The Relationship Between Caregiving and Mortality After Accounting for Time-Varying Caregiver Status and Addressing the Healthy Caregiver Hypothesis

Abstract

Background.

Previous studies have shown inconsistent associations between caregiving and mortality. This may be due to analyzing caregiver status at baseline only, and that better health is probably related to taking on caregiving responsibilities and continuing in that role. The latter is termed The Healthy Caregiver Hypothesis, similar to the Healthy Worker Effect in occupational epidemiology. We applied common approaches from occupational epidemiology to evaluate the association between caregiving and mortality, including treating caregiving as time-varying and lagging exposure up to 5 years.

Methods.

Caregiving status among 1,068 women (baseline mean age = 81.0 years; 35% caregivers) participating in the Caregiver-Study of Osteoporotic Fractures study was assessed at five interviews conducted between 1999 and 2009. Mortality was determined through January 2012. Cox proportional hazards models were used to estimate adjusted hazard ratios and 95% confidence intervals adjusted for sociodemographics, perceived stress, and functional limitations.

Results.

A total of 483 participants died during follow-up (38.8% and 48.7% of baseline caregivers and noncaregivers, respectively). Using baseline caregiving status, the association with mortality was 0.77, 0.62–0.95. Models of time-varying caregiving status showed a more pronounced reduction in mortality in current caregivers (hazard ratios = 0.54, 0.38–0.75), which diminished with longer lag periods (3-year lag hazard ratio = 0.68, 0.52–0.88, 5-year lag hazard ratios = 0.76, 0.60–0.95).

Conclusions.

Overall, caregivers had lower mortality rates than noncaregivers in all analyses. These associations were sensitive to the lagged period, indicating that the timing of leaving caregiving does influence this relationship and should be considered in future investigations.

Caregiving is stressful (1) and is often referred to as a “career” because its conditions and obligations mimic those of employment (2), but how caregiving affects mortality remains uncertain. Some studies have found higher rates of morbidity (3) and mortality in caregivers than noncaregivers (4,5), supporting the theory that the chronic stress of caregiving leads to adverse health outcomes (6). Other studies have found mixed results (7) or lower mortality rates in caregivers (8–11). However, all of these studies analyzed caregiving based solely on caregiver status at baseline and ignored whether persons stopped caregiving over the follow-up period. These analytic approaches may not have reflected the association between caregiving over time and mortality. In this article, we propose how strategies used in occupational epidemiology may address these analytic issues, and apply these strategies in a large sample of community-dwelling older women followed for 13 years.

Although stress theory explains the psychological distress (1) and immune dysregulation (12) experienced by many caregivers, it does not explain why some older caregivers have better physical functioning (7,13,14) and lower mortality rates (8–11) than noncaregivers. These latter observations illustrate the Healthy Caregiver Hypothesis, which is defined as follows: Adults who are healthier are more likely to become caregivers and to remain in that role, that continuing to perform caregiving activities helps to maintain health, and that persons stop caregiving when their health declines (10,15). Moreover, caregiving may impart health benefits through feeling appreciated by the care recipient, and having a sense of purpose in life (16). These components of the Healthy Caregiver Hypothesis correspond to the Healthy Worker Effect in occupational epidemiology. Analytic techniques used to minimize the Healthy Worker Effect could be used to obtain more accurate estimates of the impact of caregiving on mortality. The advantage of mortality over other measures of health is that its reporting is uninfluenced by self-report or access to services.

A key aspect of the Healthy Worker Effect is that workers in poorer health may move to jobs with lower exposure, leaving healthier workers in jobs where they continue to acquire exposure (17). Just as workers’ exposure to potentially harmful substances may vary over time, caregiver status varies over time (18). Yet studies to date have treated caregiver status as fixed from baseline throughout the follow-up period, which may misclassify caregiving status over time. Analytic techniques that allow caregiver status to vary during the follow-up period may more accurately reflect participants’ true exposure experience, thereby providing a truer picture of the association between caregiving and mortality.

To address the Healthy Worker Effect, a common strategy is to lag exposures. The rationale is that symptoms of disease cause workers to reduce exposure or leave work, perhaps years prior to death, while those who remain healthy continue to be employed and therefore exposed. In lagging exposure, more recent exposure is excluded in order to reflect the biologically relevant exposure period in workers who developed the outcome (19). This strategy may be applied in studies of caregiving outcomes since persons who experience symptoms of disease may cease caregiving in advance of death.

In studies investigating whether an exposure increases the risk of mortality, age is often the strongest predictor of mortality. The mean age of caregivers and noncaregivers has differed at baseline (3,10) and prior to matching (11) in studies reporting this information. In occupational epidemiology, exposure accumulates with increasing age (20). A common practice in the analysis of occupational cohorts is to tightly control for age through risk-set sampling based on age at-risk of death rather than time-in-study as the analytic time scale (17,20). This approach could reduce potential confounding by baseline age differences in caregivers and noncaregivers.

We previously reported a reduced risk of mortality over 8 years among caregivers in the Caregiver-Study of Osteoporotic Fractures (Caregiver-SOF) study, based on baseline assessment of caregiver status (10). In the current study, we apply the aforementioned analytic approaches including time-varying exposures, age-based risk-set sampling, and exposure lagging in order to reduce exposure misclassification and baseline health differences between caregivers and noncaregivers, and more tightly control for age. We hypothesize that the reduction in mortality would be greatest in current caregivers, whether measured at baseline only or as a time-varying exposure; and that this inverse association would diminish with a longer lag period.

Methods

Study Population

The participants in these analyses came from Caregiver-SOF, an ancillary study to the Study of Osteoporotic Fractures (SOF) (21). The SOF sample included 9,704 women aged 65 or older who were recruited between 1986 and 1988 from population-based listings in four areas of the United States: Baltimore County, MD; Minneapolis, MN; Portland, OR; and the Monongahela Valley, PA. Women were excluded if they could not walk without help or had a history of bilateral hip replacement. Between 1996 and 1997, an additional 662 African American women who met the same inclusion criteria were enrolled. Approximately every 2 years, SOF participants have a comprehensive clinical evaluation. The Caregiver-SOF sample is composed of members of the original and African American SOF cohorts who participated in the 6th biennial examination that took place between 1997 and 1999.

Caregiver-SOF Subsample

The Caregiver-SOF sample was identified in two phases, described elsewhere (22). In each phase, a Caregiver Screening Questionnaire was administered to SOF participants to determine if they currently helped a relative or friend with seven instrumental activities of daily living (ADL) tasks (IADLs; use the telephone, get to places out of walking distance, shop, prepare meals, manage medications, manage finances, do heavy housework (23)), and seven basic ADL tasks (walk across a room, groom, transfer from bed to chair, eat, dress, bathe, use the toilet (24)) because that person was physically, cognitively, or mentally unable to perform the task independently. Participants were categorized as caregivers if they helped one or more persons with at least one IADL or ADL task, and as noncaregivers if they did not help anyone with these tasks.

To create the Caregiver-SOF sample, we matched one or two noncaregivers on SOF site, age, race, and zip code to each caregiver who agreed to participate. To avoid effects of residual caregiving-related stress, we excluded noncaregivers who had been caregivers in the past 2 years. The resulting sample included 375 caregivers and 694 noncaregivers. This study was approved by the Institutional Review Boards at each SOF site and Boston University Medical Center. All participants provided written informed consent.

Data Collection

Within 2 weeks of the second screening phase, respondents were interviewed in-person in their home. Up to five face-to-face interviews were conducted with Caregiver-SOF participants from 1999 to 2009: The first three interviews were conducted at annual intervals from 1999 to 2004, and the last two were at approximately 18-month intervals, conducted from 2006 to 2009. Additional follow-up contacts included quarterly postcards and biennial exams for the SOF study (21).

Follow-Up and Mortality

Participants were followed from the baseline Caregiver-SOF interview until whichever occurred first: death, withdrawal from the Caregiver-SOF or SOF study, or end of the follow-up period on January 31, 2012.

Measures

Caregiving status

At each interview, respondents were classified as caregivers or noncaregivers based on whether they assisted someone with any IADL/ADL tasks, as described above.

All-cause mortality

All-cause mortality as of January 31, 2012 was documented through death certificates obtained at each SOF site, as described in previous publications (25). Beginning at the SOF baseline interview, participants were contacted every 4 months. Death certificates were obtained for those who had died. Follow-up for vital status was 99% complete.

Covariables

Sociodemographic variables collected at the Caregiver-SOF baseline interview included self-reported race (white or black), and highest education level (attended college versus less education). Age and marital status (married versus other) were collected at baseline and updated at each follow-up interview. The 14-item perceived stress scale measured the amount of general stress experienced in the past month (possible range 0-56), with higher scores indicating greater stress (26). This measure applies to both caregivers and noncaregivers, with caregivers having higher scores in numerous studies (1). Health status measures included the respondent’s self-reported limitations in performing each of the seven IADLs (0–7) and ADLs (0–7) listed above. Respondents reported whether a physician or health professional had told her that she had hypertension, heart disease, diabetes, or cancer. These conditions were chosen because of their association with mortality in elderly adults. Body mass index (kg/m²) was based on the respondent’s height, measured at her first SOF visit, and weight, measured at the current Caregiver-SOF visit.

Caregiving characteristics

Dichotomous variables indicated whether caregivers cared for a spouse versus another relative or friend, lived with the care recipient, took care of a person with dementia, spent ≥5h/d doing caregiving tasks, and the number of years they were caregiving for the care recipient before the baseline interview. Caregivers also reported the number of IADL tasks (0–7) and ADL tasks (0–7) performed for the care recipient.

Analyses

We performed age-adjusted analyses to compare sociodemographic and health characteristics of caregivers and noncaregivers at the Caregiver-SOF baseline interview.

We estimated hazard ratios and 95% confidence intervals using two sets of Cox proportional hazards models in PROC PHREG (SAS 9.3, Cary, NC). In the first set of analyses, mortality rates were estimated for baseline caregiver status using time-in-study as the analytic timeline and adjusting for baseline covariables.

The second set of analyses was based on age as the analytic timeline. First, we modeled caregiver status at baseline to examine the effect of changing the analytic timeline and including time-varying covariates. Then, we modeled caregiver status as time-varying. For these analyses, we created age-based risk sets. Age at baseline was the youngest age at which participants began to contribute person-time in order to avoid immortal person-time bias; that is, counting person-time time during which, had the outcome occurred, the person would not have been in the study (27). For example, if a woman was age 70 at her Caregiver-SOF baseline interview, she starts to contribute person-time at age of 70. All participants contributed at-risk person-months between their age at baseline and age at which their follow-up ended: death, withdrawal from the SOF or Caregiver-SOF study, or January 31, 2012, whichever occurred first. Separate risk sets were created based on the age-at-death of each participant who died during the follow-up period. All other participants who were alive and in the cohort at that age contributed person-time to the risk set (ie, if a participant died at age of 89, all participants who were alive at age of 89 at any interview contributed person-time to that risk set). Thus, a participant could contribute person-time to multiple age-based risk sets. In these age-based risk sets, caregiver status and covariables were taken from the interview immediately preceding or concurrent with her risk set age, which allowed caregiver status to vary over time (ie, a participant contributed person-months as a caregiver until her age at the interview when she reported cessation of caregiving, at which age she began contributing person-months as a noncaregiver).

For analyses of concurrent effects, we modeled participants’ caregiving status and covariables at their current risk-set age. To account for the possibility of lagged effects of caregiving on mortality, since few participants were current caregivers at the time of their death, caregiving status was lagged 1, 3, and 5 years prior to the time represented by the risk set.

For both sets of models, individual covariables, including age, race, marital status, education, medical conditions, IADL limitations, and perceived stress were evaluated as potential confounders based on a priori assumptions of their association with caregiving status and mortality, and whether their addition to a proportional hazards model containing only caregiving status changed its association with mortality by 10% or greater.

For most summary measures (ie, IADL limitations, perceived stress scale), values were imputed using mean item substitution when less than 25% of items were missing. Otherwise, observations with missing data were dropped from analyses.

Results

The sample included 1,068 women who were followed an average of 8.2 person-years or 98.6 person-months. At the baseline Caregiver-SOF interview, the mean age of participants was 81.3 (+/− 3.7) years; 88% were white and 35% were caregivers. Adjusting for age, caregivers at baseline were more likely to be married, had fewer IADL limitations, but were more stressed (Table 1). Approximately half of the caregivers were taking care of a spouse or living with the care recipient. The most frequent reasons why the care recipient required care were dementia (27%), frailty/general health decline (22%), and stroke (21%), whereas other reasons ranged from Parkinson’s Disease or other neurological conditions, to recovery from fracture/surgery, to cancer, developmental disabilities, and psychological disorders. Respondents were helping their care recipient with an average of 1.5 ADL and 3.9 IADL tasks; 28% performed these tasks for ≥5h/d and 38% had been in this role for >5 years.

Table 1.

Baseline Age-Adjusted Characteristics of 1,068 Women Caregivers and Noncaregivers in Caregiver-Study of Osteoporotic Fractures Study

Characteristic	Noncaregivers	Caregivers
	n = 694	n = 374
	Mean (SD) or %	Mean (SD) or %	p-Value
Sociodemographic characteristics
Age (years)	81.5 (0.1)	81.0 (0.2)
Race: white	88.03	88.79	.70
Marital status: married	26.56	54.99	<.01
Education: college or higher	50.57	57.23	.04
Number of Medical conditions	1.78 (0.04)	1.76 (0.06)	.74
Number of ADL limitations	0.51 (0.03)	0.37 (0.04)	<.01
Number of IADL limitations	0.88 (0.04)	0.43 (0.06)	<.01
BMI (kg/m2)	27.39 (0.19)	27.11 (0.26)	.39
Perceived stress scale score	15.44 (0.28)	17.43 (0.38)	<.01
Caregiving characteristics
Caregiver to spouse		47.33
Lives with care recipient		50.27
Care recipient has dementia		28.25
Years spent caregiving before baseline*		6.82 (9.79)
<2 Years before baseline		22.19
2–5 Years before baseline		39.57
>5 Years before baseline		38.24

*Noncaregivers were not asked about prior caregiving experience.

A total of 308 (82%) of baseline caregivers stopped caregiving and 59 (9%) of baseline noncaregivers started caregiving over the follow-up period. This corresponded to 428 transitions over this period. Between 30% and 73% of caregivers at any one interview stopped caregiving before the next interview, whereas 2%–4% of noncaregivers started caregiving between contiguous interviews. Over half of those who stopped caregiving did so because the care recipient died; less frequent reasons were that the care recipient moved to a long-term care facility, moved to live with other caregivers, or his/her health improved, or the caregiver’s health declined.

Based on participants’ caregiving status at baseline, 145 caregivers (38.8%) and 338 (48.7%) noncaregivers died during follow-up. Additionally, 44 caregivers (12%) and 86 noncaregivers (12%) withdrew from the SOF or Caregiver-SOF studies. Baseline caregivers were less likely to die than noncaregivers: adjusted hazards ratio (aHR) = 0.77, 95% confidence interval 0.62–0.95 (Table 2).

Table 2.

Association Between Caregiver Status at Baseline and Mortality Over 13 Years, Using Time-in-Study as Timeline, Caregiver-Study of Osteoporotic Fractures Sample

		Hazard Ratio (95% confidence interval)
	n (%) Died	Unadjusted		Age-Adjusted		Fully Adjusted*
Noncaregiver	338 (48.7)	1.00		1.00		1.00
Caregiver	145 (38.8)	0.66	(0.54–0.78)	0.68	(0.56–0.82)	0.77	(0.62–0.95)
Age (in years)				1.12	(1.09–1.14)	1.09	(1.06–1.11)
Education: high school or less						0.99	(0.83–1.19)
Married						0.91	(0.75–1.12)
Race: black						0.50	(0.34–0.73)
Perceived stress scale score						1.01	(1.00–1.02)
IADL limitations						1.27	(1.18–1.37)

*Model-adjusted baseline covariates.

In analyses that used age as the analytic timeline and adjusted for time-varying covariates, baseline caregiver status was not associated with mortality (aHR = 1.02, 0.83–1.25) (Table 3). However, analyses of time-varying caregiving status resulted in a stronger reduction in mortality among current caregivers versus noncaregivers (aHR = 0.54, 0.38–0.75). The 1-, 3-, and 5-year lagged effects indicated that this inverse relationship became less pronounced when evaluating the impact of caregiving on mortality in the years prior to death (aHR = 0.60, 0.68, and 0.76, respectively).

Table 3.

Association Between Caregiver Status and Mortality Over 13 Years Using Age-Based Risk Sets, Baseline and Time-Varying Survival Models, Caregiver-Study of Osteoporotic Fractures Sample

		Hazard Ratio (95% confidence interval)
	n Died	Unadjusted		Adjusted*
Baseline caregiver status
Noncaregiver	338	1.00		1.00
Caregiver	145	0.82	(0.67–1.00)	1.02	(0.83–1.25)
Concurrent caregiver status
Noncaregiver	439	1.00		1.00
Caregiver	44	0.40	(0.29–0.54)	0.54	(0.38–0.75)
Lagged caregiver status†
Noncaregiver: 1-year lag	424	1.00		1.00
Caregiver: 1-year lag	59	0.47	(0.36–0.62)	0.60	(0.45–0.81)
Noncaregiver: 3-year lag	352	1.00		1.00
Caregiver: 3-year lag	73	0.55	(0.43–0.71)	0.68	(0.52–0.88)
Noncaregiver: 5-year lag	334	1.00		1.00
Caregiver: 5-year lag	99	0.63	(0.50–0.79)	0.76	(0.60–0.95)

*Models adjusted for time-varying education level, marital status, race, perceived stress score, and IADL limitations and adjusted for age through age-based risk-sets.

^†For lagged analyses, caregiver status was based on caregivers’ report of years caregiving before baseline interview, and noncaregivers not being caregivers during 2 years before baseline interview. 3- and 5-year lagged analyses had fewer deaths due to excluding persons missing this information.

Discussion

In this study of older women followed over 13 years, we observed a reduced risk of mortality among caregivers relative to noncaregivers, similar to prior reports from the Caregiver-SOF sample (10) and other studies (8,9,11). When we lagged the exposure to caregiving from 1- to 5-years, caregivers continued to have a reduced risk of mortality, but it was attenuated. These results reflect a Healthy Caregiver Hypothesis in that lower mortality rates would be expected in caregivers than noncaregivers, which is prominent when using current caregiving status. Moreover, our use of analytic methods from occupational epidemiology that are used to minimize the Healthy Worker Effect minimized the effects of better health associated with caregiving status. Specifically, these methods provided a more accurate assessment of caregiver status over the follow-up period by measuring it as a time-varying exposure and minimized potential confounding by age and changes in health through use of age-based risk set sampling. With these approaches, we observed reduced mortality associated with caregiving, ranging from an aHR of 0.54 for current caregivers to 0.76 for a 5-year lagged effect.

Our findings may be explained by several possible mechanisms. First, caregiving tasks may have kept caregivers physically and cognitively active. Indeed, persons who spent more time performing caregiving activities had lower risk of mortality in some studies (8) but not in others (3,28). Second, although caregivers in our study reported more perceived stress than noncaregivers, they may have experienced other psychological benefits from providing care that had salutary effects, resulting in reduced mortality. Third, participants in Caregiver-SOF were older than samples in studies that found poorer health outcomes in caregivers (3,4), and it might be that caregiving has fewer adverse health effects in very old adults than in younger adults. Finally, there may have been residual confounding by unmeasured factors related to the caregivers’ better health, but it is unlikely to have accounted for these associations.

The methods used in this study contribute to analytic advances in caregiving research. A recent study by Roth and colleagues used propensity-score matching to control for baseline differences in caregivers and noncaregivers in a sample of over 28,000 participants, and found lower mortality rates in caregivers (11). Propensity scores are used in occupational epidemiology to address differences in who is hired, corresponding to who becomes a caregiver, to create comparable groups in terms of who is exposed (29). In creating the Caregiver-SOF sample, we matched noncaregivers to caregivers on SOF site, age, race, and zip code (22). We opted not to use propensity scores for two reasons: First, we were concerned that if we matched noncaregivers to caregivers on propensity scores, we would not find enough exact matches due to the small size of the Caregiver-SOF sample; second, we did not find enough variability between caregivers and noncaregivers to warrant propensity score analysis, so we are confident that our sample was relatively homogeneous at baseline. Nonetheless, both our study and Roth and colleagues paid attention to applying analytic approaches to reduce confounding. These techniques should be applied in future studies to ensure rigorous and valid assessment of the impact of caregiving on health outcomes.

Our study had several potential limitations. The sample included a heterogeneous group of caregivers, but there were too few caregivers to persons with specific conditions, such as dementia, to compare mortality risk associated with caregiving for persons with specific conditions. We may have missed observing higher mortality risk associated with greater caregiving intensity, as would be suggested by stress theory, because we did not categorize caregivers according to caregiving intensity. However, this possibility is unlikely in that previous studies found less decline in physical functioning in caregivers who helped the care recipient with more I/ADLs (13) and lower mortality risk among caregivers who provided more hours/week of care (8). Finally, because the Caregiver-SOF sample was derived from the SOF sample, the participants were all older women and were mainly white. Thus, our results may not be generalizable to men, younger caregivers, or minorities. However, as older women comprise the majority of caregivers in the United States (30), our results are generalizable to most caregivers at the present time.

This study had several strengths. The Caregiver-SOF sample was drawn from a community-dwelling population. The same criteria (ie, assistance with I/ADL tasks) were used to identify both caregivers and noncaregivers. These features minimized potential biases that may result from recruiting caregivers from patient registries and organizations serving caregivers or families of persons with specific conditions, and recruiting noncaregivers from other sources. Our measure of caregiver status ensured that all caregivers were engaged in performing caregiving tasks. Further, caregiving status was reassessed at each follow-up interview which allowed us to document cessation in caregiving among participants who were caregivers at baseline, and initiation of caregiving among those who began as noncaregivers. Rather than classifying caregivers according to their level of caregiving-related strain (4,11), we measured perceived stress in all participants, which allowed us to control for the potential confounding effect of perceived stress on mortality.

In conclusion, this study applied several analytic approaches from occupational epidemiology to examine the association between caregiving and mortality in a sample of community-dwelling older women. Across all approaches, we found lower risk of mortality in caregivers than noncaregivers. These results add to the growing evidence that caregiving, per se, does not increase the risk of mortality in older adults. This is important information given the projected increase in the population of older caregivers in the United States (30). Nonetheless, it is known that the caregiving can be very stressful. Future studies that apply the techniques used in this and previous (11) studies are needed to further explore the health effects of caregiving-related stress, caregiving tasks, and psychological aspects of the caregiving role.

Funding

This work was supported by the National Institutes of Health, grant numbers R01 AG005407, R01 AR35582, R01 AR35583, R01 AR35584, R01 AG005394, R01 AG027574, R01 AG027576, R01 AG18037, and R01 AG028144.