Grief is a complex process. The early effects for a person who loses a partner or spouse concern a) the immediate shock and psychological impact of bereavement and b) the change from cohabitee/married to single status [4]. Although depression or other psychological distress may diminish over the first year, distress over the loss itself may continue for a number of years [5] and include deterioration in morale and in physical health. In studying the impact of bereavement on longer term physical health it is important to distinguish any such changes from normal age related deterioration, and thus controlled comparisons are essential. The clearest evidence for increased mortality arises from a study of members of a health care plan in California in which 12,522 spouse pairs were followed up for 23 years. Mortality following bereavement was significantly higher in men and women after adjusting for age, education, and other predictors of mortality. The highest relative risks occurred 7–12 months after bereavement but remained elevated for two years after the bereavement [6]. The study, however, was unable to adjust for other important changes in health status in the “to-be bereaved” or those that might have occurred between the baseline examination and mortality, possibly weakening the relationship between risk factors other than bereavement and mortality. Other studies have reported absent or statistically weaker associations. A large cohort study of 7,735 men followed for a mean of 11.5 years in the United Kingdom showed no association between becoming widowed and subsequent cardiovascular or non-cardiovascular (usually cancer) death [7]. This was in contrast to men divorced during follow-up who were at elevated risk of both types of mortality. A further study of data on almost 92,000 people in the US identified 4,032 white adults who became widowed between 1963 and 1974. It was reported that mortality rates based on person-years at risk were not elevated for widows compared to matched controls remaining married, while the relative risk was significant for men in only those 55 years old and over [8]. Another large US study of mortality over 11 years estimated that the relative risk of mortality was greater for unmarried versus married partners from all ethnic backgrounds [9]. Unfortunately this study did not compare mortality in all bereaved participants versus those not bereaved.

The mechanisms involved in men's particular vulnerability are not clear short of the somewhat prosaic suggestions that men are less able to prepare nutritious meals and are more likely to smoke and misuse alcohol. In order to explore whether sex differences in such factors after bereavement might explain higher risk in widowers than widows, data were examined from a national register in Denmark of over 6,000 bereavements [10]. Although daily use of medication and primary care visits rose for up to five years after bereavement, no differences between men and women were found. However, a number of other studies have reported age and gender differences in vulnerability to bereavement [11]–[13].

Our finding of increased survival of a spouse or partner after a cancer death is not completely counter-intuitive. The release from the stress of caring may lead to better health and reduced psychological morbidity for partners as the acute phase of grief subsides. Furthermore, one possible explanation for our finding is that many patients dying of cancer in the UK receive greater supportive and palliative care services than those dying from other causes and this may moderate cohabitees' extreme stress [23] [24]. Where specialist rather than generalist palliative care is available, carers may receive more direct support. Many community and hospice based specialist palliative care teams conduct bereavement risk assessments in relatives and friends of the dying [25], [26]. Pre-bereavement formal and informal counselling may be available for families, and many palliative care services get back in touch with families in the first few weeks of bereavement to offer further support. This is particularly common practice in hospice settings, and the majority of deaths in hospices are still due to cancer. NICE guidelines in England and Wales for specialist palliative care recommend carers are fully involved in clinical decisions and are supported emotionally [27]. Death is also a more expected outcome following a diagnosis of cancer. Those who have been diagnosed early may have had considerable time to prepare for death. People with cancer are more likely to be placed on registers as part of the Gold Standards Framework, a structured care programme in UK general practice http://www.goldstandardsframework.org.uk/), and be offered opportunities to discuss future care planning and preferred priorities for care and place of death. Furthermore, cancer may be distinct from other causes of mortality in that life habits (except possibly smoking) might not be shared between partners as is often the case for COPD, and heart and liver disease. However, even if we are incorrect in this suggestion, one of the best longitudinal studies of the impact of bereavement found little evidence for impact of a shared environment [6]. There are also opportunities for surviving partners to assess and tackle their own health risks during the time from cancer diagnosis to death. Although there is increasing emphasis on good generalist and specialist palliative care for those with advanced progressive illness without a cancer diagnosis [28], many difficulties remain in making this possible. Our findings suggest there may be benefits for carers as well as cancer patients in pursuing the further development of such care for those in the last year of life and those close to them.

General practitioners may play an important part in reducing mortality after bereavement. In the 1980s it was reported that 76% of widowed people had at least one consultation with their family doctor in the five to seven months after the death, despite it being uncommon for bereaved patients to have any record of the bereavement in their GP case-notes [29]. Such attendance figures are likely to be even higher today [30] as it appears that most bereaved patients welcome the involvement of their family doctor [31]. The GP may have had an even greater impact on our exposed group as our study design means that each dyad was registered with the same general practice. When the same GP has been managing cancer in the lost loved one, he or she is more likely to have also helped the bereaved. This is borne out by our results on use of primary care services and is supported by other evidence that relatives of patients who receive terminal care at the same practice fare best, with most being visited by the GP over the year after the death [32]. Our finding of an increase in receipt of prescriptions for antidepressants and, in particular, hypnotics, both before the death of the cohabitee and thereafter, confirm that this is a very stressful time for people and that they suffer in particular from insomnia. Unfortunately, there is little evidence for effectiveness of hypnotics in this setting [33].

Our study has a number of strengths and limitations. This study is based on a very large cohort of people bereaved after the death of a person with cancer and not from couples volunteering for a study of bereavement. They were matched on age and sex with an even larger cohort of controls and data were available for 6 months before the death/index date. Thus the sample was not affected by selection, recall and non-response biases that have limited many studies of bereavement. Use of Poisson regression models, with our main outcomes modelled as time-dependent, allowed for unbiased estimates of the effects of bereavement while reducing bias caused by differential lengths of follow-up. General practice may be an important confounder in that is associated with the exposure and the outcomes. Thus, we used a random effects approach, which took account of the potential clustering effects of general practices. Our main limitation is selection of participant dyads. We cannot be certain that every participant selected was in a coupled relationship (partners or spouses) with each cohabitee; some may have been blood relatives or friends. However, the disparity of sex in 98% of the pairs strongly suggests that the overwhelming majority were cohabiting couples. Furthermore, this limitation can be applied equally to both cohorts. We may also have missed couples living with offspring older than 18 years. However, once again this would have affected both cohorts. The apparent psychological stress reflected in the increased prescriptions also gives grounds for confidence that we have selected couples where the death of the cohabitee had a major impact suggesting that most were emotionally close. We shall also not have captured bereaved people not registered with general practitioners. This limitation, however, is likely to be very small, given that on average 98% of people in England are registered with a general practitioner and that older people, and particularly those with a sick person in the household, are more likely to be registered. Bereaved people who leave their practice (and thus no longer contributed to follow-up data) might be a group at higher risk of death, for example they may include people who can no longer care for themselves and are admitted to residential homes, or move to join their families. We could not trace death certification in people who left the practices as the THIN data are anonymised. To a certain extent, however, this may also have been the case for older people who left the comparison cohort. Although there were some differences between the exposed and unexposed cohorts at the baseline, adjustments in the statistical model would have taken these into account. At first sight, 15,748 cancer deaths is lower than expected for a total practice population of close to nine million patients. The main reason why we have a smaller sample of deaths is because our entry criterion required that the deceased have lived with another adult. Many of those who die from cancer will be living alone as they are widowed or single and do not live with other family members. The other reason is that our cohort is dynamic, so even though there are nine million patients not all were a part of it for the full 15 years. Finally, there may be some cancer deaths that we did not capture as we could not identify the cause of death.

The main focus of the research was on the period after the death (or index date in the comparison patients). The 6 months before the death/index date was chosen as a consistent baseline against which we could measure change. In order to consider data on covariates, a period is needed in which patients are registered with GP. Although an alternative approach would have been to start each cancer dyad at the time of the cancer diagnosis, we would have had a greatly variable time period before the death (in some cases more than a decade), which would not always have been informative since there are many hidden biases (e.g. socio-economic status) for when patients are diagnosed. The same partner might not have been with the deceased throughout this time. Finally it would have made the selection of a start point for control patients inconsistent.