Background Prior evidence suggests that caregiving may increase risk of cardiovascular disease (CVD) onset. This association has never been examined in a nationally (USA) representative sample, and prior studies could not fully control for socioeconomic confounders. This paper seeks to estimate the association between spousal caregiving and incident CVD in older Americans.
Methods Married, CVD-free Health and Retirement Study respondents aged 50+ years (n=8472) were followed up to 8 years (1669 new stroke or heart disease diagnoses). Current caregiving exposure was defined as assisting a spouse with basic or instrumental activities of daily living ≥14 h/week according to the care recipients' report in the most recent prior biennial survey; we define providing ≥14 h/week of care at two consecutive biennial surveys as ‘long-term caregiving’. Inverse probability weighted discrete-time hazard models with time-updated exposure and covariate information (including socioeconomic and cardiovascular risk factors) were used to estimate the effect of caregiving on incident CVD.
Results Caregiving significantly predicted CVD incidence (HR=1.35, 95% CI 1.06 to 1.68) in the population overall. Long-term caregiving was associated with double the risk of CVD onset (HR=1.95, 95% CI 1.19 to 3.18). This association for long-term care givers varied significantly by race (p<0.01): caregiving predicted CVD onset for white (HR=2.37, 95% CI 1.43 to 3.92) but not for non-white (HR=0.28, 95% CI 0.06 to 1.28).
Conclusions Spousal caregiving independently predicted risk of CVD in a large sample of US adults. There was significant evidence that the effect for long-term care givers differs for non-whites and white.
- CHD/coronary heart
- social epidemiology
- social inequalities
Statistics from Altmetric.com
Caregiving for an ill or disabled spouse is extremely common: among older adult care givers, about 25% of them provide care to a spouse.1 Caregiving frequently entails substantial physical, emotional or financial burdens.2 Caregiving demands can also adversely affect the health of care givers directly through stress3 ,4 and physical strain or indirectly through behavioural changes like lack of exercise5 ,6 and self-management of chronic conditions.7 Caregiving has been shown to predict elevated levels of physiologic indicators of cardiovascular risk8 ,9; Framingham risk scores10 ,11; as well as increased risk of hypertension,12 cardiovascular disease (CVD)13 and coronary heart disease (CHD)14 ,15 onset; and increased mortality.16
Despite the potentially large public health impact of this exposure, there are important gaps in our understanding of the health risks associated with caregiving. After searching MEDLINE using relevant MeSH topics for literature in this area, we are not aware of any longitudinal study that has assessed the association of caregiving with risk of incident CVD in a nationally (USA) representative sample. Although chronicity of caregiving is important, there are no estimates of CVD for both current and long-term care givers. Furthermore, few studies have followed potential care givers prospectively or invoked special regression approaches necessary17 to address time-varying variables that may confound or also partially mediate the association between caregiving and CVD (eg, depression and care giver health behaviours). Also, there is still little known about the risk of CVD onset associated with caregiving among racial minorities.
Using follow-up data from the Health and Retirement Study (HRS),18 this study addresses two research questions. First, is either current or long-term caregiving associated with CVD incidence in a nationally representative sample of middle and older age men and women? Second, are those associations modified by race, gender or depressive symptoms?
HRS is a longitudinal survey of a national sample of US adults aged 50+ years and their spouses. Details of the study are provided elsewhere.18 ,19 Enrolments occurred in 1992, 1993 or 1998 (based on respondent's and spouse's birth years) with biennial interviews (or proxy interviews for decedent participants) through 2008. Retention rates through 2008 were above 80%. HRS was approved by the University of Michigan Health Sciences Human Subjects Committee, and these analyses were determined exempt by Harvard School of Public Health Office of Human Research Administration.
This study was restricted to married or partnered HRS participants born 1900–1947 and interviewed in 2000, which was the earliest year when caregiving assessments were consistently worded and asked with respect to spouses, and to individuals with no reported history of CVD as of the 2000 interview.
The weighted HRS sample included 11 474 respondents who were age eligible and married with both spouses in the study in 2000; persons were excluded on the basis of prevalent events at baseline and missing key covariate information. Of the 2831 (24.7%) respondents excluded because of a prior diagnosis of CVD in 2000, 38% were women and 17% were non-Caucasians. An additional 171 (1.5%) were excluded due to missing information on basic demographic factors (age, sex, race and ethnicity), leaving 8472 individuals contributing person-time to the primary analyses.
Spousal caregiving demand was calculated in each biennial interview wave (2000–2006), based on the care recipient's report regarding assistance with activities of daily living (including help with getting across a room, dressing, bathing, eating, getting in and out of bed and using the toilet) and instrumental activities of daily living (prepare meals, shop for groceries, make telephone calls, take medications). HRS respondents listed the people who provided them assistance in the last month and the number of days and hours of care they provided. An indicator variable was generated for whether a spouse was a listed care giver, and we calculated amount of care provided per week for these analyses.
After examining the distribution of weekly hours of care among care givers, we did not identify any clear threshold of hours of care provided. Thus, to be consistent with other studies of caregiving using these data,20 ,21 the hours of care were dichotomised to <14 and ≥14 h of care per week as the primary exposure classification (hereafter referred to as ‘current high caregiving’). To assess the validity of this measure of caregiving, we examined a subsample of 4648 respondents who reported on spousal demands for a supplemental psychosocial questionnaire given in the 2008 wave of HRS. Respondents who provided ≥14 h/week of spousal care were 1.7 times as likely as respondents who provided <14 h/week of care and 3.2 times as likely as respondents who did not provide any care to report their spouse made too many demands of them (p<0.001 for differences between groups; details available from authors).
For analyses of the effects of long-term caregiving, we considered any respondent who was classed as a high care giver during both the most recent two prior interview waves (separated by approximately 2 years) as a long-term care giver.
Onset of CVD between 2000 and 2008 was assessed biennially by the care giver's self-report of a doctor's diagnosis of either stroke or heart disease. For participants who had died and those unavailable for a direct interview, interviews were conducted with proxy informants, typically spouses. Incident strokes were assessed based on self-report or proxy report of a doctor's diagnosis of the event (“Has a doctor ever told you that you had a stroke?”). No information on stroke subtypes was obtained; temporary ischaemic attacks were not coded as strokes. Heart disease was assessed similarly: “Has a doctor ever told you that you had a heart attack, CHD, angina, congestive heart failure, or other heart problems?” Responses were combined from these two questions into a single variable measuring CVD, coded ‘yes’ if a respondent reported either a stroke or heart disease in a given wave and ‘no’ if they reported neither. The combined results are presented because the results of stroke and heart disease separately were qualitatively similar with wider CIs.
Six sets of covariates—demographic, socioeconomic status (SES), health risk factors, comorbid conditions, self-assessed health and care recipient memory illness—that likely confound the effect of caregiving on risk of CVD onset were considered. The demographic characteristics comprised baseline age and age squared,22 non-white race (whites/non-whites), Hispanic ethnicity23 and female sex. SES variables included years of education, Southern place of birth (yes/no),24 per capita income at baseline (household income divided by square root of household members)25 and maternal education (<8, ≥8 years).26 Care giver health risk factors consisted of vigorous physical activity (“On average over the last 12 months have you participated in vigorous physical activity or exercise three times a week or more?”, yes/no),27 alcohol use (in the last 2 weeks, any/none),28ever-smoker status and current smoker status29 and body mass index in kilograms per square metre.30 Comorbid conditions include self-reported diagnoses of hypertension and diabetes.22 Self-assessed health measures include elevated depressive symptoms31 ,32 (measured with a modified eight-item Center for Epidemiological Studies–Depression Scale, Cronbach's α=0.81,33 dichotomised at <3, ≥3),34 fair/poor self-rated health,35 and care giver activity of daily living and instrumental activity of daily living limitations (≥1/0, respectively).36 Care recipient's memory illness was defined a self-reported doctor's diagnosis of a memory illness. For observations with missing values, continuous variables were set to the mean and categorical variables to the referent group and included a missing indicator variable in the weighting models.
Several of the covariates are potentially affected by caregiving (eg, care giver and care recipient health characteristics) and may thus mediate the effects of past caregiving on CVD. To avoid adjusting for variables downstream of the primary exposure, we used the values of these covariates as assessed in the wave prior to the caregiving assessment (eg, 1998 covariates to control for caregiving demands in 2000) in weighting models to adjust for confounding (detailed below). Demographic and SES covariates are time constant and the values reported in 2000 were considered as baseline.
Methods of analysis
Elevation in hazard of CVD associated with high caregiving was estimated with discrete-time hazard models. Individuals contributed person-time to the models if they remained alive, married and had not yet reported a diagnosis of CVD. Participants were censored after the dissolution of the marriage to avoid conflating caregiving strain with bereavement effects.37 ,38
Unadjusted and adjusted hazard ratios (HRs) are presented that added the groups of covariates defined above to the models sequentially consistent with the likely temporal ordering of the confounders. We estimate current caregiving models from 2000 as the first exposure year and long-term caregiving models with 2002 as the first exposure year to calculate an exposure that includes both the 2000 and 2002 caregiving behaviour. Primary analyses were conducted in SAS V.9.2 with PROC GENMOD using a logit link, robust variance estimates and weights as described below.
Time-varying confounders were adjusted for by using inverse probability of treatment weights.17 ,39 This approach allowed adjusting for variables likely to be both mediators and confounders, for example, health behaviours that may be affected by high caregiving and may also impair a spouse's ability to provide care. Time-lagged values of the time-varying covariates were used to avoid bias from adjusting for mediators. A similar approach to handle possible bias resulting from selective dropout or survival was used. Thus, each observation was weighted by the product of (1) the inverse of the probability that individual was alive at the exposure wave, (2) the inverse of the probability that individual (conditional on having survived) was in the sample at both exposure and outcome waves and (3) the inverse of the probability that the individual received the treatment he or she actually received. Stabilised weights were calculated using previously described protocols.39 ,40 The weights were truncated at the value of 99th percentile to address the weights' skewed distribution.
Possible modifiers of the association between caregiving and CVD risk were also tested: sex, race and elevation in depressive symptoms for the care giver. Both the stratified effect estimates and a test of interaction in the pooled model are presented; the model was a multiplicative model, so the interactions tested differences in effect sizes on a multiplicative scale, rather than on an additive scale.
HRS used a multistage clustered sample. The HRS sampling weights were applied to make the population representative of the 2000 US population aged 50+ years. Models that do not account for clustering are presented, as no difference between models that account for clustering at the household level and those that did not was found.
Participant characteristics are summarised in table 1. There were 1669 total new CVD diagnoses during 50 138 person-years of follow-up. At baseline, current and long-term care givers differed from the total sample in terms of demographics (older, more non-white and Hispanic respondents), socioeconomic position (lower education, mother's education and per capita income), self-rated health (higher depressive symptoms, fair/poor health rating, functional disabilities) and prevalence of spouses with memory illnesses (higher among care givers). Over follow-up, 3.7% (person-years=1822) were ‘exposed’ as current care givers and 1.3% (person-years=652) as long-term care givers.
In the unadjusted model, current high caregiving was associated with 87% excess risk of CVD (HR=1.87, 95% CI 1.48 to 2.36; table 2). Once adjusted for demographics, SES and all health risk factors, current high caregiving remained significantly associated with a 35% increased risk onset of CVD (HR=1.35, 95% CI 1.06 to 1.68). For long-term caregiving, there was a nearly twofold excess risk of CVD onset in the fully adjusted model (HR=1.95, 95% CI 1.19 to 3.18).
The risk of CVD onset associated with current caregiving did not vary significantly by race, gender or care giver depression (table 3). While there was a significant excess risk of CVD onset for white care givers (HR=1.45, 95% CI 1.12 to 1.88), non-white care givers had a non-significant decreased estimated risk (HR=0.85, 95% CI 0.37 to 1.94) and the test of the interactions by race was not significant (p>0.10).
The estimated risk of CVD onset associated with long-term caregiving varied significantly by race but not by gender or depressive symptomatology. Among long-term care givers, Caucasians had a significant excess risk (HR=2.37, 95% CI 1.43 to 3.92) and non-Caucasians had a reduced risk (HR=0.28, 95% CI 0.06 to 1.28, p<0.01 for test of interaction). Non-whites and non-Caucasians had relatively few events (n=121) and the CI included the null. There was a significantly elevated risk among male care givers (HR=2.02, 95% CI 1.01 to 4.03) and depressed care givers (HR=2.67, 95% CI 1.14 to 6.23), but tests for the interactions by these factors were not significant (p>0.10).
In this nationally representative sample of Americans aged 50+ years, both current and long-term high caregiving predicted onset of CVD, even after a number of covariates related to demographics, SES and health status were introduced. The relationship with long-term care giving differed by race, with significant evidence of a harmful relationship among Caucasians and suggestive evidence of a protective relationship among non-Caucasians.
Comparisons with prior literature
The largest of the prior studies using cardiovascular diagnoses as outcomes14 found providing a spouse ≥9 h of care per week was associated with a near doubling of risk of CHD onset (RR=1.82, 95% CI 1.08 to 3.05) among women in the Nurses' Health Study (NHS). While our findings suggest a more modest excess risk of CVD associated with current caregiving (HR=1.31 vs 1.82), our estimates of the effects of long-term caregiving were higher than the NHS estimates (HR=1.91 vs 1.82), suggesting that the chronicity of the care may be an important aspect of caregiving to model explicitly. For example, the NHS population (predominantly white female nurses) may be more affected by the strain of chronic caregiving than a general population sample. These results are also consistent with the wider literature of studies about strain13 and distress and depression41 among spouses and care givers predicting elevated CVD risk.
This study differs with the literature that finds caregiving provides a beneficial health effect.42 Brown et al21 found that among the oldest participants in the HRS cohort, caregiving at baseline predicted reduced mortality risk over 7 years of follow-up. Differences in the population age, the outcome examined, the control variables used or the analytic approach may explain this divergence, although the qualitative difference in sign, even in unadjusted models, is surprising.
Prior research also suggests differences in the health effects of caregiving by race, specifically that non-whites often are less impacted by caregiving burdens than whites.21 In this study, risk of CVD onset varied significantly by race among long-term care givers; whites had a significant excess risk and non-whites had a decreased non-significant risk. Other studies found that positive caregiving attitudes among non-white care givers were predicted by religiosity, lower perceived burden, lower SES and differences in social support,43 ,44 such as greater quantity45 and diversity46 of support sources. It is also important to contextualise the differences between the magnitudes of the effect of caregiving in light of absolute racial disparities in health.
Stress may directly modify the risk of CVD3 ,47 ,48 associated with caregiving. For example, increased hypothalamic-pituitary-adrenal activity and cortisol levels resulting in increased C reactive protein, platelet activation and atherogenesis9; increased fibrinogen4; and documented increases in blood pressure and inflammatory markers among care givers10 ,11 ,49–51 further suggest that this stress pathway to CVD may be applicable among care givers. However, other mechanisms, such as health behaviours, may also contribute to the health risks associated with caregiving.
Although the HRS is well suited to investigate spousal caregiving, some limitations should be considered. Self-reported outcome events were subject to misclassification and have been shown to be imperfectly correlated with clinically verified outcomes in other studies.52–55 However, misclassification is substantially reduced by use of a broadly classified outcome as done here, and prior analyses of self-reported stroke in HRS indicated that results were consistent with other studies using medically verified events.56
Exposures were also subject to misclassification. In HRS, the care recipient, not the care giver, reported the amount of care provided. This may have led to misclassification of care from the recipient, though it is difficult to anticipate the direction of this potential bias. This design, however, avoids same-response bias, which may arise if the same person reports the exposure and the outcome. Also, other literature suggests caregiving burden is modified by the strain or perceived stress of caregiving. HRS does not routinely include questions about perceived stress; however, there is evidence of elevated hazard of CVD onset among care givers with higher depressive symptoms, although the test for the interaction by depressive symptomatology was not significant.
While HRS is one of the largest longitudinal studies of older US adults, CIs in many cases were relatively wide; only a small number of events were observed in several substrata, especially among long-term care givers. As more research on this topic becomes available, meta-analyses may be valuable.
Despite these limitations, these findings have several methodological advantages and build importantly on prior research. A major challenge in research on the health effects of caregiving has been avoiding the potential confounding introduced by selection into caregiving, in which healthier individuals may be more likely to take on caregiving roles. By using a longitudinal analysis with time-varying exposures, this study could model selection into caregiving roles without biasing the effect estimates by controlling mediating pathways. Though many studies of caregiving are cross sectional, this longitudinal design facilitated modelling the chronicity of caregiving with respect to risk of CVD onset.
Implications for future research, policy and practice
Caregiving burdens often multiply at approximately the same age when risk for CVD onset is increasing rapidly. Studies investigating subgroup variations in the health effects of caregiving—and particularly research on factors that may exacerbate or remediate these effects—are needed. Studies need to extend beyond using race/ethnicity alone to understand the divergent CVD risk by race; race/ethnicity may be a proxy for values, attitudes and resources of the care givers.57 Research should also examine mechanisms about the duration of care that may vary by these factors, as the risk for CVD differed in this study by race among long-term, but not current, care givers. Additional longitudinal studies that explicitly model chronicity of caregiving are also needed.
Many US adults provide care for their spouses. This care was associated with a modest significant elevation in risk of CVD onset among current spousal care givers overall and a near doubling of risk of CVD onset among spousal care givers who report providing high caregiving over two survey waves. There was also a significant difference in this relationship by race for long-term care givers. More research is necessary to elucidate sources of resilience and types of support that could reduce the detrimental health-related risks of caregiving.
What is already known on this subject
Spousal caregiving has been associated with biological markers of stress and inflammation, hypertension, deleterious health behaviours, mortality and, among women, risk of coronary heart disease onset.
No study to date has estimated the association between caregiving and cardiovascular disease onset in a nationally (USA) representative sample of older adults followed longitudinally nor whether this association varies by socio-demographic or health factors.
What this study adds
This study offers the first nationally representative longitudinal estimates of the risk of cardiovascular disease onset associated with providing care for one's spouse, both currently and consistently over a 2+-year period.
We estimate a modest effect estimate of current caregiving (31% excess risk) but an almost twofold increase in risk of cardiovascular disease associated with longer term provision of care.
Furthermore, this study finds evidence that this association varies significantly by race among long-term care givers, with an excess risk for cardiovascular disease onset among white care givers only.
Funding The authors gratefully acknowledge financial support from the American Heart Association grants 09PRE2080078 (BDC) and 10SDG2640243 (MMG), the National Institute on Aging (AG034385-01) (MMG) and National Heart, Lung and Blood Institute (T32-HL098048-01) (JRM). The HRS (Health and Retirement Study) is sponsored by the National Institute on Ageing (grant number NIA U01AG009740) and is conducted by the University of Michigan. We presented parts of this work at the American Public Health Association Annual Meeting on 10 November 2010.
Competing interests None.
Ethics approval Ethics approval was provided by University of Michigan (original study) and Harvard School of Public Health (this analysis).
Provenance and peer review Not commissioned; externally peer reviewed.
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.