Statistical analysis

The person years at risk (of dying) was calculated for each participant, from the date of the original survey until the 30 September 1999 or, if the person had died, until the date of death. This allowed us to determine the mortality in each of the three pain groups (widespread pain, regional pain, no pain). Thereafter we used Cox proportional hazards modelling to take account of the possible confounding effects of age (in five year age groups), sex, and study location. The results are presented, with the “no pain” group as reference, as mortality rate ratios with 95% confidence intervals. Similar analyses were conducted according to specific causes of death.

Results

Of the 6569 participants, at baseline 1005 (15%) had widespread pain, 3176 (48%) had regional pain, and 2388 (36%) had no pain. People with widespread pain (median age 55 years; 66% women) were older and more likely to be female than those with regional pain (median age 49 years; 59% women) and no pain (median age 42 years; 54% women). In total there were 654 deaths among participants during the follow up period. Mortality was lowest in those who originally reported no pain (10.1 per 1000 person years) and increased across regional pain (13.1/1000 person years) and widespread pain (16.2/1000 person years) groups. The mortality in both the regional pain (mortality rate ratio 1.21, 95% confidence interval 1.01 to 1.44) and widespread pain groups (1.31, 1.05 to 1.65) remained significantly increased after adjustment for age group, sex, and study location (table 1).

Most of the deaths in the study cohort were due to cardiovascular disease (40%), cancer (31%), or respiratory disease (16%), with only 11% due to other diseases. In addition, 2% of deaths were due to violence, accidents, or suicide. Participants with regional pain and widespread pain were, respectively, three times and five times more likely to die from causes not related to disease during the follow up period (table 2). However, the wide confidence intervals around the mortality rate ratios are indicative of the small number of deaths from these cause. There was no relation between pain status reported on the original survey and subsequent mortality from either cardiovascular or respiratory disease (table 2). The excess risk was almost all due to deaths from cancer. After adjustment for age group and sex, participants with regional pain (1.55, 1.09 to 2.19) and widespread pain (2.07, 1.37 to 3.13) were significantly more likely to die from cancer during the follow up period compared with those with no pain.

Widespread pain may be evidence of cancer, particularly if metastasis throughout the body is present. For this reason we subsequently used the NHS central register to identify all participants in our studies who had been diagnosed as having cancer before the original study questionnaire was completed. In total, 236 participants had been diagnosed as having cancer. They were removed from subsequent analyses. The increased risk of death from cancer among participants with regional pain (1.66, 1.13 to 2.43) and widespread pain (2.27, 1.46 to 3.54) was, however, still evident (table 3).

Such an observation may arise because of confounding factors. Therefore we used data from study A to adjust for current smoking status and level of psychological distress. However, the doubling in risk of death from cancer associated with widespread pain remained (table 4).

The three most common fatal cancers in the study, which accounted for more than half the total, were lung cancer, cancers of the gastrointestinal tract (upper and lower), and cancer of the female breast. Table 5 shows a separate analysis of these cancers in relation to widespread pain. Although these site specific analyses have large confidence intervals, they are suggestive of a general rather than site specific excess cancer risk among people reporting widespread pain.

Methodology

Several methodological aspects need to be considered in the interpretation of these results. The two population based studies involved were conducted according to similar protocols, and the results with respect to the relation between widespread pain and mortality (study A 1.24 (0.91 to 1.60), study B 1.41 (1.01 to 1.97)) and specifically cancer mortality (study A 1.93 (1.14 to 3.25), study B 2.07 (1.08 to 3.94)) are consistent. There were no selection factors involving the participants, apart from their decision on whether to take part in the original pain surveys. The prevalences of regional and widespread pain in these surveys are similar to those reported by other population surveys. 2 3 5 A comparison between the causes of death in the study population and mortality among adults aged 18-84 years in England and Wales during 1998 also shows these to be similar: cancer (30.7% versus 30.5%), circulatory (39.9% versus 40.4%), respiratory (16.2% versus 13.5%), external (2.1% versus 3.5%), other causes (11.0% versus 12.1%).11 It is highly improbable that those who chose not to take part would exhibit a different relation between their original pain status and mortality over the subsequent eight year period.

Pain was reported at a single point in time (referring to the preceding month). In study B information was available on the duration of pain reported. Of those with widespread pain, 83% satisfied the definition of “chronic pain” from the International Association for the Study of Pain (IASP).12 So in most participants with widespread pain these were not transient symptoms. Nevertheless, some people will have been misclassified according to their usual pain state, and others will have changed pain state during follow up. Such misclassification, however, would result in an underestimate of the strength of the association. Errors in vital status on the NHS central register are rare. The effect of any error in coding of cause of death would be random across pain groups. It is inconceivable that responses to a postal survey up to 10 years previously could influence the chance of an error on cause of death being made on a death certificate. Such errors would again result in an underestimate of the strength of the association.

Confounding factors

We are not aware of any previous large scale population study examining pain status with future cause specific mortality, and we did not have an a priori hypothesis that excess mortality in people with widespread pain would principally be related to deaths from cancer. Could such an association be due to artefact? Participants who reported widespread pain differed in several respects from those who reported no pain. They were older and more were women, but the excess mortality was still evident after we adjusted for age and sex. The observed association may be due to confounding factors. Smoking is one of the most important risk factors for death from cancer, and it is also more common among people with widespread pain. Similarly, psychological distress has been reported as a predictor of future death from cancer13 and is also common among people with widespread pain.14 However, even after we adjusted for these additional factors we still found an approximate doubling of risk of death from cancer among people with widespread pain. Although we lacked information on other potential confounders, the observation that the increased risk of death from cancer may be consistent across cancer site makes identification of “missing” confounding variables difficult.

Social class is a marker for risk of dying from cancer. Is reported pain status also a measure of social class, thus explaining the association? Studies in different countries among populations of widely differing social status have shown remarkably similar rates of reported regional and widespread pain. 1-3 5 8 15 Overall there is little evidence that pain reporting, particularly widespread pain, varies by social class. Specifically, a previous report from study A has shown neither a strong nor a significant link between reporting of back pain—the most common regional pain syndrome—and a measure of social status derived from occupation.16 Although there were differences in social status between the areas in which the studies were conducted, we did adjust for study area in the analysis. Within each study, because the study population was sampled from people registered with selected general practices, the variation in measures of social status (particularly if based on area of residence) would be small.

Mechanisms

If the association is true, what are the possible mechanisms? The association may be with cancer occurrence or survival, and the precise nature of any association is necessarily speculative. Mechanisms associated with increased perception of pain may also be associated with an increased risk of cancer. Secondly, patients who reported widespread pain may have worse survival when they develop cancer. High levels of psychological distress, a feature of widespread pain, particularly depression, have been associated with reduced survival from cancer. 17 18 However, in the current study excess mortality from cancer was evident both in those with low and high levels of distress (data not shown). Some studies have provided evidence that certain psychosocial factors may predispose people to the development of cancer. These include the inability to release emotion, the experience of stressful life events, psychosexual disturbance, and parental problems or separation in early life.19 Many of these factors have also been linked to widespread body pain. 20 21 Lifestyle factors subsequent to these adverse events, possibly in combination with changes in neuroendocrine function, may result in both an increased reporting of pain and an increased risk of cancer.

In summary, we have shown an association between the report of widespread pain and excess mortality from cancer in the medium and long term. This has implications for the long term follow up of patients with “unexplained” widespread pain symptoms, such as those with fibromyalgia. However, it is important to set the risk in context: the vast majority of such people did not die from cancer. The risk increased from about one in 60 among people reporting no pain to one in 20 among those with widespread pain. Future studies are needed to confirm this association and to investigate the possible mechanisms.