A PROSPECTIVE, LONGITUDINAL STUDY OF THE RELATIONSHIP
OF PSYCHOLOGICAL WORK TO DURATION OF SURVIVAL
IN PATIENTS WITH METASTATIC CANCER.

A.J. Cunningham, C.V.I. Edmonds, C. Phillips, K. I. Soots, D. Hedley and G.A. Lockwood
Department of Epidemiology, Statistics and Behavioral Science,
Ontario Cancer Institute/Princess Margaret Hospital, Toronto, Ontario, Canada.

 

Correspondence can be sent to:
Dr. Alastair J. Cunningham
Department of Epidemiology, Statistics and Behavioral Science
Ontario Cancer Institute/Princess Margaret Hospital
610 University Avenue
Toronto, Ontario, Canada
M5G-2M9
Tel: 416-946-2946
Fax: 416-946-2024


SUMMARY

This study is a prospective, longitudinal investigation of the psychological factors associated with duration of survival in patients with metastatic cancers of various kinds who were receiving group psychotherapy. A correlative approach rather than an experimental (trials) design was used in the study because our aim was to relate the psychological attributes of each individual to survival rather than to test the efficacy of the intervention.

Twenty-two patients with medically incurable metastatic cancer of various kinds received weekly group psychotherapy for up to 1 year, the great majority remaining well enough to attend the group for at least 8 months. During this time, they provided extensive verbal data, through written homework, and from notes taken by the therapists at interviews and during group sessions. These data were subjected to detailed qualitative analysis, as a result of which a number of psychological themes were defined. A quantitative rating was assigned to the data for each theme in each individual patient through team discussions. The scores for the individual themes were summed to produce a 'total psychological score', representing the degree of each patient's involvement with psychological self help work. The values for each patient were then related to his or her survival duration.

Cox regression analyses showed that this composite score, and five of six major themes, were significantly related to survival duration. These themes were: ability to act and change; willingness to initiate change; application to self-help work; relationships with others; and quality of experience. In contrast, there was no relationship between survival and four standard psychometric measures taken at the onset of therapy. However, results on a 5-point scale measuring the subject's expectancy that psychological efforts would affect the disease showed a strong relationship to survival.

To control for differences in severity of disease as a factor possibly influencing psychological work, the analyses were repeated, using the survival duration predicted for each patient by a panel of oncologists as a covariate. Closely similar results were obtained.

Limitations on the interpretation of the results are discussed. Within these limits, it appears that there is a strong association between longer survival and psychological factors related to the involvement of cancer patients in psychological self-help activities. While causality cannot be inferred, reasons are given for believing that this is not a result of the disease influencing the patients' psychology, but rather the converse.


INTRODUCTION

It has long been thought that a cancer patient's mental attributes might affect the rate of progression of disease. Stolbach and Brandt (1988) reviewed the writings of physicians from the seventeenth and eighteenth centuries who reported that feelings of anxiety, disappointment and depression were possible precursors of cancer onset. In more recent times, work with animal models has clearly shown a relationship between psychological determinants and tumour growth (La Barba, 1970; Sklar and Anisman, 1981;Visintainer et al., 1983). Research on the relationship of human personality to cancer has not yielded consistent results, however. Prospective studies have implicated a number of possible risk factors, the most consistent finding being depression or repression of emotion (Shekelle et al., 1981; Hislop et al., 1987; Jensen, 1987; Temoshok; 1987; Gross, 1989; Kaasa et al., 1989), a conclusion disputed by Kreitler et al. (1993) in their more recent review. Other factors which some have found to be protective included social support (Ell et al., 1989; Waxler-Morrison et al. 1991; Maunsell et al. 1995), smaller numbers of severe or difficult life events (Ramirez et al., 1989) and 'fighting spirit' (Greer et al., 1979; Pettingale et al., 1985). Other investigators failed to find similar relationships (Funch et al., 1983; Jamison et al., 1987; Cassileth et al., 1988), and fighting spirit was not found to be protective in a recent, prospective analysis of 578 patients (Watson et al. , 1999).

More recently, evidence for an effect of psychological factors has been sought from intervention studies. This would seem to be a more promising approach. If the mind is to affect the regulation of cancer progression via endocrine or immune intermediary mechanisms, there must presumably be some change in mental status, resulting in a change in these physiological regulators; without such influences, the disease would continue to progress at the same rate. Psychological therapy would make change more likely.

Evidence from controlled trials of interventions is mixed, however. An influential analysis by Spiegel et al. (1989) demonstrated an 18-month average prolongation of life in women with metastatic breast cancer who attended a weekly support group for up to 1 year. Some anomalies in the survival curves for the control group in this experiment (unusually rapid deaths) have been pointed out by Fox (1998). A similar experiment which, unlike that of Spiegel et al. was designed to test for prolongation of survival (Cunningham et al., 1998), showed no effects of treatment and a more typical survival curve for controls. Fawzy et al., (1993) found a significant survival advantage to patients with malignant melanoma who had taken a brief, group, psychoeducational course 6 years previously; however, this effect disappeared at a 10-year analysis (Fawzy, personal communication). Edelman et.al. (1999), using an intervention similar to that of Fawzy et.al., failed to show an effect on lifespan of women with metastatic breast cancer. Richardson et al. (1990), in an experiment primarily testing for compliance with medication and using a sequential cohort design, reported significant effects of a psychological intervention in patients with haematological malignancies. Other published studies with weaker designs have given either positive (Linn et al, 1982) or negative results (Gellert et al., 1993; Ilnyckyj et al., 1994).

Thus we are at a point where, although there is some evidence for an effect of psychological interventions on average survival, it is not consistent. In addition, the available personality literature, based mainly on an analysis of mean group responses to psychometric tests, has also yielded no consensus as to which psychological attributes might promote survival. One problem may be that (in our experience) relatively few cancer patients respond to psychological help by making significant changes to their lives. Nevertheless we have noted, as have others (e.g. Kennedy et al., 1976; Achterberg et al., 1977), that patients who do become very involved with psychological self help methods often seem to survive longer than is medically expected. This potentially important phenomenon may be missed when group means are compared in randomized trials, the impact of a few being washed out in the mean results. Randomized clinical trials test the effects of an intervention against non-intervention controls. Our focus is not on testing the efficacy of a psychological intervention but on the subjects, with the aim of relating application to psychological self help with survival duration. What evidence of this nature exists?

Early work by a number of investigators (Meares, 1980; Newton, 1982-82; Simonton et al., 1980) suggested that psychological interventions may enhance survival, although their conclusions were based on evidence that was essentially anecdotal. Since that time there has been a plethora of "New Age" speculation and assertion on this topic (eg Hay, 1984; Siegel 1986), which is potentially confusing to the cancer patient population (Doan and Gray, 1992), and is often irritating to practitioners of evidence-based medicine. Among the more scientific investigations are two kinds of study. First, meta-analyses of cases of so-called 'spontaneous remission' (Everson and Cole, 1966; Challis and Stam, 1990; Stoll, 1992) or the disappearance of a malignant tumour in the absence of medical treatment that could have caused the remission. Such cases are said to be exceedingly rare and the medical literature in this field has generally not considered the possibility that this event may be triggered by psychological change; i.e. the cases are considered in purely medical terms.

Second, there is a small literature analysing the psychological properties of so-called 'remarkable survivors', cancer patients who appear to have lived much longer than expected, with or without remission of evident cancer (Kennedy et al., 1976; Achterberg et al., 1977; Roud, 1986-1987; Pennington, 1988; Huebscher, 1992a,b; Berland, 1995). A number of design weaknesses render the conclusions of these studies unreliable. The major flaw is that, in all cases, patients who apparently outlived their predicted life expectancies were selected retrospectively, meaning that there was no way of knowing how many patients showed similar characteristics but failed to survive. Inaccuracies and biases in recall are also unavoidable with such retrospective assessment. Further, these measurements were typically carried out at a single time point; patients were not observed over any length of time, as is possible when a psychological intervention is used. The methods of psychological description were mainly clinical impression or interview. Only the studies by Pennington (1988) and Huebscher (1992a,b) apparently using rigorous qualitative methods to analyse interview transcripts (grounded theory, in both cases). In all of the studies, medical documentation was almost non-existent, and inappropriate subjects (with medically curable cancers) were often included.

Our aim in the present study was to describe, in a more reliable and rigorous way, the individual variation in psychological responses to the predicament of life-threatening cancer, and to relate this variability to survival duration, using a case-oriented, correlative approach. To avoid the bias inherent in retrospective selection of a tiny proportion of "remarkable survivors", we adopted a prospective, longitudinal design, the subjects being followed over an extended period before their ultimate fate was known. This does not appear to have been attempted before, probably because the focus has been on complete remission, which is generally thought to be rare. Rather than look only for 'cures', our approach was to document smaller variations in the length of life and to relate this to variations in psychological attributes. Furthermore, we used a psychological intervention (1 year of weekly group therapy) to increase the probability of relevant change, and to allow frequent, regular data collection. Finally, because this was an exploratory, hypothesis-generating study, and it was not known what kinds of psychological work and change might prove to be related to survival, the main technique selected for characterising patients' psychological state was a qualitative analysis of verbal data collected from a variety of sources: written homework, therapist notes and interviews. To this we added a quantitative dimension by rating responses in certain categories. The present report is concerned mainly with quantitative aspects of the relationship of psychological attributes to survival; a subsequent report will deal with the quality of the patients' experience in more detail.


METHODS

Subjects:

The study population (Table 1) consisted of 22 subjects with cancers that were considered by a panel of oncologists' (below) to be medically incurable. These were metastatic in 20 cases (the remaining 2 patients having diagnoses of cholangiocarcinoma and multiple myeloma). The most common sites of the original cancers were breast, colon or rectum, and pancreas. The interval since diagnosis was variable. These patients had all shown some prior interest in helping themselves with psychological adjunctive techniques, and had completed a six-session psychoeducational program in which they had learned basic coping skills, relaxation, thought monitoring, mental imaging, goal setting, and some communication skills (Cunningham et al., 1991). Following this initial training, those who expressed a wish to do further work, and who met the medical criteria, were enrolled in a weekly therapy group for a planned 12 months. Once they had attended this group for 2 months they were considered part of the study. If, after initial enrollment, patients became too ill or were insufficiently interested to persist beyond 2 months (a further six individuals), they were not included in the study, because very little data was collected from them over such brief contact. It is of importance that there was no other selection exercised by the investigators, in contrast to the retrospective studies reviewed above. The final 22 patients were thus consecutively enrolled individuals who met the criteria of sufficiently strong motivation and good health to spend 2 months or more in therapy. The subjects (Table 1) were mainly female, all were between the ages of 31 and 67 years, and most had above-average education.

Intervention:

The intervention was 1 year of weekly 2.5-hour group psychotherapy sessions, conducted by two of the authors (AJC and CP), with between four and seven patients in the group at any one time. New members were recruited into the group at intervals, over a period of 3.5 years, to replace those leaving. Thirteen of the 22 subjects attended the therapy groups for 12 months, five attended for 8 months, and the remaining four attended sessions for between 2 and 6 months, because of declining health in two cases and lack of interest in the other cases. A monthly support group was available to graduates.

The intervention had three components. The first was support, in the sense defined by Peteet (1982): patients were comforted and strengthened by the group in the face of life-threatening disease. Expression and clarification of feelings was encouraged, as was rational problem solving, frank communication with important others, monitoring and changing of cognitions and the daily application of coping skills. Patients were helped to become, as Spiegel (1986) put it, 'experts in living'. Fear of death and dying was openly discussed and group members who died were grieved for. The second aspect of the intervention was homework: there was an emphasis on taking responsibility, not for one's cancer or its outcome, but for the implementation of a personal programme of self help. Patients were asked to practise coping skills, such as relaxation and meditation, and to complete 20 weekly sessions of written homework, which included questions and assignments over a wide range of psychological and spiritual/existential issues (examples: What makes me feel I really want to live? Which people are important in my life, and how much support do I feel each of them gives me? Draw a picture of an important aspect of your life at the following (specified) stages. Attend at least one community spiritual or religious meeting and write about how it affected you). The third facet of the intervention was group psychotherapy: patients' entrenched opinions and habit patterns were explored and modifications suggested; e.g. common issues included guilt, denial, blaming physicians and others, exaggerated dependencies, feelings of helplessness and a perceived loss of entitlement to longer life.

Collection of psychological data:

Data collection was prospective, carried out before the patient's ultimate fate was known, and longitudinal, typically over a period of 1 year (less for some, as indicated above). There were three main sources of verbal data: detailed process notes taken by the therapists during all group meetings; written homework supplied by the patient in response to the questions and assignments in the homework manual; and individual interviews (with the first author), wherever possible, approximately 1, 6 and 12 months after entry into the group. Notes were also taken at casual contact times (e.g. telephone calls). In collecting verbal data, there was a particular focus on the perceived effects of the interventions and self-help work, and on thoughts, emotions and behaviours of the subjects that might have relevance to the experience of adjusting to the cancer. However, patients spoke over a wide range of topics, including most aspects of their lives and functioning. Clinical inferences by the therapists were also an important part of the database. Records were kept of attendance and homework done.

Because of the existence of literature on the relationship between quality of life and related variables to outcome (Coates et al., 1992; Wisloff et al., 1997), we administered five self-report questionnaires at the beginning of the therapy. This was instituted a short time after the beginning of the project, and data is available for 17 of the 22 subjects. Two of the instruments used are in common use in psycho-oncology: the Profile of Mood States (POMS), a 37-item version (McNair et al., 1971/1981; Shacham, 1983), which is a basic measure of affective state in several dimensions including anxiety and depression; and the Functional Living Index for Cancer (FLIC) (Schipper et al., 1984), a 22-item measure assessing physical, emotional and social aspects of cancer patients' quality of life. To these we added the Purpose in Life test (PIL) (Crumbaugh, 1968), which measures perceived meaning in life, and the Sense of Coherence (SOC) questionnaire, sometimes known as the Orientation to Life questionnaire, or OLQ (Antonovsky, 1979), which assesses whether or not things are expected to turn out well.

The fifth instrument, an unpublished scale which we devised and called the Expectancy Questionnaire (EQ), was a simple, 5-point scale asking subjects about their expectations of the effects of the psychological therapy on the cancer. The five points on this scale were (in slightly abbreviated form): 0, no change in the cancer; 1, may be some slight slowing of cancer growth; 2, expect to make a real difference to the cancer although it may not visibly regress; 3, expect cancer will shrink noticeably; 4, the cancer will disappear.

Prediction of medical outcomes:

This study used a heterogeneous sample of patients with diseases that are usually associated with different lengths of survival. We, therefore, had to ensure that any association between psychological variables and survival duration was not simply a result of disease severity acting on both. For example, if low psychological scores were found to be associated with poor survival, this might occur because these patients were too ill to do much self help work. Controlling for the severity of disease at the start of the intervention would remove this factor. Various methods for predicting, at intake, a patient's likely survival time were considered. A literature search was undertaken for patients comparable to each of our subjects, but it proved impossible to match many of them precisely to populations studied in published trials. We could not readily enter individual prognostic factors into the analyses, such as site of metastases or interval from primary to metastatic diagnoses, because the factors that were important varied from one patient to another. We felt, therefore, that the best available estimates could be gained by submitting the medical details for each patient to an expert panel. In effect, we used a panel of oncologists , and their medical experience, to factor in the many prognostic features of a patient's history in order to produce a numerical estimate of the severity of the disease in the form of estimated expected survival. This survival estimate was then used as a covariate to control for the heterogeneity of the sample.

As a first step, the oncologist co-investigator (DH), who had not met the patients and was blind to their psychological characteristics and progress, obtained from subjects' charts all relevant prognostic data describing their disease status at the time of entry into the study. This included information on tissue of origin and grade, metastatic spread, interval between primary and secondary disease, any tumour marker activity, treatment and general health details. Written consent for this had been provided. This prognostic information was given, at individual meetings, to 14 oncologists, none of whom had met the patients. We asked them to draw a survival curve, for each patient, on a blank graph provided, predicting the rate at which individuals like the subject would be expected to die. From the curve we could then read off the median predicted survival, i.e. the point at which 50% of such subjects would be expected to have died. Thus, for each experimental subject, we had a number of predicted survival curves; this number varied from nine to 14, because the oncologists had the option to decline making a prediction if they felt insufficiently experienced with the particular diagnosis. The median of these estimates was in turn calculated. Thus, we had a median estimate of the median, or 50% survival rates, medians being chosen over means to avoid the skewing of estimates by occasional high values.

To gain some estimate of the accuracy of the technique, the same panel of oncologists (with two missing) predicted the survival curves of 18 recently deceased patients, not included in the study, who also had a variety of types of metastatic cancer. The median survival for all patients in this sample agreed closely with the actual median (Barbera L, Bezjak A, Cunningham A, Tritchler D, in preparation. Physician prediction of survival in patients with metastatic cancer). This method of predicting survival (to control for survival differences owing to varying diagnoses), although the best we have been able to devise, remains uncertain. It has the advantage of allowing the consulting oncologists to take into account individual features of each patient, such as their age and previous disease history, thus providing a more relevant estimate than a comparison with normative population means.

We were also able to use these oncologists' estimates to rank the patients, despite their different prognoses, in order of the degree to which they outlived their predicted survivals. For example, if a subject died 18 months after study entry, and the median estimate of the panel members was that 50% of similar patients would have survived only 9 months, he or she outlived expectancy by a factor of 2.0 (observed survival divided by the median survival predicted by the panel of oncologists). This ranking has been used to prepare a graphical comparison of the psychological attributes of each subject (Fig 1). For the six subjects still alive at the time of writing, this ratio is obviously an underestimate.

We maintained contact with all surviving patients after their year of therapy was over, through occasional telephone calls and through the attendance of many of them at monthly support group meetings; the date of death, where this occurred, was verified from their charts.

Qualitative analyses:

An analysis of these data was done using non-numerical unstructured data searching, indexing and theorizing (NUDIST) (Richards and Richards, 1994) software to facilitate coding, organization and retrieval of the verbal data. A grounded method was used: dimensional analysis (Schatzman, 1991; Kools et al., 1996). The first step developed a vocabulary of concepts or categories from the extensive raw data; these were coded and clustered into a provisional index tree of some 55 categories. As a model was gradually developed, this number was reduced by collapsing closely similar categories and by a decision to focus the analysis around psychological perceptions, attitudes and behaviours within individuals (excluding e.g. the subject's social circumstances, although retaining documentation of his or her reactions to it). This resulted in 26 final categories. A key process in understanding a phenomenon through dimensional analysis is the process of dimensionalizing , i.e. naming the main components and describing their various attributes. Therefore, the 26 categories were combined into the six major themes shown in Table 2 (example: 'Application to self-help work'). These themes have been connected in a simple model that forms the core of our analysis of the process of striving to overcome cancer and will be reported in detail elsewhere.

Quantitative rating of psychological features:

For each of the categories to be rated, all relevant data were assembled and summarized by one team member. The adequacy of this summary was checked by a second member. We developed quantitative ratings, on a scale of 1 - 5, for each of these categories, with written descriptions of patterns that would qualify as '1', '3' or '5'. An assessment of the strength or intensity of that category could now be made for each subject. For example, in the category 'Dedication to self help work', a rating of '1' was given when the subject was not at all committed to self-help work, and invested little or no time in it; a rating of '5' applied when the work was the top priority in the life of a subject who was sufficiently self-motivated enough to go consistently beyond assigned homework. The score for a theme was the sum of scores for its component categories. Table 2 shows the 26 categories were combined into 6 themes. These themes were then used for our calculations of the relationship of psychology to survival. We also summed the ratings for all 26 categories to give a total psychological score.

Figure 1 is a visual display of the ratings for each theme in each subject. It shows graphically the association between theme scores and survival ratio. As can be seen from the figure, different symbols have been assigned to ratings of 80 - 100%, 60 - 79%, 40 - 59% and 20 - 39% of the possible score for a theme (the lowest rating assigned for a category was 20%, meaning that the subject obtained a rating of only 1 for each of the categories contributing to the theme). To minimise personal bias in assigning ratings, a team discussion process was used. All the salient information for each subject on each category to be rated was scrutinized and rated independently by either three or four persons. Two of the raters were the therapists, who had detailed clinical knowledge of the subjects, while the other raters were blind to the subjects. All had to be satisfied that the verbal data supported the rating ultimately assigned by discussion to consensus.

Statistical analyses:

Relationships between the psychological variables were estimated using Pearson's correlation coefficients. Cox regression analysis was used to relate the six qualitatively derived psychological themes, the total psychological score, the psychometric scores at the beginning of therapy, the medical estimate of survival and age to the actual survival measured in years. Results are presented as risk ratios from the regression equations with the associated p-values. The risk ratio indicates the decrease in risk of death that results from a 1 unit increase in the measure being tested. To further illustrate the size of the effect, Kaplan-Meier plots are presented for two of the significant measures divided at their respective medians. A log rank test was calculated to compare the survival in the upper and lower halves of the population.

Two-variable Cox regression models were examined to test whether the results observed for the qualitative and psychometric measures were changed when medical estimate of survival or the age of the subject were controlled for.


RESULTS

Distribution of ratings on the qualitatively derived themes:

Figure 1 shows the ratings of each patient on each theme. The 22 patients were arranged in order of the ratio of the observed survival divided by the survival predicted by the panel of oncologists. In all but four cases this ratio was greater than 1, indicating that the observed survival was longer than that predicted by the oncologists. The median ratio of the observed time over the predicted time was 2.25. Several patterns are obvious from this visual display; in particular, there was a strong tendency for those subjects with best survival ratios to have the highest scores. All ratings of 80-100% in any theme occurred in the top half of the figure, as did all total psychological scores of 60% or more. While a high psychological score was always associated with prolonged survival the reverse was not always true; some individuals with low psychological scores lived longer than predicted. As we move down the figure, there was a trend towards lower ratings, the poorest survivors having the lowest psychological scores. It is also apparent that the different themes were highly inter-correlated. The Pearson correlation coefficients, ranged from 0.64 to 0.97, are shown in Table 3.

Qualitative Themes and Survival:

Cox regression analyses showed that all of the qualitative themes, with the exception of appraisal of threat, and including the total psychological score, were significantly related to observed survival (Table 4). These results are unlikely to be independent of each other because of the high inter-correlation of the themes (see above). The possible independence of the themes in relation to survival was not explored further here because of the limitation of the small sample size.

In order to illustrate the effect for total psychological score, the population was divided into halves at the median total score. The Kaplan-Meier curves for the two halves of the population are presented in Figure 2, along with a log rank test comparing them. Of the 11 subjects whose total psychological score was less than the median, all died within the study period, with a median survival of 1.29 years (range 0.39 to 2.92). Of the 11 subjects with a total psychological score greater than the median, only five have died, with a median survival of 2.85 years (range 1.72 - >5.62).

Because the subjects in this study have a range of cancer diagnoses that might have considerably different survival experiences, which in turn could influence psychological work, we tested the relationship of the qualitatively derived themes, controlling for the expected survival estimated by the oncologists' panel. The results are given in Table 5 and were virtually unchanged from the uncontrolled analyses. Again, all themes, except appraisal of threat, were significantly related to survival at levels very similar to the uncontrolled analyses. The medical estimate was significant at p=0.05 when tested alone (data not shown), and remained significant at similar levels when included together with a qualitative theme (Table 5).

As a final test of the qualitative themes, the Cox regressions were repeated, controlling for age of the subject. Age was not significant when tested alone and did not change the results of the qualitative themes when included in the models with them (data not shown).

Psychometric Measures and Survival:

Psychometric data were available for 17 of the 22 subjects. Of the five psychometric measures, only the Expectancy scores taken at the onset of therapy showed a significant relationship with survival (p=0.03; see Table 4). The effect of the Expectancy scores, as illustrated in Figure 3, was shown by dividing the data at the median Expectancy score and presenting the Kaplan-Meier curves. In this case, the median (Fig 3) did not split the sample into two equal groups because the Expectancy scores only ranged from 1 to 4 and there was a large number of ties. All five subjects with scores less than the median have died, with a median survival of 0.99 years (range 0.39 - 1.64). Seven of the 12 subjects with scores greater than or equal to the median have died, with a median survival of 2.29 years (range 1.20 - >4.74). Controlling for either the medical estimate of survival (Table 5) or age (data not shown) in the Cox regressions did not change the results.


DISCUSSION

The aim of this study has been to generate hypotheses about psychological factors that may influence progression of cancer in patients. A correlative method was needed. We used what Tashakkori and Teddlie (1998) called a type 4 mixed qualitative/quantitative design: a qualitative analysis of verbal data derived from each patient, followed by quantitative rating of the emergent themes. This labour-intensive, exploratory approach was chosen because the attributes related to survival are not yet known. As noted in the 'Introduction', several decades of research using psychometric tests have not yielded consistent correlations between personality or behavioural variables and survival in cancer patients. Our sample size (22 patients), while small by trials or survey standards, is large for such a detailed qualitative analysis, and comprises data collected and analysed over about 5 years.

Median survival of the 22 subjects was 2.25 times that predicted by the oncology panel (with six subjects still alive at the time of writing). This could be either a result of some selective factor, not obvious to the panel (and therefore probably psychological in nature), or be a result of attributes developed by the subjects while in therapy. Our argument for a relationship between the psychological properties of the subjects and survival duration is not, however, dependent on a comparison of the study group with the wider population, but rests on within-group comparisons. Regardless of any selection factors operating on the sample as a whole, there was a strong and highly significant relationship, by Cox regression analyses, between survival duration and a number of psychological themes obtained by the qualitative analysis: the ability to act and change; willingness to initiate changes; application to self-help work, relationships with others; and quality of experience. The one factor in which variations were not significantly related to survival was appraisal of threat, possibly because all of the subjects entering the study acknowledged the seriousness of the threat to their lives. The five significant themes, although differing greatly from one another in their clinical nature, were highly inter-correlated. It appears that, under the conditions of our experiment, all of these psychological factors fluctuate together. Alternatively, they may all be assessing one common, underlying quality. Together they form a complex of attitudes and behaviours which we have called involvement in psychological self-help work . Our data thus support the anecdotal accounts and clinical experience that this kind of involvement is related to cancer survival. They do not prove a causal relationship, as is discussed below.

A number of aspects of the design limit the conclusions that can be drawn from this study, however. The first is the possibility that those with more serious prognoses were less able to do the work required to achieve high psychological scores. To control for this we entered into the regression equations an expert panel's estimate of probable survival duration. Controlling for this covariate did not change the significance of the relationship of psychological scores to survival. Applying the panel's estimates to each individual patient allows us to compare the degree of extended survival attained by patients with different kinds of disease, as is described in the 'Methods' section.

A second argument against the possibility that the psychological ratings were influenced by seriousness of disease is based on the attendance and survival data (Table 1). Almost all of the study group were basically well for at least 6 months, which was long enough to acquire high ratings. These reflected the quality, not the quantity, of their engagement in psychological work. As Table 1 shows, all but four patients lived for at least 14 months after entry. Eighteen attended group sessions for at least 8 months, and three of the four who dropped out earlier than this survived for at least 1 year. There was only one early death in our sample, which occurred 4.5 months after entering the study (this man was well enough to make several trips between Canada and Germany during this time). We have observed clinically that motivation for action and change is, in any case, often at its highest when there are symptoms and a more immediate threat, and frequently wanes when people feel well. Results derived from the quality of life instrument (FLIC) also support the conclusion that our qualitatively derived themes were not simply covert measures of physical health at enrollment. No significant correlation was found on the FLIC, which specifically assesses physical health status among other things, between scores at entry and any of the themes or survival.

In contrast, an important limitation in our design lay in the need to delay psychological rating until the therapy was over. This was unavoidable; because it was the first experiment of its kind, we needed to collect data and analyse them qualitatively before the categories for rating could be finalised. With the categories now defined, it will be possible in future experiments to rate at an earlier stage, when health outcomes are less determined. However, because the aim of the experiment is to assess the response of patients to the intervention, there is a limit to how early this can be done if involvement is to be assessed.

This design limitation leaves the study open to the criticism that the raters may have been influenced by having some awareness of the health status of the patients after their therapy. We attempted to minimize any such bias by the use of team consensus discussions to assign ratings, with at least one member of this team being unfamiliar with the patients. Blind rating of collected data could, in theory, circumvent this source of possible error, but would require deleting all therapists' inferences from the data base. This would remove much of the most useful and subtle information about the subjects.

The generalizability of conclusions from our study is, however, severely limited by the self selection of subjects for motivation to undertake a programme of psychological work. Although the subjects were consecutively enrolled patients who fulfilled the eligibility criteria (see 'Methods' section), they represent a very small proportion of those attending the hospital. It is currently unknown what proportion of patients not currently demonstrating motivation for this kind of work would be capable of a similar dedication to psychological self help if evidence for an effect on lifespan is found. The small number of patients is also an obvious limitation to confidence in the results; while the effects were large enough to have reached significance with small numbers, a similar outcome from further work will be needed to increase confidence in the psychology-survival relationship.

Our design, while imperfect, would appear to have a number of strengths by comparison with work carried out previously. First, its prospective and longitudinal nature; second, the thorough medical documentation of all participants, in contrast to the very limited or absent medical documentation in accounts of remarkable survivors; and third, the collection of psychological documentation based on weekly contact over a period of up to a year (around 100 hours of contact per person), which contrasts with the usual single interviews in the remarkable survivor work or the psychometric tests of the personality-survival literature. The qualitative approach also allowed an initial openness to a wide range of factors that might influence good adaptation to cancer, and a gradual focussing of our data collection and analysis as patterns unfolded during the study.

As the qualitative analysis progressed, this focus became the subjects' interest and involvement in psychological work and change. Although we had prior clinical experience to suggest the importance of these qualities, we began the study with an interest in all aspects of the early lives and current life circumstances of patients, and with their personality traits generally. We also initially expected that the patients' reported behaviours might yield the most informative data. However, it became evident that the factors most critical to understanding subjects' adaptation were their ability or willingness to explore and implement new strategies and priorities; in other words, the focus shifted from the behaviour to its cognitive precursors. This parallels the shift of interest to cognitive processes in coping in the health psychology literature in general (Lazarus and Folkman (1984), Prochaska et al., 1992). Within the theme of willingness to initiate change (Table 2), categories that emerged as important included outcomes expected from applying self-help strategies, perception of ability to make changes, and what we have called wishes, plans and goals. These were all familiar within the Health Action Process Approach of Schwarzer (1992) and the prior work of Bandura (1982), a social cognitive perspective emphasizing the complex set of expectancies and multiple stages of motivation and volition that guide attitudinal and behavioural change. The pathway from initial appraisal to willingness to change and subsequent application to self-help work was strongly influenced, in turn, by what we described as ability to act and change, a theme that included character traits concerned with defensiveness and resistance (or openness) to change. Individuals could be clustered into groups, according to their adaptational patterns, in a way that will be described in more detail elsewhere. In general, however, the dominant pattern associated with good outcome, in terms of both quality of experience and long survival, was an openness or flexibility, prompting a willingness to make substantial lifestyle changes, followed by a dedicated application to psychological self help strategies. This is a pattern that also makes intuitive sense.

In marked contrast to the significant association of quantitatively derived data to survival, results from standard psychometric tests, including a quality of life instrument, obtained from 17 of the 22 subjects at the beginning of the therapy period, did not show any significant relationship with survival in the present study. This is in agreement with the relatively large literature cited in the 'Introduction'. In those published studies, where significant prediction was achieved, additional data were generally obtained by inter-personal interactions such as interviews, as was the case in the present study (Greer et al., 1979; DiClemente & Temoshok, 1985; Morris, et al., 1992; Edmonds et al., 1999). While recent evidence from chemotherapy trials suggests that quality of life tests may predict survival (Coates et al, 1992; Wisloff et al., 1997), our experimental conditions differ from these reports in that our subjects received long-term group psychotherapy, presumably modifying their psychological status between the initial assessment and time of death. In contrast to standard instruments, the expectancy score at onset of therapy from the same small number of subjects, showed a strong relationship to subsequent survival duration. It is unlikely that expectancy is a covert measure of physical health because, as was the case for the qualitatively derived psychological themes, it did not correlate significantly (r = 0.05) with the FLIC. This intriguing result, which obviously requires replication with a larger sample, echoes the recent demonstration by Butow et al. (1999) that, after controlling for conventional prognostic factors, the strongest factor predicting survival duration in patients with metastatic melanoma was response on a 4-point self-report scale measuring the effect that patients expected from their treatment.

We are aware of only one rigorous, published study that used a correlative design to relate psychological therapy to cancer progression. De Vries et al. (1997) assessed changes in the rate of growth of tumours coincident with a 12-week period of psychotherapy. A drop in this rate was found in five of 35 subjects. This approach uses an endpoint that might reasonably be thought to predict survival, and seems to suggest that psychological change can be rapidly translated into effects on tumours in some cases. We did consider using tumour markers or size as endpoints in this study, but opted instead for survival, that oncologists advised us was more definitive, because tumours sometimes fluctuate in growth rate or size for unknown reasons.

Finally, the null findings in our recently published randomized controlled trial (Cunningham et al, 1998) need to be reconciled with the present results. A hypothesis to explain this discrepancy derives from the present observation that prolonged survival is associated with considerable work and change. In the randomized trial, subjects were actively recruited, the initiative coming from the investigators. These subjects were not generally as highly motivated to do self-help work as those volunteers in the present study who showed high levels of involvement. They resembled, rather, the subjects at the low end of the 'involvement' scale in our present study. However, it is intriguing to note (Cunningham et al., 1998) that seven of the 66 subjects in the randomized trial (from the intervention group) took the trouble to seek psychotherapeutic help outside of the research protocol. These seven survived significantly longer than the rest of the group (p=0.024). While the number is too small to be more than suggestive, it is consistent with the present finding that motivation or drive to help oneself is associated with longer survival.

This comparison between the two studies also underlines the strengths and limitations of the two designs. Randomized trials test treatment efficacy more definitively. Correlative designs allow us to relate characteristics of individuals to outcomes, thus providing more information on what sort of changes might help patients. They have the advantage that valuable contributions can be made with relatively small numbers of patients, and they also avoid the inhumane process of randomizing patients who may desperately want the therapy on offer. If the relationship between therapy-induced psychological change and survival does exist, its validity will gradually become apparent with more studies of this kind being undertaken (as has happened for the smoking-lung cancer relationship).

In the next series of experiments, with the themes already defined, improvements will be possible, notably the earlier rating of patients during the intervention, We also plan to rate subjects at a number of time points, in order to relate change in psychology to outcome. This would strengthen the argument that therapy contributes something over and above the patients' initial attributes. Future work might be carried out with different patient populations, e.g. using a medically more homogeneous study group should decrease the influence of medical variables. The disease process in patients with primary cancer might prove more sensitive to psychological factors because primary tumours are likely to be less well adapted to their host. If psychological attitude eventually proves to correlate reliably with survival then it is probable, a priori, that change in attitude will promote change in survival duration. Prolongation of life through adjunctive psychological therapy may thus be more possible than has been generally believed in biomedical circles.



ACKNOWLEDGEMENTS

This study was supported financially by the J.S. Trust and the Princess Margaret Hospital Foundation.

We would also like to thank the following oncologists for providing survival estimates for our subjects: Dr. Rachael Barstow, Dr. Rob Bristow, Dr. Andrea Bezjak, Dr. David Bloomfield, Dr. Charles Catton, Dr. Pamela Catton, Dr. Bernard Cummings, Dr. Anthony Fyles, Dr. Mary Gospodarowicz, Dr. Peter Kirkbride, Dr. Normand Laperriere, Dr. Fei-Fei Liu, Dr. Warren Mason, Dr. Sue-Anne McLaughlin, Dr. Malcolm Moore, Dr. Amit Oza, Dr. Lillian Sui, Dr. Jeremy Sturgeon, and Dr. Padraig Warde, all of the Princess Margaret Hospital.


REFERENCES

Achterberg J., Mathews-Simonton, S., and Simonton, O.C. (1977). Psychology of the exceptional cancer patient: a description of patients who outlive predicted life expectancies. Psychother. Theory Res. Pract., 14: 416-422.

Antonovsky, A. (1979). Health, Stress and Coping. San Francisco: Jossey-Bass.

Bandura, A. (1982). Self efficacy mechanism in human agency. American Psychologist, 37: 122-147.

Berland, W. (1995). Unexpected cancer recovery: why patients believe they survive. Advances J. of Mind-Body Health, 11: 5-19.

Butow, P.N., Coates, A.S. and Dunn, S.M. (1999). Psychosocial predictors of survival in metastatic melanoma. J Clin Oncol, 17(7): 2256-2263.

Cassileth, B.R., Walsh, W.P., and Lusk, E.J. (1988). Psychosocial correlates of cancer survival: a subsequent report 3 to 8 years after cancer diagnosis. J Clin Oncol, 6: 1753-1759.

Challis, G.B., and Stam, H.J. (1990). The spontaneous regression of cancer. Acta Oncol., 29: 545-550.

Coates, A., Gebski, V., Signorini, D., Murray, P., McNeil, D., Byrne, M., and Forbes, J.F. (1992). Prognostic value of quality of life scores during chemotherapy for advanced breast cancer. J Clin Oncol, 10(12): 1833-1838.

Crumbaugh, J.C. (1968). Cross-validation of Purpose-in-Life Test based on Frankl's concepts. Journal of Individual Psychology, 24: 74-81.

Cunningham, A.J., Edmonds, C.V.I., Hampson, A.W., Hanson, H., Hovanec, M., Jenkins, G., and Tocco, E.K. (1991). Helping cancer patients cope with, and combat, their disease: report on a group psychoeducational program. Advances J. of Mind-Body Health, 7: 41-56.

Cunningham, A.J., Edmonds, C.V.I., Jenkins, G.P., Pollack, H., Lockwood, G.A., Tritchler, D., and Warr, D. (1998). A randomized controlled trial of the effects of group psychological therapy on survival in women with metastatic breast cancer. Psycho-Oncology, 7: 508-517.

De Vries, M. J., Schilder, J.N. Mulder, C.L., Vrancken, A.M., Remie, M.E., and Garssen, B. (1997). Phase II study of psychotherapeutic intervention in advanced cancer. Psycho-Oncology, 6: 129-137.

DiClemente, R.J., and Temoshok, L. (1985). Psychological adjustment to having cutaneous malignant melanoma as a predictor of follow-up clinical status. Psychosomatic Medicine, 47: 81.

Doan, B.D., and Gray, R.E. (1992). The heroic cancer patient: Reflections on the relationship between illusion and mental health. Canadian Journal of Behavioral Science, 24(2): 253-266.

Edelman, S., Lemon, J., Bell, D.R. & Kidman, A.D. (1999). Effects of group CBT on the survival time of patients with metastatic breast cancer. Psycho-Oncology 8: 474-481.

Edmonds, C.V.I., Cunningham, A.J., and Lockwood, G.A. (1999). Psychological response to long term group therapy: A randomized trial with metastatic breast cancer patients. Psycho-Oncology, 8: 74-91.

Ell, K., Nishimoto, R., Morvay, T., Mantell, J., and Hamovitch, M. (1989). A longitudinal analysis of psychological adaptation among survivors of cancer. Cancer, 63: 406-413.

Everson, T.C., and Cole, W.H. (1966). Spontaneous Regression of Cancer, Saunders: Philadelphia.

Fawzy, F.I., Fawzy, N.W., Hyun, C.S., Elashoff, R., Guthrie, D., Fahey, J.L., and Morton, D.L. (1993). Malignant melanoma. Effects of an early structured psychiatric intervention, coping and affective state on recurrence and survival 6 years later. Arch. Gen. Psychiatry, 50: 681-689.

Fox, B.H. (1998). A hypothesis about Spiegel et. al.'s 1989 paper on psychosocial intervention and breast cancer survival. Psycho-Oncology, 7: 361-370.

Funch, D.P., and Marshall, J. (1983). The role of stress, social support and survival from breast cancer. J. Psychosom. Res., 27: 77-83.

Gellert, G.A., Maxwell, R.M. and Siegel, B.S. (1993). Survival of breast cancer patients receiving adjunctive psychosocial support therapy: A 10-year follow-up study. J of Clinical Oncology,11(1): 66-69.

Greer, S., Morris, T., and Pettingale, K.W. (1979). Psychological response to breast cancer: effect on outcome. Lancet, ii: 785-787.

Gross, J. (1989). Emotional expression in cancer onset and progression. Social Science Medicine, 28: 1239-1248.

Hay, L.L. (1984). You Can Heal Your Life. Santa Monica: Hay House.

Hislop, T.G., Waxler, N.E., Coldman, A.J., Elwood, J.S., and Kan, L. (1987). The prognostic significance of psychosocial factors in women with breast cancer. J. Chron. Dis., 40: 729-735.

Huebscher, R.R. (1992a). Spontaneous Remission of Cancer. Ph.D. Thesis. U.M.I. Dissertation Services, Ann Arbor, Michigan.

Huebscher, R. R. (1992b). Spontaneous remission of cancer: an example of health promotion. Nurse Practitioner Forum, 3: 228-235.

Ilnyckyj, A., Farber, J., Cheang, M.C., and Weinerman, B.H. (1994). A randomized controlled trial of psychotherapeutic intervention in cancer patients. Annals Royal College of Physicians & Surgeons of Canada, 27: 93-96.

Jamison, R.N., Burish, T.G., and Wallston, K.A. (1987). Psychogenic factors in predicting survival of breast cancer patients. J. Clin. Oncol., 5: 768-772.

Jensen, M.R. (1987). Psychobiological factors predicting the course of breast cancer. J. Personality, 55: 317-342.

Kaasa, S., Mastekaasa, A., and Lund, E. (1989). Prognostic factors for patients with inoperable non-small-cell lung cancer, limited disease. Radiother. Oncol., 15: 235-42.

Kennedy, B.J., Tellegen, A., Kennedy, S., and Havernick, N. (1976). Psychological response of patients cured of advanced cancer. Cancer, 38: 2184-2191.

Kools, S., McCarthy, M., Durham, R., and Robrecht, L. (1996). Dimensional analysis: broadening the conception of grounded theory. Qual. Health Res., 6: 312-330.

Kreitler, S., Chaitchik, S., and Kreitler, H. (1993). Repression: Cause or result of cancer? Psycho-Oncology, 2: 43-54.

LaBarba, R.C. (1970). Experiential and environmental factors in cancer. A review of research with animals. Psychosom. Med., 32: 259-276.

Lazarus, R.S., and Folkman, S. (1984). Stress, appraisal and coping. Springer: New York.

Linn MW, Linn BS, Harris R. 1982. Effects of counseling for late stage cancer patients. Cancer 49: 1048-1055.

Maunsell, E., Brisson, J., and Duschenes, L. (1995). Social support and survival among women with breast cancer. Cancer, 76(4): 631-637.

McNair, D.M., Lorr, M., and Droppleman, L.F. (1971/1981). Profile of Mood States Manual. Educational and Industrial Testing Service, San Diego, California.

Meares, A. (1980). What can the cancer patient expect from intensive meditation? Australian Family Physician, 9, 322-325.

Morris, T., Pettingale, K., and Haybittle, J. (1992). Psychological response to cancer diagnosis and disease outcome in patients with breast cancer and lymphoma. Psycho-Oncology, 1: 105-14.

Newton, B.W. (1982-83). The use of hypnosis in the treatment of cancer patients. Am. J. Clin. Hypnosis, 25: 104-113.

Pennington S. (1988). Healing Yourself. McGraw: Toronto.

Peteet, J.R. (1982). A closer look at the concept of support: some applications to the case of patients with cancer. Gen. Hosp. Psych., 4: 19-23.

Pettingale, K.W., Morris, T., Greer, S., and Haybittle, J.L. (1985). Mental attitudes to cancer: an additional prognostic factor. Lancet, i: 750.

Prochaska, J.O., DiClemente, C.C., and Norcross, J.C. (1992). In search of how people change: Applications to addictive behaviors. American Psychologist, 47(9): 1102-1114.

Ramirez, A.J., Craig, T.K.J., Walson, J.P., Fentiman, I.S., North, W.R.S., and Rubens, R.D. (1989). Stress and relapse of breast cancer. Brit. Med. J., 298: 291-293.

Richards, T.J., and Richards, L. (1994). Using computers in qualitative research. In Handbook of Qualitative Research, Denzin NK, Lincoln YS (eds). Sage Publications: Thousand Oaks, California; 445-462.

Richardson, J.L., Shelton, D.R., Krailo, M., and Levine, A.M. (1990). The effects of compliance with treatment on survival among patients with hematologic malignancies. J. of Clinical Oncology, 8(2): 356-364.

Roud, P.C. (1986-87). Psychosocial variables associated with the exceptional survival of patients with advanced malignant disease. Int. J. Psychiatr. Med., 16: 113-122.

Schatzman, L. (1991). Dimensional analysis: notes on an alternative approach to the grounding of theory in qualitative research. In Social Organisation and Social Process, Maines DR (ed.). Aldinc De Gruyter: NewYork, pp. 303-314.

Schipper, H., Clinch, J., McMurray, A., and Levitt, M. (1984). Measuring the quality of life of cancer patients: The Functional Living Index-Cancer: Development and Validation. Journal of Clinical Oncology, 2: 472-483.

Schwarzer, R. (1992). Self-efficacy in the adoption and maintenance of health behaviours: theoretical approaches and a new model. In Self Efficacy, Schwarzer R (ed.). Hemisphere Washington Pub. Co.

Shacham, S. (1983). A shortened version of the Profile of Mood States. Journal of Personality Assessment, 47(3): 305-6.

Shekelle, R.B., Raynor, W.J., Ostfeld, A.M., Garron, D.C., Bieliauskas, L.A., Liu, S.C., Maliza, C., and Ogelsby, P. (1981). Psychological depression and 17-year risk of death from cancer. Psychosom. Med., 43: 117-125.

Siegel, B.S. (1986). Love, Medicine and Miracles. Harper & Row, New York.

Simonton, O.C., Mathews-Simonton, S., and Sparks, T.F. (1980). Psychological intervention in the treatment of cancer. Psychosomatics, 21: 226-233.

Sklar, L.S., and Anisman, H. (1981). Stress and cancer. Psychol. Bull., 89: 369-406.

Spiegel, D. (1986). Psychosocial interventions with cancer patients. Psychosocial Oncology, 4: 83-95.

Spiegel, D., Bloom, J.R., Kraemer, H.C., and Gottleib, E. (1989). Effect of psychosocial treatment on survival of patients with metastatic breast cancer. Lancet, 2(8668): 888-891.

Stolbach, L.L., and Brandt, V.C. (1988). Psychological factors in the development and progression of breast cancer. In Stress and Breast Cancer, Cooper CL (ed.). Wiley: London.

Stoll, B.A. (1992). Spontaneous regression of cancer: new insights. Biotherapy, 4: 23-30.

Tashakkori, A., and Teddlie, C. (1998). Mixed methodology. Combining qualitative and quantitative approaches. Sage Publications, Thousand Oaks: California, pp. 162.

Temoshok, L. (1987). Personality, coping style, emotion and cancer: towards an integrative model. Cancer Surveys, 6: 545-567.

Visintainer, M.A., Seligman, M.E.P., and Volpicelli, J. (1983). Helplessness, chronic stress and tumor development. Psychosom. Med., 45: 75-76.

Watson, M., Haviland, J.S., Greer, S., Davidson, J., and Bliss, J.M. (1999). Influence of psychological response on survival in breast cancer: a population-based cohort study. Lancet, 354: 1331-36.

Waxler-Morrison, N., Hislop, T.G., Meares, B., and Kan, L. (1991). Effects of social relationships on survival for women with breast cancer: A prospective trial. Social Science Medicine, 33(2): 177-183.

Wisloff, F., and Hjorth, M. (1997). Health-related quality of life assessed before and during chemotherapy predicts for survival in multiple myleloma. Br. J. of Haematology, 97: 29.