort, sometimes called prospective. In case-control studies, a group of cancer patients, cancer survivors, or patients with recurrence is selected, together with a control group, and the researcher measures some putatively discriminating attribute in both groups to see whether they differ in respect to that attribute.
In cohort studies, an attribute is measured in all members of the cohort before the outcome of disease, death, or recurrence, and the researcher waits to see who gets cancer or not, survives or not, or suffers a recurrence or not. The pair of groups being tested is examined for differences in outcome that may have emerged, given the existence of differences in the attributes originally measured. For example, if the pair of groups in the population is defined as "those who are stressed" and "those not stressed," the risk factor is the condition of being stressed, and the control condition is not being stressed. The proportion who get cancer during the given follow-up period in the group with the risk factor is compared with the proportion who get cancer during this period in the group without the risk factor.
This distinction between types of study is important, because the case-control type, when used only on a sample and not the population, is subject to a number of possible biases. Such potential biases make it difficult to decide which case-control studies to accept, doubt, or reject. Cohort studies, which are also subject to some biases, but not nearly so many or so damaging as those characterizing many case-control studies, are more trustworthy and involve less risk of a false conclusion.
A variety of psychosocial factors have been used as independent variables in the past. These include not only directly measured variables but also those inferred from results on an instrument that provides an indirect measure (eg, the Rorschach test). In most case-control studies, the authors imply that because they found a relationship, the factor existed before the tumor was discovered, and they seldom remark on whether it existed before malignant transformation. The analyst is left hanging in the air because the researchers had not learned enough about the disease to know the meaning or implication of their finding.
The Table lists a number of psychosocial factors (with the most important items discussed below). Each of them has been examined at least once as a risk factor in some study in the literature (see Fox  for a bibliography). Very few mention the possibility that the factor, eg, depression or suppression of emotions, may have arisen from the biologic effects of the cancer itself or from the patient's knowledge that he or she had the disease. Personally, I do not think that such a possibility was ignored; I think the possibility never even occurred to the researchers in the first place because of their narrow disciplinary focus. Most of those researchers were psychologists or psychiatrists, although a few were somatic physicians, eg, Kissen and Thomas .
More recently, both psychologists and psychiatrists have become more aware of the need to examine the effects of cancer on the psyche, as well as the potential interference by a number of possible confounders, both demographic and biologic, with proper research methods. Nonetheless, such confounders still receive limited attention.
Stress will be defined here as the psychic and physiologic disequilibrium caused by some event, which will be called a stressor.
Stress in Animals
The earliest work on the effect of psychological factors on cancer dealt mainly with humans. But soon afterwards, a number of studies appeared in which the relationship of stress and cancer in animals was explored. Important early work on this topic was done by Riley , who carried out extensive studies on the topic, and Seifter et al . They and others who followed found clear evidence that stress in rodents led to faster growth of transplanted tumors or those caused by injection of oncogenic viruses, and shorter survival times than in nonstressed control animals. This was also true for development and growth of spontaneous tumors. It is of note that Riley's early work on spontaneous tumors was done in animals with virus-infected milk, like the mammary tumors caused by the Bittner virus.
On the other hand, some researchers, eg, Newberry and Sengbusch , found stress to inhibit tumor development and growth in animals under some conditions. In a thorough review of these and other findings, Justice  presented an extensive list of variables that stimulate and inhibit tumor development. The most important inference he drew, now well confirmed, was that viral tumors in animals are adversely affected by stress, while those induced by chemical carcinogens are favorably influenced by stress.
In view of the role attributed to the immune system in protecting against cancer growth and possibly initiation, one might be tempted to transfer these animal findings to humans. Yet several facts suggest caution.
1. Humans, guinea pigs, and certain other animals are considerably less sensitive to corticosteroid proliferation than the major rodent species used in the laboratory,  and, indeed, even among these rodent species, various strains differ in their sensitivity to glucocorticoids. Thus, the major finding of Riley , Seifter , and others that stress-induced high glucocorticoid levels in rodents led to increased tumor growth might not be duplicated in humans, even if human cortisol levels did rise with stress.
2. Humans are outbred, with a variety of responses to many physiologic stimuli; mice and rats used in laboratories historically have been inbred to produce cancer-prone strains. While there are now more varied strains, overall that history cannot be ignored.
3. Tumor transplants or heavy doses of carcinogens introduce strong antigens, that is, stimuli to immune recognition and response, with consequent greater protection against the tumor. Spontaneous human tumors take a long time to develop and probably involve relatively weak or even absent antigen proliferation, and hence limited or even absent immune recognition of and response to antigens.
4. When animals are immunosuppressed, they develop tumors in excess of normal at many sites, each strain being susceptible to tumors at the site peculiar to that strain, eg, liver, lung, testes. When humans are immunosuppressed, they also develop tumors in excess of normal, but most frequently lymphoreticular tumors; that is, the focus of the immunosuppressive stimulus is the immune system itself. The incidence of reticulum cell sarcoma in immunosuppressed individuals, for example, is 150 times that of the population, whereas the incidence of other tumors (except other lymphomas, which is also very high) is about twice that of the population.
5. Justice reported  that stress inhibited growth in chemically induced animal tumors. While many stress experiments in animals involve viral tumors, with associated stimulation of tumor growth, in humans, the proportion of viral tumors is small, on the order of 3% to 4% of all tumors.
Thus, if one extrapolated directly and incautiously to humans from animal findings, one would conclude that overall, stress is a depressant of tumor development and growth, as found by Newberry  and others in animals, rather than a stimulant to tumor growth, as many researchers suggest. For a number of reasons, such a conclusion should not be drawn. Although the animal findings are important, they should form the bases for hypotheses, not conclusions, about the effects of stress on human cancers.
This discussion will focus on cohort studies, as there are several reasons for giving little emphasis to case-control studies. First, cancer can and does produce physical, psychological, and attitudinal changes, mostly negative, that can bias conclusions . Second, these changes in patients are known to increase reports of stressful events when compared with controls .
Third, and perhaps most important, one can never be sure that the patient group sample is unbiased, and that the control group is matched to the patient group in regard to variables that might lead to erroneous conclusions. A very limited list of such variables would include the tumor's site, stage, histologic grade, depth of invasion, and size; the degree of lymphocytic invasion at the tumor site; the degree of microvascularization; the patient's age, sex, socioeconomic level, race, smoking status, prior tumor history, alcohol habits, body mass index, status of certain genes (eg, p53), compliance with treatment regimen, and, for breast cancer, age at menarche, age at menopause, oral contraceptive use, estrogen-receptor level, and menstrual stage at operation. In a large cohort study, one hopes that the sheer size of the sample will cause most of these variables to even out among the groups with and without the risk factor.
Case-Control Studies--Several early workers, using the case-control model, reported a greater number of stressful events occurring earlier in life in patients with cancer than in the noncancer groups, eg, Greene , and LeShan and Worthington . But most later case-control studies showed no excess of traumatic events among patients, eg, Schonfield  and Greer  (a short review of his studies).
Among the more recent case-control studies, we find similar contradictory results. Ramirez et al  compared frequency of traumatic events between diagnosis of breast cancer and first recurrence among 50 patients with recurrence with the frequency of such events during a similar period in 50 patients without recurrence. They found an excess of reported stressful events among those patients with recurrence.
In contrast, Priestman et al , who studied 300 women, 100 with malignancy, 100 with benign tumors, and 100 controls, found that the severity and nature of the stressors did not differ among the groups. In fact, the controls experienced more stressful events than did the benign group, and the benign group experienced more than those with cancer.
Early Cohort Studies--The earlier cohort studies found no more cancers among stressed than unstressed members of the cohort. For example, Keehn et al  found no greater cancer mortality among 9,813 soldiers of World War II discharged for psychoneurosis than among 9,942 controls over the period January, 1946, to December, 1969.
Keehn  studied cancer mortality among prisoners of war in World War II from 1946 through 1975, and the Korean conflict from 1954 through 1978. No excess cancer mortality was found for either Pacific or European World War II veterans (n = 6,023), or for Korean veterans (n = 3,959) over their respective controls (n = 5,223 and n = 3,953).
Joffres et al  looked at 4,581 Japanese men in Hawaii and found no more stressful events among cancer patients than among controls. While this is a case-control analysis, it should be noted that if such an analysis is done on all the cases and all the controls in a population, the results are no more biased than those in a cohort study, in which all the later cases and later controls are similarly identified. Stronger bias is found in smaller sample case-control studies in which both the controls and cases are selected. In the Joffres study, neither group was selected beforehand, and it could therefore be judged to be as trustworthy as a parallel cohort study.
Recent Cohort Studies--There have been relatively few recent cohort studies. An example is a study by Barraclough et al , who found no relationship between stressful events and breast cancer survival. In a widely cited series of studies, Grossarth-Maticek et al  did find a relationship between stressful events and later cancer, but this work has been criticized most severely  and will not be further dealt with here.
Readers should be aware of a demonstrated bias that may affect these studies . It appears that, on the whole, cancer patients tend to recall more stressful events than noncancer controls, even though some studies report otherwise . A few defining studies have shown that cancer patients' reports of stressful events do not reflect actual events experienced. In a thorough review of memory as it is influenced by affect (that is, emotion, mood, or feeling), Blaney  concluded that people with negative affect report more negative events than people with average or positive affect. Studies reflecting his conclusions are those of Brett et al  and Cohen et al . Almost all patients who have cancer and know it have negative affect to varying degrees. Even if not all but a substantial number had negative affect, their excessive recall of negative events would be enough to bias the average recall level of the whole cancer group.
These three sets of findings lead us to the following conclusions :
1. Case-control studies yield mixed results but are subject to biases.
2. Most cohort studies show no excess stressful events associated with cancer incidence, mortality, or survival.
3. People with negative affect report having experienced more stressful events than those with average or positive affect, when the true frequencies are alike.
The last fact explains many, if not all, of the case-control findings. It is also the basis for predicting that in the absence of such bias, there will be no excess of stressful events among the cancer group. The prediction is confirmed by the findings of the cohort studies, in which that particular bias cannot exist. Thus, it is almost certain that stressful events do not occur more often among those who later get cancer, die of it, or survive a shorter time than among controls.
Bereavement is an important category of stress that has been studied for its possible effects on later cancer incidence and mortality. Holmes and Rahe  ranked loss of spouse as the most stressful among 43 possible stressful events. Yet, several writers have observed that bereavement is not always accompanied by sadness, distress, or regret. For example, the death of a spouse with a fatal and painful disease can produce relief rather than sadness or distress. Although the number of such cases is quite probably relatively small, I know of no studies on this matter.
An overall measure of the effects of bereavement combines data from those who are distressed by the death and those who are relieved. The result is conservative, since the possible effect of distress in producing cancer will be diminished overall by the lack of reported stress among those experiencing relief. The result will be even more conservative if the bereaved who are relieved at a close person's death do, in fact, have reduced susceptibility to later cancer.
Some of the earlier case-control studies reported increased cancer incidence among widowed persons. These have been intensively analyzed, as have prospective studies up to 1986, and their problems and difficulties have been carefully delineated .
In the cohort studies, as before, the bias that may exist in case-control studies has been removed. None of the large cohort studies carried out over an extended period have shown excess cancer deaths in bereaved spouses, compared with still-married spouses. In the exceptions, the excess lasted 6 months, a year, or, in one study, as long as 2 years. Since the development time to diagnosis of most cancers is on the order of years--3, 10, 15--those findings could not have referred to cancer initiation. One such large cohort study, a 1987 Finnish study of 95,647 persons widowed in 1972 , showed no excess deaths during the subsequent 4 years among 7,600 cancer cases. One can ignore the excess mortality seen among widows in the first week following death and in the first month among widowers as not being attributable to cancer.
Another study, conducted in Washington County, Maryland, of 4,032 white persons widowed between 1963 and 1974 and followed for approximately 12 years  showed no excess of cancer deaths. A third large cohort study looking at this question  reported a similar lack of excess cancer deaths over 10 years among persons widowed in 1971. Here, the sample was 1% of the whole population of England and Wales, observed from 1971 to 1981. The authors write, "No peak of postbereavement mortality from malignant disease is clearly established in either sex ". In a fourth such study of 1,782 breast cancer patients--all those diagnosed in Denmark from March 1, 1983, to February 29, 1984--and 1,738 controls, Ewertz  found no difference in the death rates of married and widowed patients.
In summary, while some studies have reported a short-term excess of cancer following bereavement, large cohort studies have not, in general, found excess cancer incidence or death over the long term. This conclusion is consistent with the previous one that stress other than bereavement cannot be said to increase later cancer incidence or death.
A few studies have looked at cancer survival among widows, eg, Neale , and one or two have included widowers' survival. Also, one or two studies have looked at disease-free interval. However, the results have been mixed, some finding reduced survival among the widowed and some not. In either case, those results cannot be applied to the current issue-bereavement as a psychological factor-since none of the studies on survival even mention the time of bereavement in respect to the cancer diagnosis. Only one thing can be certain in these studies: the marital status of the patient at the time of cancer diagnosis. Thus, the cancer could have been detected 1 day after bereavement or 20 years after bereavement. This fact allows no conclusions to be drawn from the marital status studies with regard to bereavement as a psychological factor.