Article Text

Download PDFPDF

The relationship between informal caregiving and mortality: an analysis using the ONS Longitudinal Study of England and Wales
  1. Susan Ramsay1,
  2. Emily Grundy2,
  3. Dermot O'Reilly3
  1. 1Department for Epidemiology and Public Health, University College London, London, UK
  2. 2Department of Geography, University of Cambridge, Cambridge, UK
  3. 3Institute of Clinical Sciences, Queen's University Belfast, Belfast, UK
  1. Correspondence to Dr Susan Ramsay, Department for Epidemiology and Public Health, University College London, 19 Torrington Place, London WC1E 7HB, UK, susanramsay{at}


Background Many studies have suggested that caregiving has a detrimental impact on health. However, these conclusions are challenged by research which finds evidence of a comparative survivorship advantage, as well as work which controls for group differences in the demand for care.

Methods We use a large record linkage study of England and Wales to investigate the mortality risks of carers identified in the 2001 Census. The analysis focuses on individuals aged 35–74 living with others in private households and a distinction is made between those providing 1–19 and 20 or more hours of care per week. Logit models identify differences in carers’ health at baseline and postcensal survival is analysed using Cox proportional hazards models.

Results 12.2% of study members reported providing 1–19 h of care and 5.4% reported providing 20 or more hours. While carers were significantly more likely to report poorer health at baseline, survival analyses suggested that they were at a significantly lower risk of dying. This comparative advantage also held when the analyses were restricted to individuals living with at least one person with poor health.

Conclusions The comparative mortality advantage revealed in this analysis challenges common characterisations of carers' health and draws attention to important differences in the way carers are defined in existing analyses. The survival results are consistent with work using similar data for Northern Ireland. However, the study also affords more uniform conclusions about carers’ baseline health and this provides grounds for questioning existing hypotheses about the reasons for this advantage.

  • Mortality
  • Ageing
  • Record Linkage

Statistics from


Most care for people who need help with everyday activities is provided by family members and other ‘informal’ carers.1 ,2 Delivering this support can represent a considerable undertaking and it has often been suggested that the attendant stress and physical strain might result in health outcomes that are comparatively poorer than those of non-carers. Many studies of carers’ physical and mental-health outcomes provide evidence to support this perspective.3–15 However, less attention has been paid to mortality. The results from existing analyses are inconsistent, with research by Christakis and Allison16 and Schulz and Beach17 suggesting that carers have a higher risk of dying and work by O'Reilly et al18 and Brown et al19 indicating that they experience a comparative survivorship advantage.

In particular, the ‘lower risk’ findings present a clear challenge to the notion that carers are necessarily more likely to have poor-health outcomes, and in doing so, they also complement other work which suggests that it is the fact of having a relative in need of care, rather than the provision of care itself, which may be associated with poorer health.20 Overall, this suggests that it is important to assess the validity of these provocative findings and to try and understand how they might relate to the broader body of work on carers’ health outcomes. Bearing this in mind, the following work investigates the comparative mortality risks of informal carers in England and Wales.

The study begins by examining the relative health status of carers at baseline. It then assesses the relationship between care provision status and mortality before moving to consider whether this pattern differs in subsamples where the carer and non-carer groups are all faced with the stress of living with a sick household member. All of the analyses are carried out using the Office for National Statistics-Longitudinal Study of England and Wales (LS) and this also enables us to assess and refine our understanding of the relationships identified in O'Reilly et al's work with LS’s Northern Ireland sister study.


Study population

The LS is a representative record linkage study for the population of England and Wales. It contains approximately 1% of the total population of these countries and includes census information, as well as linked data on mortality, births to sample mothers, emigrations and cancer registrations. The 1st and 9th deciles of the age distributions for carers providing 1–19 h/week of care or 20 or more hours were 30|68 and 33|75 respectively. Given this and the fact that certain Census questions are only asked of those aged under 75, the following work focuses specifically on analysing the mortality risks of non-institutionalised LS members aged 35–74 years at the time of the 2001 Census. As with its Northern Ireland counterpart, care provision status is defined directly using LS members’ responses to the 2001 Census question on caring activities.

These questions ask those aged 16 and over to indicate the number of hours they looked after, or gave any help or support to, family members, friends, neighbours or others because of long-term physical or mental ill health or disability, or problems related to old age. In keeping with other work on informal care in England and Wales, we make a distinction between those who reported providing no care, those who reported 1–19 h and those who reported providing 20 or more hours a week.21 ,22 Throughout the paper, those providing 1–19 h are referred to as ‘light carers’ and those providing 20 or more hours as ‘heavy’ carers. This terminology is common in research on caregiving. However, one should be careful to avoid assuming that it necessarily denotes differences in the overall difficulty of the work provided by the two caregiving groups.

To maintain consistency with the subsequent analysis of LS members living with sick household members, the main results presented here only relate to those living with others in private residences on Census day. These restrictions provide a raw sample of 217 410, which then reduces to 178 368 LS members with complete records for all variables. The majority of these sample members reported providing no care (82.4%), with 12.2% providing light care and 5.4% providing heavier support. This pattern is similar to the distribution of carers among those living alone and those living with others but with incomplete records for all variables. Models for those living alone and those aged ≥16 or ≥75 were also fitted as part of this project. The results derived from these samples are similar to those reported in the main text. More information about the work carried out with these alternative samples is provided in the appendix (see online supplementary appendix 1).

To assess whether the relationship between mortality and caregiving is likely to be distorted by group differences in the household demand for care, the analysis also made use of linked Census data for those individuals who were living with LS members at baseline. Drawing on responses to the Census questions on self-reported health (SRH) and limiting long-term illness (LLTI), binary variables were constructed, which indicated whether a Study member was living with at least one individual who reported that they had an LLTI or that their health was ‘not good’ (rather than ‘good’ or ‘fair’). These variables were then used to subset the main sample in order to analyse carers’ comparative mortality risks in samples where all individuals are faced with the stress of having at least one household member with poor health.

Mortality data

Carers' mortality risks were analysed on the basis of LS members’ linked death records for the period from 29 April 2001 (Census day) until 31 December 2009. At the time of writing, this is the most up-to-date death data that is available for the dataset. Overall, it provides a total of 10271 deaths for the main sample of interest, 3643 for those living with one or more persons with an LLTI and 1906 for those living with one or more persons with health that was ‘not good’. Of the deaths in the main sample, 220 (2.1%) occurred to those who reported having ‘not good’ health in 2001; 2036 (19.8) occurred to those who reported having an LLTI; and 3138 (30.6%) occurred to those who reported having an LLTI and having health that was ‘not good’.

Control variables

All of the analyses controlled for variables that previous research with the LS has found to be associated with caregiving.21 ,23 More specifically, they included variables for age, marital status, ethnicity and various indicators of socioeconomic status such as housing tenure, educational attainment, economic activity and area-level deprivation quintiles. The variable coding for each of these variables is provided in the appendix (see online supplementary appendix 1). In the survival analyses, census variables for SRH and LLTI were used to control for the baseline health status of carers. The SRH question asks respondents to rate their health as ‘good’, ‘fairly good’ or ‘not good’, whereas the LLTI question asks respondents to report if they have a ‘health problem or disability which limits their daily activities or the work they can do, including problems that are due to old age’.

Statistical modelling

The study began by using logit models to look at baseline differences in the health of the carers before fitting Cox proportional hazard models in order to investigate their comparative risk of mortality in the postcensal period. All of the models were first fitted using a grouped age variable and then rerun with age entered as a continuous, quadratic and cubed variable. Whilst the alternative age terms sometimes resulted in small improvements in the overall model fit, the model coefficients tended to be very similar and did not favour an alternative interpretation of the estimated relationships. The grouped models are presented in the following sections for the purposes of simplicity and overall comparability.

Table 1

Characteristics of those living with others and aged 35–74 in 2001 by care provision status

The initial postestimation tests suggested that the fully adjusted Cox models were biased by a lack of proportionality in the control variable for baseline SRH. Following Cleves et al,24 this problem was addressed by allowing the hazard of death to vary freely for each stratum of the SRH variable. In contrast to the initial specification, the postestimation tests for the revised models suggested that there was no significant violation of the assumption of proportionality. This adjustment did not alter the main conclusions about the influence of caregiving and the HRs were very similar to those produced by the uncorrected model.

Overall, this suggests that the lack of proportionality in the uncorrected data has a fairly minimal impact on the results. Nevertheless, the following sections present results for models where the hazard of death has been allowed to vary freely by baseline SRH.


Baseline differences in the health of carers and non-carers

The table overleaf provides some descriptive statistics for the main sample. Overall, it shows that non-carers have a profile that is more similar to light carers and it suggests that any differences between the two groups tend to be in the opposite direction of those between the heavy carers and the non-carers. For example, a comparatively larger proportion of those doing light care report living in owned accommodation and a smaller proportion report renting, whereas a comparatively larger proportion of heavy carers report renting and a smaller proportion report being owner occupiers.

Table 2 illustrates the differences in the health of carers and non-carers at baseline in 2001 by modelling carers’ comparative odds of having an LLTI or poorer SRH. Overall, the results for the LLTI and SRH models paint a similar picture of carers’ health. With the exception of female heavy carers in the LLTI model, the ORs for the fully adjusted models suggest that, when controlling for observed differences in age and socioeconomic characteristics, both heavy and light carers tend to have a significantly higher odds of reporting poor health at baseline. As we shall see in the discussion, the uniformity of these conclusions is important in helping to interpret the relationship with mortality that is revealed in the subsequent survival analyses.

Table 2

Differences in the health of sample carers in 2001

Carers’ comparative risk of dying

By the end of 2009, 5.8% of the main sample members had died. The deaths constituted 5.9% of those providing no care, 4.1% of those providing light care and 7.5% of those providing heavy care. The top three lines in table 3 provide the results of sex-specific Cox models predicting carers’ hazard of mortality relative to non-carers. The HRs for the fully adjusted models indicate that those who reported providing heavy or light care in 2001 were significantly less likely to die than those who reported doing no care. This comparative survivorship advantage is more pronounced for light carers (men: HR 0.81, CI 0.75 to 0.89; women: HR 0.74, CI 0.66 to 0.83) than it is for those who reported providing more hours of care (men: HR 0.87, CI 0.79 to 0.97; women: HR 0.80, CI 0.71 to 0.89).

Table 3

Hazard of dying for those providing care in 2001 relative to those providing no care

A likelihood ratio test for a combined model with a sex–carer interaction suggested that there were no significant differences in the mortality risks of male and female carers. The results from the fully adjusted and age/SE models are similar. However, when there is no control for baseline health status, the comparative survivorship advantage is non-significant for heavy male carers (HR 0.93, CI 0.84 to 1.03).

Mortality of individuals living with a sick household member

The bottom two lines in table 3 provide the results of fully adjusted Cox models predicting the comparative mortality risks of carers in two subsamples where all individuals were living with at least one person who reported poor health. More specifically, these lines show the comparative mortality risks of carers who were living with at least one household member who reported having an LLTI or ‘not good’ health in 2001. In contrast to the main sample, the individuals in these samples were comparatively older, had higher proportions of individuals with poor health, and were more socioeconomically disadvantaged when viewed in terms of housing tenure, economic activity and educational attainment (see online supplementary appendix 1).

Given that these baseline variables are associated with higher mortality, it is perhaps unsurprising that the proportion of postcensal deaths was also higher for these reduced samples, with 8.8% of the ‘not good’ sample and 8.7% of the LLTI sample having died in the postcensal period compared with 5.8% in the sample as a whole. Nevertheless, the likelihood ratio tests for combined models with an interaction between the caregiving and household contexts suggested that there were no significant differences in the comparative mortality risks experienced by carers living with a sick co-resident and those who were not.

In keeping with the results for the main sample, the hazards for both subsamples suggest that light carers tended to have significantly lower mortality than non-carers. Lower mortality hazards were also estimated for those providing heavy care, although the relationship is only significant in the model for females in the subsample where one or more co-residents reported having an LLTI (HR 0.84, CI 0.74 to 0.96). Broadly speaking, these results suggest that the pattern of carer mortality advantage holds when focusing on a population where carers and non-carers are living in a more similar context of care demand.


The analyses presented in the previous sections suggest that those who report providing care tend to have lower mortality than their non-caring counterparts. This finding is consistent with analyses by Brown et al19 and O'Reilly et al18; yet it is also at odds with other studies which have concluded that carers are comparatively more likely to die.16 ,17 The existing analyses of carer mortality do not attempt to reconcile these differing conclusions. However, a simple methodological comparison draws attention to some important differences in the way that these contrasting studies distinguish between carers and non-carers.

The ‘lower risk’ conclusions about caregiving are all drawn from work where carers are directly identified on the basis of self-reported questions about an individuals’ care provision status and the intensity of any support that they provide.18 19 In contrast to this, Schulz and Beach's “higher risk” findings are specifically related to the experience of stressed spousal caregivers17 and Christakis and Allison's complementary insights are derived from comparing the mortality risks of married individuals who experience a spousal hospitalisation to those of other married individuals who do not.16 On closer reflection, it would seem quite possible that the contrasting conclusions about carers’ mortality risks stem, at least in part, from these ‘classification’ differences.

Moreover, whilst it is clear that the groupings used in the ‘lower risk’ analyses can afford insights into the general relationship between caregiving and mortality, it is perhaps more accurate to think of the work by Schulz and Beach and Christakis and Allison as providing insights into the experiences of specific subgroups of the caregiving population. When viewed from this perspective, the results from the ‘lower risk’ analyses provide a more defensible description of the general relationship between caregiving and mortality. However, this should not be taken to imply that the other results are without value. Although the ‘higher risk’ analyses are more limited in scope, they do draw attention to variations within the caregiving population.

It is difficult to carry out such fine-grained analyses with the LS because the dataset contains no information about the nature of the care being provided, other than the number of hours a week that individuals spend engaged in these activities. However, recognising such differences can help us to develop a more nuanced understanding of the caregiving experience, and it would seem important for future research to try to ‘unpack’ the overall mortality relationship in more detail. For example, it would be interesting to know whether there are differences in survivorship associated with the carer's relationship to the care recipient. Or whether there are differences that are associated with the way in which the care is scheduled and provided.

Ideally, when confronted with subgroups of the care population who appear to experience higher risks, one would also take account of group differences in the degree to which individuals are experiencing stress. Indeed, it is entirely conceivable that the aforementioned ‘high-risk’ analyses might actually be picking up group differences in the care context rather than the experience of providing care (eg, the comparatively higher levels of stress reported by Schulz and Beach's spousal carers or an increased level of stress that is associated with the experience of a spousal hospitalisation rather than any caregiving that is assumed to accompany this event).

Given the comparatively advantageous nature of the mortality risks illustrated in the preceding analyses, it is harder to argue that these sorts of differences might account for the pattern of lower mortality risks that is reported in this study. This position is strengthened by the subsample models, which suggest that there is little change in the nature of the relationship between caregiving and mortality when looking at a population where the different care provision groups are likely to face a similar demand for care. However, in the absence of any direct measures, it is difficult to develop more definitive conclusions about the extent to which stress influences the observed relationships and it is important that any discussion is careful to acknowledge this limitation.

Nevertheless, despite the limited scope for the development of more detailed conclusions, analysing the LS does provide strong evidence to suggest that carers experience a comparative survivorship advantage. These results challenge the common characterisation of carers as a comparatively vulnerable group, and as a consequence, they provide grounds for thinking more deeply about the nature of the observed relationships between an individual's health outcomes and their reported care provision status. Interestingly, some researchers have suggested that carers’ comparative resilience to negative physiological outcomes might derive from a qualitative difference between those who care and those who do not.

For example, both Taylor et al25 and Ross et al26 suggest that the tentative patterns of advantage that are apparent in their own analyses might, in part, arise because the caregiving groups are drawn from a relatively healthier subset of the overall population. The existence of this sort of selection bias is supported by some longitudinal research, which has found that physically healthier individuals are more likely to enter into a caregiving role.27 O'Reilly et al find mixed evidence to support the idea that carers are healthier than non-carers, with baseline analyses finding evidence to suggest that carers were less likely to have an LLTI but that heavy carers were also less likely to characterise their health as being ‘good’.

The authors employ various techniques to try and control for possible distortions in the relationship with mortality, which might be created by selection effects. However, the ambiguous character of the baseline health relationships means that it is difficult to develop any firmer ideas about the actual existence of any such effect. It is here that this work on England and Wales is particularly valuable. While analysing the LS reveals the same carer mortality advantage observed in the Northern Ireland analysis, the Study provides more uniform conclusions about comparative differences in carers’ health at baseline in 2001.

In keeping with other work on the LS, the baseline health analyses show that carers are more likely than non-carers to report having an LLTI or poorer SRH. Importantly, by failing to support the idea that carers are necessarily healthier than non-carers, these findings raise doubts about the existence of the selection effect and suggest that it is important for further research to look in more detail at the health of carers before the onset of caregiving, as well as their health during the period when this support is being provided. In particular, it would be useful to carry out additional longitudinal analyses that focus on different types of health outcome so as to work out whether the selection hypothesis can be rejected or whether the health variables existing in these data are simply incapable of capturing the underlying differences between carers and non-carers.

What is already known on this subject?

  • Many studies of carers’ health outcomes have found evidence to suggest that informal carers are at a higher risk of experiencing poorer health. However, there is little work looking at carers’ mortality risks and the results from existing analyses are inconsistent.

What this study adds?

  • This study analyses carers’ comparative mortality risks using the Office for National Statistics Longitudinal Study for England and Wales and finds that carers are significantly less vulnerable to mortality than non-carers. These results are in keeping with similar research on Northern Ireland. However, the England and Wales data also afford more uniform conclusions about a carer's baseline health status. Importantly, these findings contradict the idea that carers are necessarily healthier than non-carers, and in doing so, they suggest that there is a need for more work which focuses specifically on assessing the validity of the selection hypothesis as an explanation of carers’ survivorship advantage.


Census output is Crown copyright and is reproduced with the permission of the Controller of HMSO and the Queen’s Printer for Scotland. The permission of the Office for National Statistics to use the Longitudinal Study is gratefully acknowledged, as is the help provided by staff of the Centre for Longitudinal Study Information & User Support (CeLSIUS). CeLSIUS is supported by the ESRC Census of Population Programme (Award Ref: RES-348-25-0004). This paper was produced as part of the CeLSIUS project No. 30147. The authors alone are responsible for the interpretation of the data.


Supplementary materials

  • Supplementary Data

    This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.

    Files in this Data Supplement:


  • Contributors All of the authors jointly conceived the study, and read and approved the final version of the manuscript. SR designed the study, analysed the data and drafted the manuscript. EG contributed to the study design, advised on the data analysis and commented on the manuscript. DOR contributed to the study design, advised on the data analysis and commented on the manuscript.

  • Competing interests None.

  • Ethics approval The LS' original ethical approval was granted by the Patient Information Advisory Group (PIAG). This organisation was replaced by the National Information Governance Board for Health and Social Care (NIGB) in 2008.The permission of the Office for National Statistics to use the Longitudinal Study is gratefully acknowledged, as is the help provided by staff of the Centre for Longitudinal Study Information & User Support (CeLSIUS). CeLSIUS is supported by the ESRC Census of Population Programme (Award Ref: RES-348-25-0004). This paper was produced as part of the CeLSIUS project No. 30147.

  • Provenance and peer review Not commissioned; externally peer reviewed.

  • Data sharing statement The analyses reported in the manuscript are based on a dataset which includes microdata from the Census of England and Wales. As a consequence, specific legislation governs the way in which the data are accessed and released. In particular, researchers must apply formally to the Office for National Statistics (ONS) in order to gain access to the data and all analytical work must be carried out on-site at the ONS. Before any results can be released into the public domain, it is also obligatory to submit them to ONS for clearance. Unfortunately, these restrictions mean that it is not possible to share unpublished data from this project.

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.