1
Introduction
1.1
Background
It is estimated that over 6 million people provide informal unpaid care in the UK,
a number expected to rise to 9 million by 2037 (Maher and Green, 2002; Carers UK,
2010). Informal caregivers have been described as an ‘invisible healthcare system’
(Arno et al., 1999). Caring for a relative with an illness or disability may result
in chronic stress (Schulz and Sherwood, 2008). Research suggests that carers are more
vulnerable to both physical (Da Roza Davis and Cowen, 2001; Pinquart and Sorenson,
2007; Perlick et al., 2005) and mental health problems (Gysels et al., 2012; Phillips
et al., 2009; Pinquart and Sorenson, 2003; Hirst, 2005), including depression and
anxiety (Papastavrou et al., 2012; Cooper et al., 2007; Damjanovic et al., 2007).
1.2
Impact of care
The adverse consequences of the caregiver role can be observed across several areas,
including disturbances in sleep (Phillips et al., 2009), immunological responses (Damjanovic
et al., 2007; Kiecolt-Glaser et al., 1991; Clark et al., 2013), endocrine functioning
(Kiecolt-Glaser and Glaser, 1999; Kring et al., 2010), and elevated rates of mortality
(Schulz and Beach, 1999). There is evidence to suggest that former carers are also
in worse health than non-carers (Lee and Gramotnev, 2007). The negative health sequelae
of the caregiving role (e.g. emotional distress) are frequently enduring (Vedhara
et al., 2000). The reduction of caregiver burden and improvement in carer wellbeing
thus remains a major responsibility for government and public health bodies (Department
of Health, 2010; Carers UK, 2010).
1.3
Models of caregiving
The stress-process model (Pearlin et al., 1990) and its variants (e.g. Szmukler et al.,
1996; Mackay and Pakenham, 2012) have been used to examine the mechanisms whereby
caregiving may lead to poor mental health. In this model, factors directly relating
to the nature and magnitude of the difficulty of the caring role are labelled as primary
stressors (Pearlin et al., 1990; Schulz and Sherwood, 2008) (see Fig. 1, developed
following Hilgeman et al. (2009) and Judge et al. (2009)). These can, for example,
include dealing with difficult behaviours, cognitive impairment and aggression from
the care recipient (Schrag et al., 2006), or with difficult emotions and relational
deprivation (Winn et al., 2007; Gilhooly and Whittick, 1989). Caregiving occurs within
a social context and background contextual factors such as socioeconomic status, education,
gender and ethnicity of the carer can have an important influence on the impact of
caregiving on health. The model is also designed to account for the effect of secondary
stressors on carer mental health. Secondary stressors include role strains such as
strains within relationships with other family members, interference with the carer's
occupation, financial strains and restriction of social life due to caring. Intrapsychic
strains, such as impacts of caring on dimensions of self-concept such as self-esteem,
mastery, role captivity (trapped unwillingly in the carer role), loss of sense of
self and competence in caring are also types of secondary stressors. More positively,
caring may lead to feelings of personal gain associated with successful achievement
of caring which also comes under the heading of secondary stressors. We were unable
to measure intrapsychic strains and the detail of strains within family relationships
in this study. A significant proportion of mental ill-health in carers may be attributable
to stressors associated only indirectly with their caring duties, such as stressful
life events, economic strains, job-caregiving conflicts and constriction of social
and recreational life (Carers UK, 2007, 2008). For instance, there is evidence that
carers often experience other stressors including social and economic disadvantage
(Hirst, 2004a), difficulties with finances and employment (Yeandle et al., 2007; Schulz
and Martire, 2004; Mangalore and Knapp, 2007) and housing (Gilleard, 1998), and a
reduced socioeconomic status (SES) relative to non-caregivers (Pinquart and Sorensen,
2003; Schulz and Sherwood, 2008). The caregiver role often restricts opportunities
to engage in social and recreational activities, and this may increase the risk for
depression (Neiboer et al., 1998). Thus, caregivers often have smaller social networks
and reduced levels of social support (Magliano et al., 2006; McManus et al., 2009),
with consequent negative effects on their mental health (Singleton, 2002; Bromley
et al., 2004). Thus mental ill-health in carers may be attributable partly to these
secondary stressors rather than only the primary stressors involved in aspects of
direct caring. The third aspect of the stress process model is access to mediators
such as social support. If available, social support can buffer or mediate the strain
of the caregiving role and reduce the risk of mental disorder (Pearlin et al., 1990).
1.4
Carer studies
Understanding the precise impact of the caregiving role on carer health outcomes should
be an important target for service providers. However, methodological problems such
as varying definitions of caring, small self-selected samples, the restriction of
samples to those caring for people with specific disabilities (e.g. Awad and Voruganti,
2008; Haley et al., 2010), a lack of non-carer comparison groups and adjustment for
socioeconomic position, make it difficult to draw firm conclusions about carer health
and lifestyle (Pinquart and Sorenson, 2003; Neugaard et al., 2008; Butterworth et al.,
2010). While it is certain that mental health problems in carers will reflect a ‘multitude
of stressors’ (Hirst, 2004a), few studies have used multivariate analyses to isolate
pathways, whereby the contribution of the caregiving role to common mental disorders
(CMD) relative to other known risk factors can be determined (Noh and Turner, 1987;
Hirst, 2005; Holmbeck, 1997). This quantifies the association of a wide range of secondary
stressors, related to caring, to mental ill-health in a nationally representative
sample and seeks to understand how much of the association of primary stressors of
caring and mental ill-health can be explained by secondary stressors and the effects
of social support as a mediator on the association of caring and mental ill-health
(see Fig. 1). We know that caregivers have a higher prevalence of CMD than non-carers
(Smith et al., 2014). We therefore hypothesized, first, that carers will report more
stressors and lower levels of social support than non-carers, and, secondly, that
the association between caregiving and CMD will be partially explained by the higher
levels of secondary stressors and lower levels of social support in carers.
2
Methods
The Adult Psychiatric Morbidity Survey (APMS) 2007 is a nationally representative
random probability survey containing questions on mental and physical health, lifestyle
and socio-demographic factors. Households in England were randomly selected using
the Small Users Postcode Address File, and one eligible person per household was selected
using the Kish grid method (Kish, 1965). 13,171 households were visited, and 57% (7461)
of the selected participants agreed to take part (70% of those who could be contacted).
The APMS interview consisted of a 90 min computer-assisted personal interview conducted
by experienced interviewers (McManus et al., 2009).
2.1
Measures
2.1.1
Primary stressors
Carers were identified by a question taken from the 2001 England & Wales census (ONS,
2001), asking whether respondents currently “looked after, gave help or support to
family, friends, neighbours or others because they have long term physical or mental
ill health or disability.”
2.1.2
Secondary stressors
Stressful life events in the last six months were assessed using the List of Threatening
Experiences (Brugha et al., 1985). We constructed measures reflecting the experience
of any personal life event and any material life event in the last six months. Illness
in a close family member was excluded from the list as potentially being a confounder
in the association of caring and common mental disorder. Debt was measured by asking
respondents if they had been seriously behind paying for bills and loans in the last
12 months. Ten questions adapted from the Domestic Violence and Abuse module of the
British Crime Survey (Walby and Allen, 2004) elicited experiences of non-sexual abuse
by a current or previous partner in last 12 months. A dichotomous variable of any
domestic violence was derived.
2.1.3
Mediators
Perceived social support was assessed from 7 questions taken from the Health and Lifestyle
survey (Cox et al., 1987). Respondents were asked if they thought that family or friends
were available for support. Scores were summed and categorized with a maximum score
(21) indicating good support, 18–20 moderate support, and <18 low support (Brugha
et al., 2003). Social participation, included visits to evening classes, leisure centres,
or libraries and active involvement in any recreational or social club, attendance
at a local or voluntary community group (at least once month; at least once year;
not in last year/never).
2.1.4
Common mental disorders and other health outcomes
Common mental disorders (CMD) were assessed using the Revised Clinical Interview Schedule
(CIS-R) (Lewis et al., 1992), a standardised clinical psychiatric interview providing
six ICD-10 diagnoses, based on symptoms in the past week. A dichotomous measure based
on the presence of any CMD, involving all 6 ICD codes, was derived. CIS-R questions
were also used to assess suicidal ideation: whether respondents, in the last week
or year, or at any other time, had wished they were dead, had thoughts about suicide
without intentions to act, or had attempted to take their life.
Alcohol misuse was measured with the AUDIT scale. A score of ≥8 was used to define
a drinking problem (Saunders et al., 1993). Questions on drug use were based on questions
from the Diagnostic Interview Schedule (Malgady et al., 1992). Drug use was here defined
as acknowledged consumption of any of 14 illicit named drugs in the last 12 months
(McManus et al., 2009).
A score of 2 or more from the CIS-R fatigue subscale was used to indicate high levels
of fatigue (Lewis et al., 1992). Problems with activities of daily living (ADL) (Martin
et al., 1988) were assessed by 7 questions asking about difficulties with activities
such as bathing, dressing, taking pills, gardening and decorating. Participants were
divided into those who had and who did not have problems with any ADL.
2.1.5
Sociodemographic background factors
Sociodemographic details including age, gender, highest education level, equivalised
household income, employment status, Registrar General's Social Class (Standard Occupational
Classifications ONS 2000), marital status, number of dependent children in household,
and housing tenure were also recorded (McManus et al., 2009).
2.2
Statistical analysis
Analysis was conducted using the survey design functions in STATA v13. to account
for weighting procedures. Weights were applied to represent the structure of the national
population and to account for the probability of selection and non-response (McManus
et al., 2009).
All analyses were adjusted for age, gender and socio-economic status (including equivalised
household income, education, and occupational class). Risk factors and outcomes were
largely treated as dichotomous variables in order to examine the magnitude of effect
of multiple variables using odds ratios. Analyses were completed on the weighted complete
record subset (N = 5170) created when all covariates are completely observed from
the fully adjusted multivariate logistic regression model.
Step 1 involved confirming that the predictor of interest (Caring) was significantly
related to the outcome (CMD). The associations of caring with CMD, with individual
mental disorders and with fatigue and suicidal ideation were examined using logistic
regression.
Step 2 established whether the predictor (caring) was related to the potential mediating
factors (stressors and social supports). The associations between caring, stressors
and social support were examined using logistic regression models.
Step 3. Those stressor and social support variables satisfying step 2 were added to
a series of multivariate logistic regression models to test whether or not the relationship
between caring and CMD diminished as a result of their inclusion.
Step 4. The mediation effect of personal life events, material life events and perceived
social support between CMD and caring (adjusting/controlling for the other covariates
in model 4) was investigated separately. As the putative mediators and the outcome
are binary variables we followed the method in MacKinnon and Dwyer (1993) to find
the coefficients of the mediator paths and their corresponding standard errors. These
were then included in three mediation tests, the Sobel, Arorian and Goodman Test.
The Sobel assumes the path between caring to the mediator and the path between the
mediator to CMD are independent. The Arorian does not assume this and adds the covariance
term in the denominator, whereas the Goodman test subtracts it to create an unbiased
estimate of the variance. We apply all three to examine how robust inferences are
to the chosen method. We put more emphasis on the Arorian test for our conclusions
as it has been shown to perform better in a Monte Carlo study (MacKinnon et al., 1995).
All three tests make the assumption that the distribution of the product of the regression
coefficient is normal; however this is unlikely to be true, it is generally asymmetric
with high kurtosis. We thus apply an adjustment to the Arorian test as described by
MacKinnon et al. (2002) (first variant) and use the table of critical values to assess
significances.
3
Results
3.1
Caring in the general population
Just over a quarter (weighted N 1367) of the population reported caring duties. Women
were more likely to be carers than men (OR = 1.20 95% CI 1.04–1.39), and carers were
most likely aged between 55 and 64 years (OR = 2.71 95% CI 1.86–3.94: reference group,
16–24 years). People in the lowest income group (<£10,575) (OR = 1.31 95% CI 1.07–1.61)
and those in the second lowest income groups (>=10,575 to <16,195) (OR = 1.52 95%
CI 1.24–1.86) were significantly more likely to have caring duties than those in the
highest income group. People working in personal service occupations were more likely
to be carers (OR = 1.46 95% CI 1.10–1.94: reference group, managers and senior officials) (see
Table 1).
3.2
Associations between caring and CMD
Carers had an increased risk of having a CMD (OR = 1.64 95% CI 1.37–1.97) adjusting
for age, gender, standardised occupational classification, equivalised household income
and education. After adjustment for age, gender and SES, carers had increased rates
of Generalized Anxiety Disorder (OR = 1.57 95% CI 1.19–2.08), Mixed Anxiety and Depressive
Disorder (OR = 1.45 95% CI 1.13–1.87) and Panic Disorder (OR = 2.38 95% CI 1.43–3.96)
compared with non-carers (Table 1). Carers were twice as likely to report thinking
about suicide in the last week (OR = 2.71 95% CI 1.31–5.62) and more likely to report
wishing they were dead in the last week (OR = 2.10 95% CI 1.15–3.82).
3.3
Associations between caring, stressors and social support
Table 2 describes the odds of caring in relation to household tenure and various individual
measures of stressors and social support. More carers reported domestic violence in
the last 12 months (OR = 1.57 95% CI 1.14–2.17), and they were more likely to report
debts (OR 1.60 95% CI 1.20–2.13) in the last year than non-carers (Table 2).
Carers were less likely to be divorced/separated/widowed (OR = 0.78 95% CI 0.65–0.92)
than non-carers. Carers and non-carers did not differ significantly in visits to evening
classes, leisure centres, or libraries, but carers were more likely to participate
in a voluntary or local community group in the last year (OR = 1.36 95% CI 1.07–1.72)
and were more likely to belong to a recreational, social or community club (OR = 1.23
95% CI 1.06–1.42). More carers reported suffering high levels of fatigue (OR = 1.33
95% CI 1.14–1.54) than non-carers. Self-reported general health did not differ significantly
between groups.
3.4
Multivariate models of caring and CMD
Four multivariate models tested the relationship between caring and CMD after stressors
and social support were added to the models (Table 3). Model 1 demonstrated an association
after adjustment for basic socio-demographic factors.
In model 2 the addition of personal and material life events, debts, alcohol problems
and domestic violence diminished the relationship between caring and CMD, (OR = 1.52
95% CI 1.26–1.83). Domestic violence, a personal life event in the last six months,
debt, income less than £10,575 and no educational qualifications were significant
independent predictors of CMD.
The addition of informal social support (perceived social support and marital status)
in model 3 did not further reduce the association between caring and CMD (OR = 1.55
95% CI 1.29–1.87). Perceived social support was a significant independent risk factor
for CMD, with those reporting low support at a 2.03 (95% CI 1.55–2.66) increased odds
of CMD. In model 4 where community group participation and club membership were added
to the model, there was again no substantial change in the association between caring
and CMD (OR = 1.58 95% CI 1.30–1.91). Being a member of a club was associated with
a 0.73 (95% CI 0.61–0.88) reduction in odds of CMD. Debt, a personal life event, domestic
violence, marital status, and moderate/low perceived support also remained independently
associated with CMD in the final model.
In Table 4 we report whether personal life events, material life events or perceived
social support mediated the association of caregiving and CMD. There was no evidence
to support mediation tested by Sobel, Aroian or Goodman tests. The data did not appear
to be sensitive to the choice of mediating test. We applied the MacKinnon et al. (2002)
(first variant) method to the Aroian test statistic by using the table to find the
proportions under the curve beyond the test value we obtained and doubling it for
a two tail test. The method is only approximate as our sample size is larger than
the table shows. The method supports our previous findings that personal life events,
material life events and perceived social support do not have a significant mediating
effect between caring and CMD separately.
4
Discussion
The current findings confirm that caregiving remains an independent risk factor for
CMD in a model including stressors and social support. The odds of CMD were 58% higher
in carers than in non-carers. The effect of caring on CMD is comparable to the effects
of living in a low income household, or leaving school without qualifications. Carers
were also more likely to report difficulties with debt and domestic violence.
In multivariate models, the inclusion of stressors had the largest impact on the relationship
between caring and CMD, but only reduced the odds of CMD in carers from 1.62 to 1.52.
The addition of social support variables did not reduce the odds of CMD in carers
further. Stressors and social support variables included in the models showed independent
effects on CMD after adjustment for all other variables.
4.1
The stress process
It is likely that some of the stressors reported by carers are a direct result of
the caring role. Elevated rates of trauma have been reported in carer populations
(Noble and Schenk, 2008). In severe mental health disorders, carers are also more
likely to be the targets of aggressive behaviour from care recipients (Nielssen et al.,
2007; Nordstrom and Kullgren, 2003). Further, approximately 4% of the current sample
acknowledged suicidal thoughts in the preceding year. Suicidal ideation has been reported
in other carer populations such as dementia carers, albeit at higher rates (O'Dwyer
et al., 2013). However, since details about the source of some of the reported stressors
were not given in the current survey, it is difficult to establish whether they were
a direct result of caring duties. The caring role is likely to be stressful at least
in part because of the accumulation of a variety of small ties and obligations, which
would not be captured by our measures. There does appear to be a dose effect: caring
has been shown to involve a greater risk of mental disorder in those carrying out
heavy caring duties and spending more time caring per week, and carrying out caring
at home (Hirst, 2004a). The physiological response associated with recurrent exposure
to stressors and the stress proliferation involved in caregiving may lead to allostatic
processes attempting to compensate for the effects of stress that increase the risk
of physical ill-health and mortality rates in caregivers (McEwen and Seeman, 1999;
Pearlin, 2010; Schulz and Beach, 1999) although this could also be mediated through
CMD.
The cross-sectional nature of our analyses also constrains inferences about causality.
However, whether a consequence of caring or a confounding factor, several stressors
are more frequent in people who have taken up a caring role than in non-carers.
4.2
Comparison to other studies
The risk of CMD found in carers in this study is comparable to other general population
surveys that include a representative sample of carers (Singleton, 2002; Maher and
Green, 2002; Hirst, 2004b). Hirst (2004a) also found a similar association between
caring and mental health.
Panic and anxiety disorders were most prevalent amongst carers in the current study,
as noted by others (Vanderwerker et al., 2005; Cooper et al., 2007). Fatigue has also
been widely reported in carers, which may be an indirect reflection of the degree
of primary stressors but may also be seen as an intermediate outcome on the pathway
to CMD (Singleton, 2002; Maher and Green, 2002).
It is not very surprising that carers reported stressful life events. The finding
that carers report more debt problems is consistent with other studies highlighting
financial difficulties in caregiver groups (Carers UK, 2007, 2008; Mangalore and Knapp,
2007). Our finding that debt problems might contribute to the relation between caring
and CMD is in line with studies linking debt with CMDs (Meltzer et al., 2013), and
with theories suggesting vulnerability to CMD in people of lower socioeconomic status
(Lorant et al., 2002). It is suggested elsewhere that financial difficulties increase
with time caring, possibly partly because caring may interfere with paid work (Arksey
and Hirst, 2005).
Reduced levels of perceived support and social participation with a club membership
did not account for the higher prevalence of CMD in carers in multivariate models.
This is encouraging evidence that carers from this sample were not subject to significant
restrictions in their social and recreational activities.
4.3
Methodological issues
This study corrects for some previous methodological problems. Bias in sample selection
was reduced by using a representative sample, including anyone that defined themselves
as a carer. Earlier studies have used comparable definitions, encompassing all carers,
but have reported smaller proportions of carers (Horsley et al., 1998; Green and Maher,
2002; Singleton, 2002). As our definition of caring does not seem more inclusive this
might reflect a proportionate increase in carers (Smith et al., 2014). Few epidemiological
studies of carers have included the range of secondary stressors measured in this
study (Butterworth et al., 2010; Pinquart and Sorensen, 2003; Neugaard et al., 2008).
However, in this study we did not have all the variables we needed for the stress
process model. We did not have several role strain variables such as precise indicators
of difficulties in relationships with other family members in relation to caring,
and we did not have measurements of intrapsychic strains-this may have weakened our
ability to test the full effects of secondary stressors.
Reverse causation may explain some of our findings, e.g. the higher prevalence of
CMD in carers may lead to a tendency to report more stressful life events (Kessler,
1997) or lower levels of social support (Sarason et al., 1991). Longitudinal data
would assist in the attribution of causation. Our study was limited in that it lacked
information on locus of care or type of care and its statistical power was insufficient
to investigate subgroups of carers. The stressors and supports for those caregiving
to people may differ by disease cared for and by age and relationship (e.g. spouse,
child or parent) (Awad and Voruganti, 2008; Papastavrou et al., 2012; Zegwaard et al.,
2013). Finally, our study might have missed those with severe mental distress or more
time consuming caring duties as they are less likely to take part in a general population
survey.
There has been a call for developing comprehensive models of mental disorder in caregiving
(Pinquart and Sorenson, 2003; Neugaard et al., 2008). Astrength of our study is the
incorporation of an extended range of stressors in a single model. Our analyses were
shaped by the stress process model (e.g. Pearlin et al., 1990), and examined possible
mediating pathways in caregiver CMD. We mainly focused on stressors indirectly associated
with caring (secondary stressors and contextual factors) but our findings leave much
of the relationship between caring and CMD to be explained. Primary stressors directly
related to the caring relationship are likely to explain more of the caring and CMD
relationship, and require to be further investigated alongside indirect factors. Personality
characteristics (e.g. Lautenschlager et al., 2013; Hooley and Hiller, 2000) and coping
styles (e.g. Papastavrou et al., 2012; Onwumere et al., 2011) of the caregiver would
also be useful to include in future surveys (Schrag et al., 2006; Pinquart and Sorensen,
2003). In some studies caregiving is associated with positive benefits to mental health
and increased mastery and sense of achievement (Pearlin et al., 1990). We were not
able to assess these aspects of caregiving which are important because they emphasise
that caregiving is not invariably a negative experience.
5
Conclusion
Caring is moderately associated with CMD, an association that is not explained by
social support and stressors. All carers in the English population should be recognized
as being potentially at increased risk of CMD, although for some caregiving has positive
effects on mental health. This study provided evidence that carers are also exposed
to high levels of stressors such as debt problems, domestic violence, and stressful
life events which are associated with increased risk of CMD in both carers and non-carers;
these secondary stressors did not explain the association of caring and CMD. It was
not clear in this cross-sectional study whether the stressors examined were caused
by caring duties but this study shows that there is inequity between carers and non-carers
that may lead to mental health problems. Carers may experience a multitude of risk
factors for CMD which have an additive effect. To ensure optimal health outcomes for
care recipients, whose functioning and community tenure may be dependent upon the
care provided by an informal carer, it would seem important for health professionals
to be aware of how additional sources of stress might add to the existing stress of
caring. Strategies are needed to assess carer stressors in the wider life roles and
the context in which they provide care. Further research that makes explicit the process
by which the caregiving role can lead to CMD is likely to aid the development of specialist,
evidence based carer interventions that can apply to a broad range of mental and physical
conditions.