Introduction
An increasing number of older adults with care needs continue to live in their homes because of personal preferences and policies on ageing in place. Depending on the societal context, the needs of community-dwelling older adults can be met by care from multiple sources: publicly provided services, informal care-givers and private services purchased out-of-pocket. However, as a result of the expected increase in public expenditure for long-term care services (Lipszyc et al., Reference Lipszyc, Sail and Xavier2012), many countries are reforming their care policies, and the boundaries between the provision of public-, family- and market-based services are likely to shift (Rauch, Reference Rauch2008; Rostgaard, Reference Rostgaard2011; Geerts and Van den Bosch, Reference Geerts and Van den Bosch2012; Rostgaard and Szebehely, Reference Rostgaard and Szebehely2012).
This policy development calls for a better understanding of care utilisation among community-dwelling older adults: when and under which circumstances do older individuals rely on formal versus informal sources of care? Societies need answers to these questions not only to better understand the relationship between formal and informal care, but also to be able to attend to sub-groups of older adults who may be affected differently by shifting care policies.
The relationship between formal and informal care
Although a number of scholars have suggested a conceptualisation of the link between formal and informal care provision, no clear consensus exists (for a review, see Lyons and Zarit, Reference Lyons and Zarit1999). There are two main competing theories explaining the relationship between formal and informal care on a systemic level. On one hand, the substitution model suggests an inverse relationship between formal and informal care, according to which an increase in the provision of one form of care will result in a decrease in the other form of care (Greene, Reference Greene1983). On the other hand, the complementarity model suggests that formal and informal care have different structural characteristics (e.g. in terms of the care-givers’ geographical and emotional proximity to the care recipient, their training and skills set, and their flexibility in time use and timing of service provision), which foster a dual specialisation, resulting in formal and informal care functioning as supplements rather than substitutes (Litwak, Reference Litwak1985; Noelker and Bass, Reference Noelker and Bass1989).
Empirical evidence generally supports the hypothesised substitution effect between informal and formal care (e.g. Bonsang, Reference Bonsang2009; Van Houtven and Norton, Reference Van Houtven and Norton2004). However, when taking into account individual-level factors, such as the level of needs of the care recipient, the substitution effect persists only as long as needs are low and require unskilled care (Bonsang, Reference Bonsang2009). This implies some degree of specialisation (i.e. between formal and informal sources of care) according to the individual needs of the care recipient. In addition, recent studies support a bridging hypothesis suggesting that receiving informal support may function as a bridge to formal services, e.g. when informal care-givers help facilitate the ongoing contact between the care recipient and the formal care system (Geerlings et al., Reference Geerlings, Pot, Twisk and Deeg2005).
Individual factors explaining variation in care utilisation
The needs and resources of the care recipient are likely not only to vary between individuals but also to change and develop over time within individuals. To understand the individual circumstances underpinning variations in care utilisation in old age, a number of studies have employed Andersen and Newman's 1973 behavioural model of health service use (Andersen and Newman, Reference Andersen and Newman2005; for a review, see Kadushin, Reference Kadushin2004). This model distinguishes between predisposing, enabling and needs factors. The predisposing factors are characteristics that increase the propensity to use services, but exist prior to the onset of specific episodes of illness, such as age and gender. The enabling factors are resource characteristics that affect service availability or accessibility, such as income, health insurance and a social support network. The needs factors represent the most immediate causes of service use, including physical and cognitive health problems, functional decline and disability (Andersen and Newman, Reference Andersen and Newman2005). Empirical evidence supports Andersen and Newman's model, in that factors such as age, number of functional limitations in (instrumental) activities of daily living, living alone and a low level of available informal support are associated with an increased likelihood of using formal home care services (for a review, see Kadushin, Reference Kadushin2004). Recently, however, scholars have pointed out that previous studies have often taken a static approach, failing to account for dynamic trajectories of care utilisation over time (Geerts and Van den Bosch, Reference Geerts and Van den Bosch2012; Kemp et al., Reference Kemp, Ball and Perkins2013; Barrett et al., Reference Barrett, Hale, Butler, Barrett, Hale and Butler2014).
Only with the development of high-quality panel databases on ageing have scholars started to investigate the dynamics of care utilisation. For example, Geerlings et al. (Reference Geerlings, Pot, Twisk and Deeg2005) predicted transitions in utilisation of professional and informal care between consecutive waves of the Longitudinal Ageing Study Amsterdam (LASA) and found that variations in needs factors, as well as age, partner status and income, are important predictors of transitions between different sources of care. Using the same data, Pot et al. (Reference Pot, Portrait, Visser, Puts, van Groenou and Deeg2009) examined transitions in utilisation of acute and long-term care in the last phase of life and found that variations in needs factors were essential in explaining the increase in long-term home care towards the end of life. Our study adds to the growing literature on the dynamics of care utilisation, by exploring 15-year trajectories of care utilisation among community-dwelling older adults and by examining the individual factors underpinning variations in utilisation of care over time.
Socio-political factors explaining variation in care utilisation
Recent comparative research has also demonstrated that welfare state context plays a role in care utilisation, and that social policies could be viewed as a macro-level enabling factor in line with Andersen and Newman's model. Comparing care utilisation in 11 European countries, Suanet et al. (Reference Suanet, van Groenou and Van Tilburg2012) found that older adults were more likely to receive only formal home care or a combination of formal and informal care in countries with more extensive welfare state arrangements (i.e. more home-based services, higher pension generosity, etc.) than in countries with fewer home-based services and less residential care, where older adults were more likely to receive informal care only. Nevertheless, Geerts and Van den Bosch (Reference Geerts and Van den Bosch2012) found in their study of transitions in formal and informal care utilisation among older Europeans that, although the likelihood of becoming a formal or informal care user varied between policy settings across the examined countries, becoming a formal care user was strongly related to informal support becoming unavailable.
The Nordic welfare model is often distinguished (e.g. from the models in central and southern European welfare states) by its commitment to universalism, and by the formal care regime with its de-familising potential (Esping-Andersen and Korpi, Reference Esping-Andersen and Korpi1986; Leitner, Reference Leitner2003). In such a formal care regime, older adults mainly receive care from the formal care system, just as older adults in those settings tend to prefer formal care services instead of family-provided informal care (Eurobarometer, 2007: 97). Comparative studies have, however, indicated that, although adult children's provision of care-specific tasks (i.e. help with bodily activities) is lower in the Nordic countries compared to the southern European ones, the rates of practical help from adult children to their parents (i.e. help with the housekeeping, gardening, transport) are higher (Brandt et al., Reference Brandt, Haberkern and Szydlik2009; Haberkern and Szydlik, Reference Haberkern and Szydlik2010).
Given these previous findings, there are reasons to expect that the dynamics of care utilisation in the context of a Nordic welfare state may appear different than in other societal settings. By investigating care utilisation among older adults in Denmark, this study adds the perspective of a comprehensive welfare model to the existing literature on the interplay between formal and informal care and the underpinning factors.
Aim of the study
The aim of this study is twofold. The first aim is to examine the dynamics of care utilisation over a 15-year period, exploring the patterns in older adults’ transitions between different sources of care (formal, informal and a combination). The second aim is to examine the individual level factors underpinning variations in care utilisation, in the context of a comprehensive Danish welfare state.
To guide our investigation, we have formulated three main expectations, as follows. First, we expect to find transitions between different sources of care (i.e. formal, informal and the combination) over time. Second, we expect that predisposing, enabling and needs factors will determine individual patterns of care utilisation. We expect, however, that individual-level enabling factors, such as financial resources and social network, will have limited importance in a comprehensive welfare state context.
We investigate these expectations by performing two sets of analysis. First, in order to explore patterns of care utilisation and transitions between formal and informal care over time, we adopt a novel sequence-based method, which allows for an explorative and dynamic examination of care utilisation. Second, in order to examine factors underpinning variations in care utilisation, we analyse which predisposing, enabling and needs factors are associated with distinct trajectories of care utilisation in old age, in the context of a formal care regime, Denmark.
Data and methods
Data
The data for this study come from the first four waves of the Danish Longitudinal Study of Ageing (DLSA), collected in 1997, 2002, 2007/2008 and 2012/2013. The DLSA has a panel structure, with the same respondents contacted in each wave. The core sample includes a random sample of cohorts born in 1920, 1925, 1930, 1935, 1940 and 1945, drawn from the Danish Central Population Registry that includes all Danish residents. Thus, the DLSA is highly representative of the target population (Kjær et al., Reference Kjær, Poulsen and Siren2016). An additional cohort of 52 year olds is added to each wave, along with refreshment samples, to repair attrition; however, in this study, we included only the core sample (i.e. respondents sampled at baseline). The number of interviewed respondents at baseline was N = 5,864 (response rate: 70%), and a total of N = 2,406 respondents participated in all four waves of the survey (retention rate: 41%). For the purpose of this study, the DLSA was further linked to records from the national registries on immigration, income and mortality to determine the origin and socio-economic position of respondents, along with reasons for attrition (e.g. mortality) on an individual level (for a detailed overview of the registries, see Thygesen et al., Reference Thygesen, Daasnes, Thaulow and Brønnum-Hansen2011).
Study sample and attrition
For the purpose of this study, we further restricted the study sample by two criteria. First, because only respondents beyond full retirement age (i.e. 65 years) responded to the items regarding care services, we limited the sample to community-dwelling respondents, aged 67 and above at baseline. Second, because the analysis of care trajectories required a full sequence of data, we limited the sample to respondents for whom we had full information on items related to care utilisation in all four waves. Thus, the study sample consisted of 473 community-dwelling respondents who self-reported utilisation of care in four consecutive waves of the DLSA. In the next section, we reflect on potential biases from sample selection and attrition.
Figure 1 illustrates the timeline and age for each cohort included in the study sample. There are four observed positions: t 0, t 5, t 10 and t 15, which correspond to the timing of each wave (i.e. 1997, 2002, 2007/2008 and 2012/2013). In the Methods section, we reflect on the implications of choosing an external timeline (waves) instead of an internal timeline (respondent's age). Three cohorts (born 1920, 1925 and 1930) are included in the study, and the respondents’ age throughout the study period, so that the sample's age ranges from 67 to 77 years at t 0 and from 82 to 92 years at t 15.
Figure 1. Timeline and age for three cohorts included in the study population.
Attrition analysis
Attrition resulting from death and physical or cognitive decline can cause bias if the retained sample is selective towards those who remain healthy enough to participate (Chatfield et al., Reference Chatfield, Brayne and Matthews2005). To estimate the extent to which attrition influenced our results, we investigated observable baseline characteristics associated with later attrition from the study sample. Further, by linking survey data to mortality records (i.e. date of death), we were able to divide attriters by the cause of attrition (mortality versus other; in line with Banks et al., Reference Banks, Muriel and Smith2011; Kelfve et al., Reference Kelfve, Fors and Lennartsson2017).
Table 1 presents the results of the attrition analysis. Passive attrition includes baseline participants who passed away before t 15 (i.e. due to mortality), whereas active attrition includes baseline participants who were alive by t 15 but dropped out of the study for any other reason (e.g. refusal to participate). Passive attriters were older, in worse health and more often care recipients already at baseline than active attriters and participants, suggesting that our analysis may suffer from survival bias. However, the comparison between active attriters and included participants (χ2 test in the far-right column) indicates that active attriters and retained respondents had similar distributions on most observable baseline characteristics. At baseline, active attriters and retained participants were not significantly different with regards to care utilisation, gender, origin, education, co-habitation or marital status. Nevertheless, a larger share of active attriters did report poor health already at baseline than among retained participants (29 versus 22%), and active attriters were older than retained participants. This could be explained partly by the fact that the active attrition category also includes individuals who were excluded from the study sample, because they moved to an institutional care facility. Thus, we expect the study sample to be representative of a broader population consisting of surviving and community-dwelling older adults.
Table 1. Selected baseline characteristics of participants included in and excluded from the study sample
Notes: The comparison is only for baseline respondents, who were 67 or more at baseline. The category ‘passive attrition’ includes participants who passed away between t 0 (1997) and t 15 (2012/2013). The category ‘active attrition’ includes participants who either dropped out of the study or moved to an institutional care facility or had item non-response on selected items between t 0 and t 15. The far-right column reports tests of difference in distributions between included participants and active attriters. It was possible to receive more than one type of care at the same time, which is why the care categories do not sum to 100 per cent. 1. The category ‘unmarried’ included never married, divorced and widowed. df: degrees of freedom.
Measurements
Dependent variable(s)
Our dependent variable was the (dynamic) receipt of care in the home from different sources. We distinguished between those receiving no care at all, those receiving informal care only, those receiving formal care only, and those receiving a combination of formal and informal care. Because only 4 per cent of respondents relied solely on private care, we did not consider private care in the operationalisation of the dependent variable. However, in a later analysis, we considered private care receipt as a descriptive characteristic.
First, informal care covered the receipt of practical help (i.e. help with instrumental activities of daily living (IADLs)) from extra-household family members and/or friends, but excluded any help (potentially) received from a co-habiting partner, because this type of help is not covered in the DLSA survey, and because we were interested mainly in examining extra-household care. Our indicator for informal care covers mainly practical help, because the DLSA survey does not cover personal care from informal sources. Respondents who replied that ‘children or other usually takes care of’ any of seven daily activities (i.e. heavy cleaning, light cleaning, shopping, clothes and ironing, breakfast, cold meal, hot meal), or who replied ‘yes’ to having received help from ‘children or other family’ or ‘friends and others’ with any of the nine activities (‘cleaning, washing, shopping, cooking, gardening, maintenance of the house, money issues or contact with public institutions, transport to treatment or medical examination, getting outside to an activity or the like’) within the past monthFootnote 1 were coded as recipients of informal care.
Second, formal care covered the receipt of publicly financed home-care services including help with both practical (IADL activities) and personal care (ADL activities) and/or receipt of home nurse services (i.e. help with wound care, review of medications, etc.). Respondents who replied ‘yes’ to any of the items ‘Do you receive home help – permanent or temporary?’ and ‘Do you have regular visits from a home nurse?’ were coded as recipients of formal care. As a result, our indicator for formal care covers both practical help, and personal and medical care.
Explanatory variables
Our explanatory variables were divided into predisposing, enabling and needs factors. As predisposing factors, we considered cohort (born 1920, 1925 or 1930), gender (female versus male) and level of education (seven years of schooling versus eight years or more). Ethnic origin was not included in the analysis, owing to a very small share (1%) of people with ethnicity other than Danish in the sample. Because the predisposing factors were at all times constant, we relied on the baseline measure of each variable.
As enabling factors, we considered both financial resources and social network. Financial resources are important to consider because they provide access to market-based services in particular. We included yearly household disposable income in quartiles (mean = €26,406, standard deviation = €12,294).
Social support is an important resource for informal care, often provided by adult children, friends and neighbours (Cantor, Reference Cantor1979; Geerts and Van den Bosch, Reference Geerts and Van den Bosch2012; Kalwij et al., Reference Kalwij, Pasini and Wu2014). The following variables were included: marital status (married versus unmarried (including never married, divorced and widow(er)); co-habitation (co-habiting versus single-living); frequency of contact with adult children (no child or less than monthly versus monthly or more often); and having a ‘confidant’ (i.e. someone to turn to when in personal problems versus no one). Because marital status and co-habitation can vary over time, we examined both the level and change. Marital status was coded in three categories: continuously married, continuously unmarried and incident widow(er), where the last category refers to someone married at t 0 but widowed at t 15. This variable captures marriage and co-habitation with a partner, but not co-habitation with other household members. However, because previous studies have demonstrated that having other household members present is also associated with lower use of formal care (see Chappell and Blandford, Reference Chappell and Blandford1991; Rodríguez, Reference Rodríguez2013), we also included an alternative indicator capturing co-habitation in general, with four categories (continuously co-habiting, continuously single-living, incident single-living and incident co-habiting). These two alternative indicators were tested against each other when deciding on the final model fit.
We included the frequency of contact with children instead of alternative operationalisations (e.g. such as having any children or not, or the geographical distance from children), because we wanted to know the extent to which the respondent's children could be counted on as a potential source of support, rather than whether the respondent had any children or not (see e.g. Tomassini et al., Reference Tomassini, Kalogirou, Grundy, Fokkema, Martikainen, Van Groenou and Karisto2004). Having a confidant was considered a proxy for socio-emotional support, and thus we chose to include this in the study as a single-item subjective indicator to reflect support from anyone (e.g. friends and neighbours, family, etc.)
Needs factors included four indicators reflecting different areas of need. For ease of interpretation, each indicator was coded into a dummy reflecting low versus high needs. The first indicator was self-rated health (SRH; very good or good versus fair, poor or very poor). The second indicator was physical functioning, operationalised as ADL dependence, based on self-reported dependence (able to perform activity independently versus cannot do it without help) in any of six ADLs (i.e. cutting toenails, climbing stairs, walking outside, walking inside, wash or shower, and dress and undress; in line with Shanas, Reference Shanas1968: 26f.). The third indicator was medical conditions (one or more conditions versus none) based on self-report of eight general practitioner-diagnosed conditions (i.e. hypertension, diabetes, bronchitis, osteoarthritis, myalgia, osteoporosis, back disease and depression). The fourth indicator was a dummy for self-reported memory problems (none versus some or many) based on a single item. For each need factor, we examined both the level and the change over time. For example, changes in SRH between t 0 and t 15 were coded into four categories: stable good SRH, stable poor SRH, emerging good SRH and emerging poor SRH.
Table 2 shows the distribution of the study sample on key variables. It is evident that enabling factors and needs factors in the sample change over time. Similarly, or likely as a consequence, care utilisation increases over time.
Table 2. Distribution on key variables in the study sample
Notes: N = 473. SD: standard deviation. ADL: activity of daily living.
Data analysis
To carry out a dynamic assessment of care utilisation, we used social sequence analysis to cluster individual care trajectories into a typology, which we then used as a dependent variable in the second step of the analysis. Social sequence analysis is a method that can be used to examine full sequences without making any assumptions about patterns in the data (Abbott and Tsay, Reference Abbott and Tsay2000). All sequences consisted of four observations (positions t 0, t 5, t 10, t 15), according to the defined state alphabet of care utilisation (i.e. no care, informal care only, formal care only, and combined formal and informal care). Sequences were analysed and clustered in R with the TraMineR and Cluster packages for mining and visualisation of categorical sequence data (Gabadinho et al., Reference Gabadinho, Ritschard, Mueller and Studer2011).
The analysis of care trajectories takes its starting point in an alphabet of possible sequence states, corresponding to each source of care as well as their combinations: (A) no care, (B) informal care only, (C) formal care only and (D) a combination of formal and informal care.
Figure 2 illustrates an example of three possible care trajectories, according to the selected state alphabet. We visualise the individual longitudinal succession of care states from t 0 to t 15, as well as (through the length of each segment) the duration (i.e. number of successive positions) spent in each state. For example, Observation 1 spends two positions (t 0 and t 5) receiving no care, then spends one position (t10) receiving informal care only, followed by one position (t 15) receiving formal care only. Observation 2 receives no care at all over the course of 15 years (t 0 to t 15), whereas observation number three transitions from no care (t 0) to informal care (in t 5), followed by a combination of formal and informal care (from t 10).
Figure 2. Illustration of three hypothetical care sequences according to the selected state alphabet.
The sequence analysis then followed three steps. First, individual care trajectories were matched by their longest common sub-sequence. Because our sequences were structured according to an external timeline (i.e. waves) instead of an internal one (i.e. ageFootnote 2), the order of events was meaningful, while the timing of events had limited significance. For this reason, we chose a distance metric that is based on Longest Common Subsequences (LCS) in the data (Elzinga, Reference Elzinga2007). The LCS dissimilarity metric was useful in our case, because it gives priority to the similarity of the order of events in the data (without giving weight to contemporaneous similarityFootnote 3). Second, in order to produce a typology of care trajectories, we combined the output from the matching (i.e. the dissimilarity matrix) with a data reduction procedure based on Agglomerative Hierarchical Clustering (‘agnes’) using the Ward clustering method in R (Studer, Reference Studer2013). This procedure organises sequences into clusters in a way that maximises the similarity of cases within each cluster while maximising the dissimilarity between groups. Third, in order to decide where to cut off the number of clusters for a meaningful typology of care trajectories, we first assessed the graphical representation of the clustering by examining a Ward Dendogram, and next computed several quality measures of clustering and plotted their standardised values. Based on local spikes in the standardised scores of these tests, we were able to decide on an optimal cut-off point at four clusters. In sum, the followed procedure resulted in four distinct clusters of care utilisation. Although these clusters are distinct in the sense that they each represent a different overall pattern of care use over time, the clustering method does allow for variability within clusters (i.e. some degree of deviation from the overall pattern of the cluster). For example, each of the four clusters also covers a smaller sub-set of trajectories that are less similar to the overall pattern of the group. However, the key idea is that each participant was clustered in a way that maximised the similarity within each cluster.
To examine factors explaining variation in care utilisation, we ran a multinomial logistic regression model. For this purpose, the four-cluster solution was imported to Stata and used as the dependent variable in a multinomial logistic regression model, with predisposing, enabling and needs factors as the explanatory variables. Some of the explanatory variables were highly correlated, so to decide on the best model fit, likelihood ratio tests were used to assess the contribution of different sets of variables to the model fit (results not reported here). The analysis indicated that predisposing variables should be limited to age and gender, because adding educational level did not add more to the model fit. Enabling variables were limited to contact with children, marital status, a dummy for having a confidant and income quartiles. Co-habitation was chosen over marital status because it added more to the model fit. The needs variables included in the final model were ADL dependence and medical conditions, after which neither SRH nor memory problems added more to the model fit. In the Results section, we report only on the final model specification, which included the following variables: Cohort, gender, contact with children, marital status, having a confidant, income, ADL dependence and medical conditions. For ease of interpretation, the coefficients from the multinomial logistic model were transformed into mean marginal effects.Footnote 4 We do not suggest any causal interpretation of the results; rather, we aim to describe different trajectories of care by examining predisposing, enabling and needs factors associated with each trajectory.
Results
Trajectories of care utilisation
To investigate variations in patterns of care utilisation over time, we performed a sequence analysis based on data from 15 years of care utilisation. We clustered together similar trajectories based on matching of LCS. The analysis resulted in four distinct clusters of care utilisation.
Figure 3 shows the set of frequency plots for the four clusters. Each plot displays the ten most frequent sequences in the cluster, and sequences are displayed bottom-up in decreasing order of their frequencies. Most clusters are dominated by transitions in care use over time. Cluster 1 (57 per cent of the sequences) is dominated by transitions from no care use to informal care and back to no care use again; Cluster 2 (21 per cent of the sequences) is dominated by transitions from no care use to use of formal care only, and from using formal care only to using a combination of formal and informal care; Cluster 3 (13 per cent of the sequences) is dominated by transitions from no care use to use of informal care only; and Cluster 4 (10 per cent of the sequences) is dominated by transitions from no care use to use informal care only, followed by a transition to a combination of formal and informal care.
Figure 3. Sequence frequency plots for a four-cluster solution.
Factors explaining variations in care utilisation
To investigate the determinants of variations in which care utilisation occurs, we used multinomial logistic regression to analyse how cluster membership was associated with individual-level predisposing, enabling and needs factors known to predict variations in health service utilisation.
Table 3 shows the results from a multinomial logistic regression model treating the four clusters of care trajectories as outcomes. For ease of interpretation, coefficients of all covariates were transformed into mean marginal effects. Thus, the results for each covariate (in Table 3) reflect the expected percentage-point change in probability of cluster membership, compared with the reference group (with all other covariates, as observed). A positive value reflects a higher probability of cluster membership, and a negative value reflects a lower probability.
Table 3. Multinomial logistic regression model predicting cluster membership by predisposing, enabling and needs factors
Notes: N = 473. Multinomial logistic regression. All other covariates as observed. SE: standard error. Ref.: reference group. ADL: activity of daily living.
Significance levels: † p < 0.10, * p < 0.05, ** p < 0.01.
Factors explaining trajectories dominated by no or sporadic informal care use (Cluster 1)
Cluster 1, which we labelled ‘sporadic informal care’, is the largest cluster and is characterised either by no care use at all from t 0 to t 15, or alternatively by single periods of informal care followed by no use of care. Nearly all predisposing, enabling and needs factors included in the model are negatively associated with membership in this cluster. On average, participants have significantly lower probability of belonging to this cluster if they are older, female or single-living, as compared with younger, male and co-habiting participants. Further, stable or emerging ADL dependence decreases the likelihood of belonging to this cluster, as opposed to stable good health, which increases the likelihood. Having a confidant and belonging to the highest income quartiles are factors that increase the likelihood of cluster membership. In sum, older adults who are not predisposed to service use, who have trajectories characterised by low health needs, and who have available resources in terms of both income and social support are more likely to experience a care trajectory characterised by limited reliance on any source of care or by sporadic receipt of informal care.
Factors explaining trajectories dominated by increasing formal care (Cluster 2)
Cluster 2, which we labelled ‘increasing formal care’, is characterised by trajectories involving a transition from no care use to reliance on formal care only. For a sub-set of this cluster, the care trajectory ends in receiving a combination of formal and informal care. Among the examined predisposing factors, older age (i.e. cohort born in 1920) significantly increased the likelihood of belonging to this cluster by 19 percentage points compared with the youngest age group (i.e. cohort born in 1930). Furthermore, this care trajectory was strongly associated with a high needs level: stable and emerging ADL dependence was associated, respectively, with a 14 and 11 percentage point greater risk of membership in this cluster than stable independence in ADLs. In addition, contact with children was negatively associated with receipt of formal care only; thus, on average, participants with no children or limited contact with their children had a 14 percentage point increased likelihood of cluster membership over participants with frequent contact with their children. Taken together, reliance primarily on formal care was associated with being predisposed to service use, having high or escalating health needs, and having limited resources in terms of social support.
Factors explaining trajectories dominated by increasing informal care (Cluster 3)
Cluster 3, which we labelled ‘increasing informal care’, is characterised by trajectories involving a transition from no care use to reliance on informal care only. This cluster is different from Cluster 1 in that informal care is received continuously rather than sporadically. Among the predisposing factors, only older age was negatively associated with membership in this cluster so that older respondents (i.e. cohort born in 1920) had a 7 percentage point smaller likelihood of belonging to Cluster 3 than the youngest age group (i.e. cohort born in 1930). Regarding the enabling factors, we found that participants with frequent contact with their children were 14 percentage points more likely to belong to this cluster than participants with limited contact. Of needs factors, ADL dependence did not significantly alter the likelihood of belonging to this cluster, but having one or more medical conditions did increase the likelihood by 11 percentage points over having none. Taken together, a trajectory characterised by informal care only was associated with younger age, some care needs and having adult children as a source of informal support.
Factors explaining trajectories dominated by combined care use (Cluster 4)
Cluster 4, which we labelled ‘bridging to combined care’, is dominated by a combination of formal and informal care, which is often preceded by receipt of informal care only. We interpret this as a form of bridging and/or dual specialisation. First, this cluster was (more than any other cluster) dominated by old age and a high level of needs. Respondents in the two oldest age groups (cohorts born in 1920 or 1925) had an approximately 8 percentage point greater likelihood of belonging to this cluster than respondents from the youngest age group (cohort born in 1930). Participants with stable or emerging ADL dependence had, respectively, a 14 and 6 percentage point greater likelihood of belonging to this cluster than independent individuals. Second, enabling factors played an important role in determining membership in Cluster 4. Continuously and incident single-living individuals had, respectively, a 5 and 16 percentage point elevated risk of belonging to this cluster. Participants with frequent contact with children had a 10 percentage point greater likelihood of receiving a combination of formal and informal care than those without contact. Furthermore, those in the highest income quartile had a smaller likelihood of belonging to this cluster, i.e. receiving a combination of formal and informal care. Taken together, a trajectory dominated by reliance on formal and informal care in combination was associated with higher age, an elevated needs level, fewer financial resources and access to sources of informal support.
Additions to the care-state alphabet
Because we excluded receipt of private care and service from the state alphabet of the sequence analysis, we examined separately the use of private services and whether it differed across clusters. In Denmark, privately purchased care services most often include practical help (i.e. household maintenance) and less often personal care. A cross-tabulation and χ2 test of clusters, and use of private help and care in t 15 (results not reported here), indicated that respondents belonging to Cluster 2 (i.e. care trajectories characterised by reliance solely on formal care) utilised private help to a larger extent. In other words, those lacking a social support network may more often be in need of (or choose to) supplement formal care with private services purchased out-of-pocket, whereas those with a social support network need not do so.
Also, because our care-state alphabet did not distinguish between different types of formal home care (practical help versus personal care, both provided by the municipality), we examined the receipt of personal care across clusters (results not reported in the text). We compared only Cluster 2 with Cluster 4, which were clusters characterised by receipt of formal care. We found that the share of cluster members receiving formal personal care was significantly greater in Cluster 4 than in Cluster 2.
In Table 4, we summarise the characteristics of each cluster and the predisposing, enabling and needs factors that significantly predict cluster membership.
Table 4. Summary of cluster characteristics
Notes: 1. Results stemming from additions that were made to the original care-state alphabet are in parentheses. ADL: activity of daily living.
Discussion
In this paper, we explored and explained dynamic variations in care utilisation over a 15-year period in a formal care regime, Denmark. There were two main findings. First, the sequence analysis of transitions between different states of care (i.e. formal care, informal care and the combination) confirmed our expectation that care is a dynamic process, in which care recipients transition between different sources of care over time. Second, the results confirmed that sources of individual care utilisation change over time because the determining individual factors change. Our analyses of patterns of care utilisation revealed a typology of four distinct care trajectory clusters that differed in individual-level predisposing, enabling and needs factors, supporting Andersen and Newman's behavioural model of service use (Andersen and Newman, Reference Andersen and Newman2005). Contrary to our expectation, however, enabling factors such as financial resources and social network were not overruled by the generous welfare context, but played a role in care utilisation.
Two of the clusters were characterised by low care needs and use of informal help only (Clusters 1 and 3), and the members of these clusters were younger. The two remaining clusters (Clusters 2 and 4) were characterised by older respondents with a high level of care needs who transitioned to the use of formal services over the 15-year period. Nevertheless, although those in Cluster 4 relied on a combination of informal and formal care while transitioning, members of Cluster 2 received no informal care. Also, although members of Cluster 4 received informal care for practical tasks and formal care predominately for personal care, members of Cluster 2 received formal care (both privately and publicly paid), especially for practical tasks.
Although Clusters 2 and 4 were similar in both predisposing and needs factors, they differed in terms of enabling factors. Members of Cluster 2 were more likely not to have children (or contact with children), and members of Cluster 4 were more likely not to have high disposable incomes. The differences found between Clusters 2 and 4 support the model of dual specialisation, as the informal care receipt in Cluster 4 seems not to crowd out formal help on personal care, but tends to focus on help with practical tasks. Similarly, members of Cluster 2 who lack informal care-givers have greater likelihood of receiving formal help, especially with practical tasks. The found combination of informal and formal care among those who have both care needs and a social network also supports the bridging hypothesis, suggesting that informal support functions as a bridge to formal services (Geerlings et al., Reference Geerlings, Pot, Twisk and Deeg2005).
Somewhat contrary to our expectation, we identified disposable income as a factor influencing the receipt of care. Members of Cluster 4, who received a mix of formal and informal care, generally had a lower income level. There are two potential explanations. First, individuals with fewer financial resources are likely to have more care needs because of the socio-economic gradient in health; second, they have fewer financial resources to allow other sources of care (such as private care), resulting in a reliance on a combination of informal and formal care services.
This study extends the literature on the relationship between formal and informal care in a number of ways. First, it provides evidence of the dynamic nature of care needs and receipt in later life (e.g. Barrett et al., Reference Barrett, Hale, Butler, Barrett, Hale and Butler2014). In our longitudinal analysis, a majority of the study participants transitioned from no care use to some type of care use in the 15-year period studied, but the trajectories consisted typically of several transitions and there was a variation in the type of care received.
Second, supporting Andersen and Newman's behavioural model, our study indicates that individual predisposing, enabling and needs factors are important predictors of variations in care utilisation in later life. As also found elsewhere, old age, functional limitations and medical conditions, and living alone predict utilisation of formal and informal care. However, our study adds the perspective of a comprehensive welfare context to the existing literature and furthers the scholarly understanding of the interplay of macro- and micro-level factors that underpin care utilisation in old age. Contrary to our expectation, enabling factors such as social support and income are important resources for care receipt. This is an important finding in that it raises questions about the comprehensiveness of the contemporary Nordic care regime. While access to care is ideally universal in such a regime, our study shows that financial resources and the availability of informal care still affect care utilisation in later life.
Compared with recent findings from other policy settings (e.g. Kadushin, Reference Kadushin2004; van Groenou et al., Reference van Groenou, Glaser, Tomassini and Jacobs2006; Geerts and Van den Bosch, Reference Geerts and Van den Bosch2012; Kemp et al., Reference Kemp, Ball and Perkins2013; Rodríguez, Reference Rodríguez2013), our findings reveal some similarities and differences. Compared with recent findings from Spain (Rodríguez, Reference Rodríguez2013), the predisposing and needs factors influence care utilisation in a way similar to our formal care regime context and in the more familistic care policy context of Spain. However, in our study, trajectories that were characterised by formal care use were more strongly associated with health-related needs than trajectories characterised by informal care. This finding supports the characterisation of Denmark as a formal care regime in which publicly financed services are provided universally, yet are based on an assessment of needs (contrary to a means-based provision), and where public bodies rather than the family caters to highly care-dependent citizens (Esping-Andersen and Korpi, Reference Esping-Andersen and Korpi1986; Schulz, Reference Schulz2010).
Nevertheless, although access to services is universal and needs based, having children is still an asset for receipt of care. We demonstrate that even in the context of a strong formal care regime, the family and informal social network plays a role in caring for older adults in three out of four typical care trajectories. This result is in line with previous findings from Sweden (see Larsson and Silverstein, Reference Larsson and Silverstein2004). These findings from the Nordic countries indicate that even in the context of a strong formal care regime, the family and informal social network plays a role in caring for older individuals. Furthermore, our analyses demonstrated that those relying solely on formal care were more prone to supplement with private services purchased out-of-pocket. This dualisation of care (where some older adults rely on (free) family-provided informal services whereas others turn to marketised services) has been argued to reflect recent cuts in tax-funded home care (e.g. Rostgaard and Szebehely, Reference Rostgaard and Szebehely2012).
Our findings regarding the determinants of care raise concern about whether the care system is comprehensive enough to overcome individual differences with regards to enabling factors. Future studies should look into whether older adults with care needs who lack financial and social resources are at risk of experiencing care-deficits, and to what extent the Nordic welfare system is able to buffer this risk.
Limitations
Our dynamic investigation introduces a novel methodological approach to the field of care research. The sequence-based method had two main advantages. First, it allowed for an assessment of transitions between different sources of care; second, it made possible an explorative examination of variations in individual trajectories of care utilisation over a 15-year period. However, the present study also had certain limitations.
The restriction of the sample to retained respondents (i.e. the panel) came at the cost of representativity, and thus generalisability. Thus, the findings from this study can be extended only to the population consisting of surviving, community-dwelling older adults. In addition, because we relied on prospective panel data, our trajectories were limited to four observation points covering a 15-year period. As a result, instead of full trajectories of care utilisation, we observed snapshots of the process. Furthermore, because of the sampling design of the survey, we had to align the sequences around an external rather than internal timeline (i.e. waves instead of age), and hence, we chose the dissimilarity measure (LCS) accordingly.
Although the sequence analysis is a novel method, it requires using a simple care-state alphabet (with few and mutually exclusive states of care) rather than a complex one (with overlapping types of care). Hence, we relied on a care alphabet with only three states (i.e. informal, formal and combined care). Future research using alternative methods of analysis should look into care trajectories and consider other types of care as well, such as long-term institutional care (e.g. nursing home admission) and short- and long-term hospital admissions.
Finally, the data did not allow operationalisation of any institutional changes occurring in the 15-year period studied. Although Denmark has arguably experienced fewer institutional changes in its formal care policies than the other Nordic countries (e.g. Rauch, Reference Rauch2008; Rostgaard and Szebehely, Reference Rostgaard and Szebehely2012), institutional variation may explain some of the transitions found between different sources of care over time.
Conclusion
Previous research has often taken a static approach to understanding care utilisation, thereby failing to account for variations in individual needs and resources over time. This study responds to this gap by providing new insights into the dynamic nature of care use, into the relationship between formal and informal care, and into the factors that underpin different patterns of care utilisation in old age. Based on our dynamic analysis, we point to several distinct trajectories of care use that vary in terms of the formal and informal sources of care utilised. Furthermore, our findings provide new evidence on the associations between care use and multiple determining factors. Our results emphasise how patterns of individual care utilisation change over time as the individual and societal determinants change. Most importantly, we identify a sub-group of older adults with high needs and limited social resources that tend to rely solely on formal services. This is a group likely to be (negatively) affected by care policies, shifting the care responsibility from the state to the family or the market.
Acknowledgements
The authors wish to thank Professor Tine Rostgaard for her constructive comments on an earlier version of this paper.
Author ORCIDs
Agnete Aslaug Kjær, 0000-0003-3644-9689.
Author contributions
The authors declare that both authors have made substantial contributions to conceptions and design, acquisition of data, and analysis and interpretation of data, that both authors have participated in drafting the article or revising it critically for important intellectual content, and that both authors have approved the final version.
Financial support
This work was supported by a PhD grant jointly financed by the Department of Political Science at the University of Copenhagen and the Danish Center for Social Science Research; a research stay at Yale University financed by the Fox International Fellowship Program; and a research grant from Innovation Fund Denmark (grant number 6158-00002B).
Conflict of interest
The authors certify that they have no affiliations with or involvement in any organisation or entity with any financial or non-financial interest in the subject matter or materials discussed in this paper.
Ethical standards
Data collection, processing and analysis of personal data of respondents in the Danish Longitudinal Study of Ageing and linkage to register data have been approved by the Danish Data Protection Agency (J.nr. 2015-57-0083).
