Potential reasons for excess mortality among unmarried persons with cancer in general

It is possible that married individuals, because they are taken care of by their spouse, are more prone than the unmarried to visit a physician at occurrence of symptoms, thus possibly discovering tumors at an earlier stage [8–10]. Early detection may increase the chance of a successful treatment. It may, however, also be positively associated with measurements of survival simply by increasing the time between diagnosis and death (the so-called lead-time bias). As we control for such differentials, with some limitations discussed below, the remaining discussion is centered on other causal pathways.

One probable mechanism for the excess mortality among the unmarried is that they might have poorer overall physical health at time of diagnosis. In support of such a relationship, several studies have reported lower scores of self-rated physical health among the unmarried than the married [21, 22]. An important reason for this pattern is probably that social support or pressure from the spouse and economic advantages achieved by sharing a household and having a spouse who contributes lead to a healthier lifestyle; with for example better nutrition and less smoking and alcohol abuse [1, 23–25].

Also the mental health at time of diagnosis may affect cancer survival. Studies have shown that mental health problems are more common in the unmarried population, presumably in part because of lack of social and emotional support [1, 24, 26]. Common problems are e.g. depression, anxiety-disorders, and loneliness [26–28]. These conditions, perhaps the latter two in particular, may result in psychological stress [29]. This could in turn lead to more risky health behaviors and poor sleep, thus adversely affecting also the general physical health status [30]. Additionally, stress has been shown to have a more direct effect on physical health [4, 31], and some studies even suggest effects on tumor growth [32], though there are also studies where such effects have not appeared [33].

In addition to physical and mental health, treatment is of course an important determinant of cancer survival. These factors are actually linked, because psychological stress and depression may cause poorer adherence to treatment regimens [30]. It is possible, even in a supposedly egalitarian country such as Norway, that married individuals receive better treatment from hospitals than the unmarried. Adherence to treatment regimens, however, is perhaps likely to play a more important role. A meta-analysis suggests that marriage influences adherence to treatment positively, partly through the partner's support [7]. Besides, one might expect that married individuals have a better chance of avoiding unhealthy behaviors after a diagnosis has been made, thereby improving prognosis [34].

In addition to affecting the survival prospects through factors such as spousal support, marriage may have an effect through parenthood. Raising children appears to have a positive effect on cancer survival [35], probably because children induce a healthier lifestyle and (especially if they are adults) may provide support during treatment and later. Unfortunately, our data only included information about children for the youngest individuals.

Finally, selection obviously contributes to the difference in cancer survival between married and unmarried individuals. For example, men with much knowledge and high income (potential) are seen as desirable partners and therefore tend to display high marriage rates (though not while studying) and low divorce rates, while the corresponding effects of women's socio-economic resources are more ambiguous and probably (as we return to below) have changed over time [36, 37]. Education and income are also important determinants of health [38], and may through such differentials in health, or in treatment, also affect the cancer survival [39]. We have controlled only partially for this confounding effect of socio-economic resources by including education. Also, values may play a part. Individuals who are engaged in religious activities, for instance, appear more prone to avoid risky health behaviors [1]. In addition, they are less likely to divorce their spouses [40]. The values also include lifestyle preferences, with implication for entry into and out of marriage as well as health behavior. Next, healthy individuals are probably more likely to enter and remain in a marriage than the less healthy [41], although there are also studies indicating a negative health selection into marriage [23]. Furthermore, the health of the spouse is obviously a determinant of widowhood, and is linked to the health of the person under study. Finally, childbearing is not only a result of marriage; it is also a determinant. For example, married individuals with children, non-adult in particular, are less likely to divorce than those without [40]. As mentioned, children may affect cancer survival as well.

Potential causes for increased excess mortality among unmarried cancer patients

When discussing trends in excess mortality, we first consider the never-married, for whom the changes have been most pronounced. In principle, changes in any of the mechanisms described above could help explain the observed increase in excess mortality in this group.

Starting with the health factors, it is possible that the never-married have had an increasingly poor health at the time of diagnosis compared to the married. In support of that idea, it has been shown by some researchers that the never-married have experienced a less favorable development in all-cause mortality over the last few decades [e.g. [11, 13–16]] and in mortality from cardiovascular diseases [16, 42], the latter in particular being indicative of growing differences with respect to health-related lifestyle. The very few studies that have investigated the changing differentials in self-rated health have provided mixed evidence. An American study found an especially pronounced health improvement in the never-married population compared to the married [43], while a Finnish study suggested the opposite [21].

The reasons for the relative deterioration in general health among the never-married are far from obvious. We can only offer some suggestions for why it may have become more important to have a spouse who provides support or exerts some pressure. One possibility is that the social cohesion in the society may have decreased over time [44]. A growing importance of self-realization in the population may have reduced the willingness to care for others aside family and friends [45], and increased workloads and work-related demands may have had similar effects. In a setting of reduced social cohesion, it is not unlikely that the never-married individuals would be particularly vulnerable. Especially the older never-married population might be at risk, considering that elderly individuals could have more difficulties in maintaining social connections outside the family than the younger.

The increasing excess mortality among the never-married cancer patients may in principle also be linked to the substantial improvements in diagnostic techniques. As mentioned, married individuals tend to be diagnosed with cancer at an early stage [8]. They are more likely to visit a physician at early symptoms of disease, and more eager to undergo examination even without feeling symptoms [e.g. [46, 47]]. The latter has become an increasingly relevant issue because of the technological development, and the consequence may be that, among patients recorded with a localized tumor, the married have the smallest ones - those that to a lesser extent have infiltrated surrounding tissue. Although stage is adjusted for in this study, this control is not complete, as it does not account for sub-stages. It is, however, not likely that the possibly earlier detection of cancer among the married can contribute much in explaining the increasing excess mortality among the never-married. The estimates were very similar when we did not include tumor stage in the models (not shown), which suggests that additional control for sub-stages would also matter little.

To the extent that there are marital status differentials in treatment, it is not impossible that these have increased. One reason is that support from others may be important for a patient's compliance with the treatment recommendations, and that those without a spouse may find it increasingly difficult to find alternative sources of such support, as mentioned above. Considering that treatment regimens are more complex today than earlier, and that more care is performed in the outpatient setting, support in adhering to treatment is perhaps of particular importance nowadays. Furthermore, it seems to be a common perception among health personnel that their workload is increasing. If that is the case, it is not impossible that physicians perhaps are more likely to yield to pressure from next of kin, possibly giving married individuals an advantage in receiving better treatment.

Finally, there may have been a change regarding the selection factors. In particular, a number of studies have suggested that a high wage potential now increases a woman's chance of being married, while the opposite was the case a few decades ago, when specialization within the household (with the man having paid work and the woman taking responsibility for the housework) probably to a larger extent was considered a key advantage of marriage [37, 48]. This change may not be adequately captured by the included education variable.

Regarding the widowed and divorced populations, trends in excess mortality are less clear than among the never-married. The arguments about partnership perhaps being more important for the general health because of weakened social cohesion in society - or even having a larger effect through treatment - should be relevant also for these groups. However, it is much more common among the divorced and widowed than among the never-married to have children, which may compensate for a lack of spouse. Furthermore, if there really has been a gradually more positive selection into marriage, leaving a less resourceful group of never-married, one would expect a similar increasing disadvantage among the divorced, because many of the factors that stimulate entry into marriage also tend to increase the chance of remaining in marriage [40]. The selection with respect to widowhood is very different. In marriages in general, a partner's health affects one's own health. A widowed person might thus have poorer cancer survival because of having shared an unhealthy environment with a spouse who deceased. It does not seem likely, however, that there has been much change in this mechanism during the time period in question.

Methodological considerations

The major limitation in this study is that it has not been possible to distinguish between single and cohabitants within the unmarried population. Among elderly people, the proportion which cohabits is still low. For example, only 3% cohabited at age 70-79 in 2010, while one-third were unmarried [49]. The increasing prevalence of cohabitation among the younger in our study population, however, has implications for the interpretation of the estimates. For example, 13% were cohabitants at age 50-54 in 2010 - after a doubling over the preceding 15 years - in comparison with 40% unmarried [49, 50]. (The corresponding figures at age 30-34 were 30% and 62%, but this is less relevant for our analysis because of the few cancer cases in that age group.) If cohabitants enjoy many of the same benefits as the married, which is not unlikely [21], the increase in cohabitation could contribute in explaining the less pronounced increase in the excess mortality of the never-married among the youngest in our analysis.

Another potential limitation is that we have restricted the analyses to 13 common cancer forms. If we included all other localizations into a 14^th category (i.e. not taking into account that some of these other cancer types are more aggressive than others and that the most aggressive types may occur more frequently in some marital status groups than others), very similar results were seen. This change in the analysis increased the sample size by one-third.

There were no obvious reasons to expect that the change over time in the relationship between marital status and cancer survival would differ across the 13 cancer types. The suggested mechanisms should be generally relevant, though perhaps with some differences in their relative importance, and earlier studies have not shown clear and interpretable differences across sites in the overall effect of marital status on cancer survival [51]. Therefore, we did not estimate models separately for each of the sites considered.

The corrected-survival approach that we have used may in principle not always give a good impression of how aggressive the disease is, because it is often difficult to identify a primary, underlying cause of death. An alternative would be the relative-survival approach, which is a comparison of all-cause mortality in cancer patients with that in individuals of the same age and sex in the "normal population" or even (as done in a few studies) those with similar marital status, education or other socio-demographic characteristics. This is, however, a more cumbersome procedure, and it has been shown that the results are almost identical with respect to marital status differentials [52]. Another alternative could be the observed-survival approach, where the focus is on all-cause mortality among cancer patients. We performed also these analyses, and very similar estimates resulted.

The patients' educational level is controlled for in the models, and this has some impact on the estimates. For the observations after 1980, it is the educational level during the preceding year that is included, while for earlier observations only the level in 1970 (which is up to ten years earlier) is available. In our study population only the few individuals in their low 30 s in this time period are likely to have experienced any changes in their educational level over the previous ten years, and the lack of continuous education data should thus not influence our results markedly.

This study has, however, several obvious strengths. The time-span covered is rather large, and the data include the entire Norwegian population. It is also important that we can control for tumor stage at diagnosis.