Study population
The current study uses data from the SHARE project (Survey of Health, Ageing and Retirement in Europe), an European multidisciplinary and cross-national panel database of micro data on health, socio-economic status and social and family networks [24]. A probability sample of the target population, i.e. individuals aged 50+, was interviewed in the SHARE project [24]. .Nevertheless, people who were incarcerated, hospitalized or out of the country during the entire survey period, or who were unable to speak the country’s language(s) or had moved to an unknown address were excluded from the survey. The interviewers used computer-assisted personal interviewing (CAPI) to collect the data. Proxy interviews were allowed when respondents were unable to do an interview, for example, for health reasons. For more methodological details of the SHARE project, please see Börsch-Supan et al. [24].
Due the fact that Portugal only started participating in the SHARE project in wave 4 (2011) and did not participate in wave 5 (2013), the current study uses data from wave 4 (2011) and wave 6 (2015) (release 6.1.0.).
We restricted our sample to SHARE Portuguese respondents aged 50+ who participated in waves 4 and 6, and who did not have missing information for question “Is there someone living in this household whom you have helped regularly during the last twelve months with personal care, such as washing, getting out of bed, or dressing?” in both the waves analyzed (4 and 6) (N = 1262).
Measures
Outcome variables
Health was examined by two measures: physical health and depressive symptoms.
Physical health was assessed using a latent continuous measure. This variable was created according to procedures in Ploubidis and Grundy [25] and Di Gessa et al. [26] and includes one objective health indicator (maximum grip strength, using one or both hands) and six subjective ones. The subjective variables used are: self-perceived health (Would you say your health is …) using a 5-point ordinal scale (poor (1), fair (2), good (3), very good (4) or excellent (5)); the presence of long-term illness (Some people suffer from chronic or long-term health problems. By chronic or long-term we mean it has troubled you over a period of time or is likely to affect you over a period of time. Do you have any such health problems, illness, disability or infirmity?): coding 0 if yes and 1 if no; limited activities because of health (For the past 6 months at least, to what extent have you been limited in your activities because of a health problem): coding 1 for severely limited; 2 for limited, but not severely; and 3 for not limited; the doctor told that you have or had a heart attack: coding 0 for yes, and 1 for no; the doctor told that you have or had a stroke: coding 0 for yes, and 1 for no; and the doctor told that you have or had a chronic lung disease: coding 0 for yes, and 1 for no. This physical health measure was implemented in MPLUS, version 7, using WLSMV estimator (Muthén & Muthén, 1998–2012).
According to Ploubidis and Grundy [25], this measure is less subject to measurement error and has greater repeatability and reliability compared to individual health indicators used separately. In both waves (waves 4 and 6), our model revealed a good model fit: the Root Mean Square Error of Approximation (RMSEA) was 0.030 in wave 4 and 0.033 in wave 6 (values less than 0.06 indicate good fit); the Comparative Fit Index (CFI) was 0.986 in wave 4 and 0.983 wave 6 and the Tucker-Lewis Index (TLI) was 0.979 in wave 4 and 0.974 in wave 6 (for adequate models, both indices should have values above 0.95).
Depressive symptoms were assessed using the EURO-D 12-item scale that includes questions about feelings of depression, pessimism, wishing death, guilt, irritability, tearfulness, fatigue, sleeping troubles, loss of interest, loss of appetite, reduction in concentration, and loss of enjoyment over the last month [27]. Each question was scored one, if the feeling was present, or zero, if the feeling was not present, with a minimum possible score of zero and a maximum possible score of twelve.
We follow the Dewey and Prince [28] procedures and define clinically significant depressive symptoms as a EURO-D score greater than 3; and no clinically significant depressive symptoms as a EURO-D score equal or lower than 3. According to the same authors [28], this cutpoint was validated in the EURODEP study across the continent, and against a variety of clinically-relevant indicators. A EURO-D score greater than 3 would be likely to be diagnosed as suffering from a depressive disorder, for which therapeutic intervention would be indicated [28]. Psychometric evaluation of the 12 individual scale items for Portugal revealed good Cronbach’s Alpha in wave 4 (0.82) and acceptable in wave 6 (0.76) [22].
Independent variable
In the current study, informal care was defined as non-professional, unpaid support given to a family member, friend, neighbour or someone with another type of relationship living inside or outside their household who requires help with everyday tasks [14, 29]. The provision of informal care inside the household was analyzed by question: Is there someone living in this household whom you have helped regularly during the last 12 months with personal care, such as washing, getting out of bed, or dressing? In this question, SHARE considers that regularly means daily or almost daily informal care provided for at least 3 months. Taking this question into consideration, we defined as an informal caregiver inside the household all the Portuguese aged 50+ who responded affirmatively.
Covariates
Based on the literature, the current analysis included several control variables.
Age at the time of interview, sex (1 = female and 0 = male), marital status (1 = married and living together and 0 = all the other situations) and current job situation (1 for employed and 0 for all the other categories: retired, unemployed, permanently sick, homemaker and other). Education was coded according to the International Standard Classification of Education 1997 (ISCED-97). Respondents were grouped into the following categories: 1 as primary education (ISCED-97 score = 0–2), 2 as secondary education (ISCED-97 score = 3), and 3 as post-secondary education (ISCED-97 score = 4–6) [30]. Income was constructed using the variable total household net income (version A) that is obtained by a suitable aggregation at the household level of all individual income components. Income was adjusted for purchasing power parity and the square root of household size and divided into tertiles. The lowest tertile was coded as 1, the middle as 2, and the highest as 3. Cognitive function was constructed according to the procedures in Leist et al. [31]. The sum of five z-score measures was used: immediate recall (immediately recalling as many words as possible from a 10-word list that had been read out); delayed recall (recalling the ten-word list after a short delay); numeracy (assessed by five arithmetical subtraction tasks); Verbal fluency (naming as many animals as possible in 1 min) and orientation (score of orientation in time test). For the construction of this variable, we only consider individuals who had valid values for at least three of the tests.
To assess physical inactivity, SHARE respondents were asked how often they engage in vigorous activity (i.e., sport, heavy housework, or a job that requires physical labour) or moderate activity (i.e., activities requiring a low or moderate level of energy such as gardening, cleaning the car, or walking), with four response options: 1 - more than once a week; 2 - once a week; 3 - one to three times a month; 4 - hardly ever or never. In this study, we used the generated dummy variable that characterizes physically inactive individuals as those who have never practised vigorous or moderate physical activity.
Social network scale was a summary scale that combines five social network characteristics within a single index [30]. These characteristics include (1) the number of persons cited (network size); (2) the number of cited social network members living within 25 km (proximity); (3) the number of cited persons with weekly or more contact (frequency); (4) the number of cited persons with very or extremely close emotional ties (support); and (5) the number of different types of relationships present within the network (diversity). The first four measures were scored as follows: 0 = 0 SN members; 1 = 1 SN member; 2 = 2–3 SN members; 3 = 4–5 SN members; 4 = 6–7 SN members. The fifth measure calculated the number of different relationship categories (1- spouse; 2- other family, including children; 3- friend; and 4- other) present in the network. This last measure score ranged from 0 to 4, with a score of zero meaning no social network (i.e. no persons named) and the remainder reflecting the number of different relationship types, from 1 to 4. The total social network scale varies between 0 and 20, with 0 representing no named people in the social network and higher scores representing more social capital. This scale was divided into five levels (0 to 4), with the lower level (0) representing no social network (i.e. no people named), level 1 representing scores 1 to 5, level 2 scores 6 to 10, level 3 scores 11 to 15 and level 4 scores 16 to 20. Psychometric evaluation of the five individual scale items for Portugal revealed high Cronbach’s Alpha in both waves analyzed (0.93 in wave 4 and 0.81 in wave 6).
Finally, Quality of Life (QoL) was assessed using the CASP-12 scale, the short version of CASP-19 [32], which comprises four dimensions: Control, Autonomy, Self-realization and Pleasure. The total number of points on the CASP-12 scale varies between 12 and 48 points, with a greater QoL corresponding to higher values. Cronbach’s Alpha in wave 4 was 0.83 and in wave 6 it was 0.50.
Statistical analysis
Firstly, we assessed the prevalence of informal caregiving inside the household provided by individuals aged 50+, comparing Portugal with other SHARE countries. Secondly, we compared baseline characteristics (wave 4, 2011) of Portuguese individuals aged 50+ who provided informal care inside the household (co-residential caregivers) with the characteristics of those who do not provide informal care inside the household (non-caregivers). Statistical tests (chi-square test and T test) for two-group comparison were applied. Statistical test results with p < 0.05 were considered to be significant, and with p < 0.10 were considered marginally significant. Thirdly, a longitudinal linear mixed model with fixed effects, random effects and an error was used to analyze the impact of providing informal care inside the household on the physical health of Portuguese individuals aged 50+. The model was fitting by maximum likelihood, using the following equation: Yij = Cijβ + Xijδ + Ui + Zij, i = 1, …,n;j = 1,2 , where, Yij denotes the dependent variable – physical health – for the individual i at time j, Cij is a dummy for caregiver inside the household or not, Xij is a vector that includes all control variables, the Ui is the individual random effect, Ui~N(0, ν2), and Zij is the measure error, Zij~N(0, τ2), with Ui and Zij being independents. Thus, β gives the effect of providing care inside the household (vs. not providing care inside the household) for the individual i [33]. Lastly, a longitudinal generalized mixed model with a logit link function was applied to analyze the impact of providing care inside the household on the depressive symptoms of Portuguese individuals aged 50+. To this end, we consider constant correlation and the equation.
$$ \boldsymbol{\ln}\left(\frac{{\boldsymbol{\pi}}_{\boldsymbol{i}\boldsymbol{j}}}{\mathbf{1}-{\boldsymbol{\pi}}_{\boldsymbol{i}\boldsymbol{j}}}\right)={\boldsymbol{C}}_{\boldsymbol{i}\boldsymbol{j}}\boldsymbol{\beta} +{\boldsymbol{X}}_{\boldsymbol{i}\boldsymbol{j}}\boldsymbol{\delta} +{\boldsymbol{U}}_{\boldsymbol{i}}+{\boldsymbol{Z}}_{\boldsymbol{i}\boldsymbol{j}},\kern0.5em \boldsymbol{i}=\mathbf{1},\dots, \boldsymbol{n};\boldsymbol{j}=\mathbf{1},\mathbf{2}, $$
where πij denotes the probability of success (being depressed) for the individual i at time j (P(Yij = 1)), and the \( \left(\frac{{\boldsymbol{\pi}}_{\boldsymbol{ij}}}{\mathbf{1}-{\boldsymbol{\pi}}_{\boldsymbol{ij}}}\right) \) is called the odd ratio (OR). The Cij is a dummy variable for providing care inside the household (vs. not providing care inside the household), Xij is a vector that includes all control variables, the Ui is the individual random effect, Ui~N(0, ν2), and Zij is the measurement error, Zij~N(0, τ2), being Ui and Zij independents. In this way, exp(β) gives the likelihood of a caregiver inside the household being depressed over a non-caregiver for the individual i. All analyses were performed using software R, version 3.4.3.