Self-reported burden of caregiver of adults with depression: a cross-sectional study in five Western European countries

The data for this retrospective cross-sectional study were retrieved from the 2016 National Health and Wellness Survey (NHWS), a nationally representative, cross-sectional, internet-based general health survey of the adult population aged ≥18 years (EUR5 N = 80,600; France N = 19,500; Germany N = 19,500; Italy N = 13,000; Spain N = 9100; UK N = 19,500). A targeted sampling was used to ensure that the NHWS sample mirrors the demographic distribution with respect to gender and age in each of the EUR5 countries (France, Germany, Italy, Spain and the United Kingdom). In most cases, the survey was a self-administered structured questionnaire of close-ended questions. To further ensure a representative sample, particularly in the 65+ year-old population, online recruitment was supplemented by computer-assisted web interviews. Respondents who were recruited by telephone had the choice to complete the interview on the phone while the interviewer entered the responses online or were e-mailed a link to the survey to complete on their own. All respondents were aged 18 or older and could read and write in the primary language of the country at the time of the survey and provided informed consent. Survey participation rates was 10.5%. In 2016, a probability sampling was used to select a subsample of respondents to complete each specific module that enabled inclusion of respondents with different medical conditions to provide detailed information while limiting the average interview length and respondents’ burden. In the current study, additional variables were identified from the caregiver module (i.e., caregiver-specific variables) and from the symptom module (i.e., depression screening scale using the Patient Health Questionnaire [PHQ-9]). The study was conducted in accordance with International Society for Pharmacoepidemiology Guidelines for Good Pharmacoepidemiology Practices and applicable regulatory requirements. The 2016 NHWS protocol and questionnaire were reviewed by the Pearl Institutional Review Board (Indianapolis, IN, USA) and met the exemption requirements under 45CFR46.101(b)(2) (16-KAN-123).

Caregivers were identified using the question “Are you currently caring for an adult relative with any of the following conditions (e.g., depression, cancer, etc.)?” Based on their responses, survey participants were stratified into three groups: CG-UD, to be compared to CG-OD and non-CG respondents. Participants categorized as CG-UD (n = 1380) included those who were currently caring for an adult relative with unipolar depression. Excluded were those caring for an adult relative with bipolar disease or schizophrenia, as these patients have disorders that may experience depressive episodes, however their psychopathology differs. Participants categorized as CG-OD (n = 6470) included those who were currently caring for an adult relative with any of the following mental or physical chronic conditions: Alzheimer’s disease, bipolar disorder, cancer, chronic kidney disease on dialysis, chronic obstructive pulmonary disease, dementia, diabetes (type 1), epilepsy, heart disease, immune thrombocytopenic purpura, macular degeneration, multiple sclerosis, muscular dystrophy, osteoarthritis, Parkinson’s disease, stroke and/or schizophrenia. Non-CG (n = 69,334) included respondents currently not caring for an adult and excluded the participants (n = 3416) who declined to answer or were caregivers of any other conditions not included in the CG-UD or CG-OD groups (Fig. 1, Table S1).

All measures were collected for the study participant i.e., the caregiver and not the patient they are caring for.

The study examined characteristics and outcomes of CG-UD compared to CG-OD and non-CG. Descriptive statistics were reported for all study variables, including means, standard deviations or medians for continuous variables and frequencies and percentages for categorical variables. Bivariate analyses were performed to provide an understanding of the survey data prior to multivariable analysis and to examine the associations between potential covariates (e.g., demographics) and dependent study variables (e.g., HRQoL). For continuous variables, t-tests or the Mann-Whitney U-tests were used to determine the significant differences between groups, whereas chi-square tests were used for categorical variables. The results were fed into a more robust multivariable analysis.

For multivariable analysis, generalized linear mixed models (GLMMs) and linear mixed models with gaussian distribution, were used to determine the association of the population group with health outcomes (HRQoL and CRA), after controlling for confounders evaluated in the bivariate results including age, sex, marital status, employment, number of children in household, alcohol use, BMI, education, smoking status, exercise in past 30 days, and CCI. Respondents with responses of ‘decline to answer’ or with missing data were excluded from this analysis. Since skewness was observed on the distributions of both WPAI and HRU, GLMMs with negative binomial distributions were used. GLMM was used to account for clustering within country. Estimated mean values, standard errors, 95% CIs, and P-values were calculated for each dependent variable. Effect size were calculated using Cohen’s d approximation adjusted for use of multilevel modeling [42]. A p-value < 0.05 was considered statistically significant. In order to examine the potential bias of using (1) a heterogenous caregiver comparison group and (2) including other serious mental illnesses (bipolar disorder and schizophrenia) in the CG-OD group, we performed further sub-analyses. First, we reported the health status of caregivers for each category in the ‘other’ group showing the range of outcome values independent of disease type (Table S1). Next, we re-analyzed several outcomes (HRQoL [MCS and PCS], health status (EQ-5D-5L) and caregiver-specific measures) in the adjusted models while excluding caregivers of adults with bipolar disorder and schizophrenia (n = 5944) (data not shown). Regression models did not differ with regards to statistical differences of parameter estimates. Marginal mean estimates for the CG-OD excluding bipolar disorder and schizophrenia compared to the CG-OD group for HRQoL differed by + 0.3 for the MCS and + 0.2 points for the PCS. No differences in marginal mean estimates were noted between the CG-OD with and without bipolar disorder and schizophrenia for the EQ-5D-5L or any of the caregiver outcomes. These results suggest that the inclusion of these two serious mental illnesses in the CG-OD group did not bias the results.

All statistical analyses were conducted using R 3.5.2.

Of the total 77,184 EU study respondents, 1380 were identified as CG-UD (1.8%), 6470 as CG-OD (8.4%), and 69,334 (89.9%) as non-CG. The baseline demographic and clinical characteristics of CG-UD group compared with CG-OD and non-CG groups are shown in Table 1.

Bivariate analysis showed that CG-UD compared with CG-OD and non-CG were more likely to be female (62% vs 58 and 55%, respectively), younger (43.59 years vs 48.11 and 48.47 years, respectively) and employed (60% vs 56 and 55%, respectively). CG-UD respondents when compared with non-CG group were more likely to report greater comorbidities as well as sleep problems (p < 0.001). CG-UD group also reported higher anxiety, depression and moderate to severe PHQ-9 scores (for all, p < 0.001) when compared to both CG-OD and non-CG.

For outcome measures, bivariate analysis showed significant differences between CG-UD and CG-OD as well as non-CG group. CG-UD respondents reported significantly lower health status, SF-6D health utilities and EQ-5D-5L scores, when compared with CG-OD and non-CG group (all p values < 0.001) (Table S2).

Significantly increased absenteeism, presenteeism, and higher overall work and activity impairment was reported by CG-UD group compared with CG-OD and non-CG respondents (all p < 0.001) (Table S2). Further, CG-UD respondents reported higher HRU in terms of visits to the HCP and ER and hospitalizations compared with CG-OD as well as non-CG (all p < 0.001) (Table S2).

There were no significant differences in caregiver involvement in bathing or grooming, logistics, decision-making, or managing finances between CG-UD and CG-OD groups (Table S3). Differences were observed for the CRA for esteem, family support, and impact on finances and schedule.

The study strengths are specifically related to the self-reported outcomes of caregivers and notably includes both female and male caregivers. It is important, however, to consider the implications of these results in line with study limitations. The NHWS is a panel-based survey and although care is taken to have the panel mirror the population as closely as possible and age and gender are controlled in the NHWS sampling, sampling bias may still occur. For example, the use of an internet-based survey may bias recruitment toward a younger and/or healthier caregiver population. Data from the NHWS are self-reported and participant responses may reflect recall biases and other forms of measurement error. However, the survey represents a low-stakes event and does not present any incentive to purposely misrepresent one’s responses. In addition, missing data is minimized by the programming methods and providing ‘don’t know’ or ‘decline to answer’ as options. Also, given the cross-sectional design of the study, statements of causality cannot be made from the results and temporal trends in the relationships between study variables cannot be ascertained. Limitations in data analysis arise due to a finite set of included measured variables, which were accounted for in regression, yet there is the possibility of groups differing on unmeasured variables such as disease severity, duration of disease, and treatment status, as well as country-level differences, which may impact these outcomes [22]. The study was not designed to confirm self-reported patient symptoms, treatment history, diagnosis of depression, and other relevant diagnoses (e.g., those used in the comorbidity index). Other parameters like the number of caregiving hours, intensity of caregiving as well as severity of depression in the patient, nature of employment of the caregivers, etc. were also not explored in the study. Finally, the comparison group of caregivers in this study comprised a diverse mix of diseases and conditions, which vary, for example, in their associated caregiving activities and burden [44, 53]. In this study, burden of caregivers of adults with depression was compared to caregivers of other chronic conditions including 8% with other serious mental illness. The aggregation of these other diseases was used as a first step in further dissecting the expected heterogeneity of burden between different diseases. The potential bias of aggregating diseases and including serious mental illness in the CG-OD group would likely be an attenuated effect; however, excluding those with serious mental illness in subgroup analyses resulted in minimal alteration of the estimates. The results of this study, although small in their effect size, provide the foundation for further investigation of the differential burden of caring for depression vis-à-vis other caregiving conditions.