Our overall findings show that widowers had elevated risks of being diagnosed with prostate cancer, particularly with tumours that had spread at the time of detection. To our knowledge, our study is the first to have investigated specifically the association between marital status and prostate cancer grade at diagnosis. Single men were found to be at increased risk of high-grade prostate cancer. By contrast, there was no evidence that separated/divorced men had higher risk of prostate cancer than men who were married/with a partner. Results were consistent across pooled, meta- and sensitivity analyses.
Associations between marital status and prostate cancer incidence have been reported previously. However, in the vast majority of reports, marital status was not the main focus of the analyses [22-24, 26, 29, 32] and was presented as an ancillary result, without an in-depth investigation. Most studies observed that men who were not married had lower risks than those who were married [22, 25, 26, 29, 30, 32]. In particular, a nationwide population-based case-control study conducted in Sweden observed a 31% increased risk of developing prostate cancer among married men, compared to those who were never married (OR = 1.31; 95% CI = 1.29-1.33)[32]. This elevation in risk was largest for low-risk prostate cancer (based on stage, grade and prostate-specific antigen (PSA) level), leading to the interpretation that these findings might reflect a higher uptake of PSA testing among married men. Conversely, a few studies reported no association with marital status [23, 27, 28, 31]. Only one study, conducted in Alberta, Canada, suggested an increased risk of overall prostate cancer among never married men (OR = 1.93; 95% CI = 1.08-3.44), while no association emerged for separated, divorced or widowed men [24].
A recent systematic review investigated the association between marital status and cancer stage at diagnosis across several cancer sites [41]. Of the three studies reporting on prostate cancer, two found a decreased risk of presenting with metastatic disease or advanced stage among married men compared to unmarried men, and one reported an increase of locally advanced prostate cancer among separated, divorced, or widowed men. Findings for widowers concur with ours.
Marriage or partnership can affect prostate cancer risk through different pathways. It is one of the most important sources of emotional support and social interaction. Being in a committed relationship has been found to be related with a healthier diet, less smoking and alcohol consumption [9, 13], which may relate to prostate cancer risk [2]. Moreover, married men are more likely to adopt and maintain healthy behaviors as marriage and its symbolic meaning is accompanied by a sense of responsibility and norms that increase their will to stay healthy in order to take care of their family members [42]. Social support provided by marriage can also influence health outcomes through stress-buffering mechanisms. It increases the ability to cope with stress by providing the needed material and psychological resources and therefore reduces the negative effects of stressful events, such as the adoption of unhealthy lifestyle behaviours and the activation of physiological responses detrimental to health [43].
In our study, widowed men had a higher risk of prostate cancer than single or separated/divorced men. This may reflect the emotional impact of widowhood and its related health effect. The “widowhood effect” or the increased risk of mortality, including from cancer, among widowed persons compared with those who remain married is well documented [44]. Explanations include what is called the selection effect into widowhood, which makes widowers more likely to die or to develop disease because of shared household characteristics and behaviours with the lost one, the direct effect of the psychological shock, and lifestyle modifications accompanied by widowhood [45, 44].
We found that widowed men were more likely than those who were married/with a partner to be diagnosed at a more advanced stage. This is in line with the notion that marriage promotes health seeking behaviors which increase the likelihood to be diagnosed at an earlier stage of the disease, because women are more likely to seek regular care, such as screenings, for themselves and their husband [14, 46]. This is supported by a recent study, which suggests that married/with a partner men are more likely than never married, divorced, separated, or widowed men to undergo screening and prostate biopsy [47]. Without the encouragement of a spouse to seek medical attention, cancer in widowed men may remain undetected and diagnosis can be delayed, leading to a more advanced disease and poorer prognosis.
For single men, we observed different odds ratios according to cancer grade. The latter appears to differentiate early in the carcinogenesis process, with no evidence of direct progression from low- to high-grade prostate cancers [48]. This suggests that low- and high-grade prostate cancers may have different etiologies and sets of risk factors. This is supported by findings where risk factors varied according to cancer grade [49, 50]. It may thus be that the underlying etiological factors behind marital status are associated with different prostate cancer grades.
We are not aware of previous studies investigating associations between marital status and prostate cancer incidence across ethnic groups. Albeit based on limited numbers, our findings are suggestive of elevated risks of distant stage and high-grade cancers among widowed and single men of European descent. These risks were particularly pronounced for men of Asian descent, but not among men of African ancestry. Cultural differences in reaching out for regular medical follow-ups, in the absence of a partner providing social support, might explain these observations.
This study presents some limitations. First, a sizable proportion of cases (52%) and controls (32%) in the 12 studies included had missing information on marital status (Supplementary Tables 5 and 6). In order to evaluate whether their exclusion from the analyses could have resulted in selection bias, we compared study participants and those excluded for lack of information. Subjects without marital status information tended to have a slightly higher BMI, to smoke and to have more often a family history of prostate cancer, but they were less likely to have received no educationowever, within the sample of subjects with complete information, none of these variables was associated with marital status. Furthermore, restricting analyzes to studies with less than 25% of missing data for marital status reduced heterogeneity across studies and reinforced findings for widowers (distant cancers) and single men (high-grade cancers). These observations provide reassurance against selection issues operating in the study and explaining our findings. An additional indicator of the comparability of our study population to others comes from our replication of higher risks, in a similar order of magnitude, of prostate cancer among men of African descent and those with a first-degree family history of the disease.
Secondly, even if we adjusted for the specific study contributing to the pooled analyses, we cannot rule out that the observed associations could have been explained, at least in part, by the heterogeneity among studies. This could reflect variation in terms of study population, healthcare access and screening practices, recruitment of participants and assessment of exposures. However, our meta-results, taking into account study heterogeneity using a random effect model, generated results that were consistent with those from the pooled analyses. Furthermore, heterogeneity was lower in our sub-analysis of studies with fewer missing values for marital status and the overall interpretation of findings was unaltered.
Another limitation relates to errors inherent in the studies included in the analyses, such as a possible misclassification of reports on marital status, although we would expect reporting error to be non-differential between cases and controls, attenuating the associations observed. Moreover, the duration of the marital status captured at diagnosis/interview was not known.
Information on screening practices was not available, hampering our ability to evaluate the possible underlying role of screening in our findings. Nevertheless, a stratified analysis according to study countries, which may serve as a crude proxy for screening practices, suggested no difference in the results (data not shown).
Finally, the data at hand had too sparse information to conduct formal mediation analyses to identify the causes behind the associations observed.
Our study has several strengths. The PRACTICAL consortium is a unique resource for identifying risk factors that can be related to prostate cancer risk by combining a large amount of data from many studies. Inclusion of 14,760 prostate cancer cases and 12,019 population controls provided high statistical power and the ability to effectively investigate associations with cancer stage and grade. In addition, we made a comprehensive assessment of potential confounders using a DAG, although none emerged here as particularly important. Through the E‐value, we estimated the strength of association that an unmeasured factor would require to explain away the association that we observed [38]. The results suggest that relatively weak confounder associations could explain some of the results. Thus, residual confounding by unmeasured confounders is possible, although few risk factors for prostate cancer have yet been clearly identified. Our ability to conduct several sensitivity analyses confirmed the robustness of our findings.