The frail older person does not exist: development of frailty profiles with latent class analysis

For this study we performed a secondary data-analysis on The Older Person and Informal Caregiver Survey Minimum Dataset (TOPICS-MDS), a large data-sharing initiative in the Netherlands (for more information see [24]). In 2008, the Dutch Ministry of Health, Welfare and Sports started the National Care for the Elderly Programme (NCEP) which aimed at reorganizing health and social care according to the needs of older people. Between 2008 and 2014 several implementation and research projects were carried out and funded by the NCEP. Within the NCEP, the TOPICS-MDS instrument was developed, a standardized instrument to study the effects of these projects on older people and their informal caregivers. The instrument was based on other validated instruments on morbidity, quality of life, functional limitations, mental health, social functioning and health service utilization. Researchers in all projects collected the data consistent with the TOPICS-MDS so a national, uniform dataset was created. The TOPICS-MDS currently contains pooled data from 54 research projects which differ across study design, sampling framework an inclusion criteria. TOPICS-MDS is a fully anonymized dataset available for public access, and therefore the analysis in this study is exempt from ethical review (Radboud University Medical Centre Ethical Committee review reference number: CMO: 2012/120) [24]. For our study, we selected the baseline data of the respondents aged 60 years and older (n = 43,704).

Baseline measurements entailed: Self-reported health is assessed with two items from RAND-36. The first item allows older people to evaluate their own current general health in the following answer categories: excellent; very good; good; fair; poor. The second item is self-reported health compared to 1 year ago with five answer categories: much better; somewhat better; about the same; somewhat worse; much worse [25]. Cognitive functioning is measured by one item from EQ-5D + c focused on problems with memory, attention and thinking, and had three answer categories: no problems; some problems; extreme problems with memory, attention and thinking [26]. Social functioning is measured with one item on how often social activities are hampered by physical health or emotional problems. The possible answers are: none of the time, a little of the time, some of the time, most of the time, all of the time [25]. Mental health is measured on a five-item RAND-36 scale with items that question how often the respondents have felt nervous, calm and peaceful, down-hearted and blue, happy, or so down in the dumps that nothing could cheer them up. The scores range from 0 to 100 and a higher score implies better mental health [25]. Morbidity status is self-reported: participants could indicate their morbidities on a 17-item list of conditions (no/yes), such as heart failure, joint damage and hearing disorders [24]. The number of morbidities were summed and the score ranges from 0 to 17. Functional limitations are measured with the modified Katz-15 instrument that assesses the ability to perform 15 activities of daily living (ADL) and instrumental activities of daily living (IADL) (yes/no) such as getting dressed, shopping and taking medication [27, 28]. The number of activities that respondents cannot do is summed, ranging from 0 to 15 with a higher score indicating more functional limitations.

Frailty index is calculated from 45 health deficits in the TOPICS-MDS [29], including the before mentioned self-reported health, cognitive functioning, social functioning, mental health, functional limitations and the five items of the EQ-5D [26]. The number of health problems of the older person is divided by the total number of 45 health problems; the score ranges from 0 to 1 with a higher score indicating a higher level of frailty [29, 30].

Demographic variables: gender, living arrangement (independent; in residential care or nursing home), marital status (married or cohabiting; widowed or single), ethnicity (native Dutch; first/second generation migrant), educational level (primary school or less; practical/secondary vocational training; some college/university degree) and age.

The analyses were done in five steps combining quantitative and qualitative methods. First, we described the total sample, giving frequencies and percentages for the categorical variables and mean, standard deviations and range for the continuous variables (Table 1).

Second, we did latent class analysis (LCA) to identify subpopulations within a larger population of older people. LCA is a person-centred approach to identify unobserved groups of similar individuals (latent classes) based on observed variables. The aim of LCA is to find the best class solution; meaning the smallest number of latent classes describing the associations among a set of observed variables [31]. The observed variables we used in the LCA were self-reported health, social functioning and cognitive functioning as categorical variables and morbidity status, mental health, functional limitations as continuous variables. To avoid local likelihood maxima and inaccurate parameter estimates, we used 1000 multiple start values and 100 iterations [32]. For each class solution, we present the Akaike Information Criteria (AIC), Bayesian Information Criteria (BIC) and adjusted BIC (aBIC) which combine goodness of fit and parsimony [33]. We based the number of classes on the adjusted Lo-Mendell-Rubin likelihood test (bootstrap). The quality of the classification was determined by the entropy measure [34]. The various class solutions and model fit are presented in Additional file 1: Table S1. We used the Mplus 7.4 program. We based the final number of classes on the highest entropy score as it indicates the best quality classification.

Third, we described the final class solution according to the observed variables self-reported health, social functioning, cognitive functioning, morbidity status, mental health and functional limitations to identify the differences between them (see Table 2 and Additional file 2: Table S2).

Fourth, we assessed the quality of the LCA classification with a focus group of older people (see Additional file 3 for the focus group protocol). In LCA, the value of the classes should also be interpreted qualitatively. The focus group participants were members of the Elderly Forum of the Geriatric Network Rotterdam, one of the eight regional networks in the NCEP. All 15 members of the Elderly Forum were invited to the focus group and eight (five males and three females) were able to attend. The profiles were presented textually for each of the final classes as identified by LCA: Older people in this profile experience their health as [excellent/very good/good/fair/poor] and state that their health is [much better/somewhat better/about the same/somewhat worse/much worse] compared to a year ago. They experience [no/some/serious] problems with their cognitive functioning. They experience problems with social activities [a little/some/a good bit/most/all] of the time. Their mean score on mental health is [0–100]. They have [0–17] morbidities and need help with [0–15] daily activities. Besides the textual presentation, the final profiles were also presented together to provide a clear, visual overview. To validate the profiles, we asked the participants whether they recognized the profiles and if (how much) they could relate to them. In addition, we asked them to state which specific domain contributed the most to frailty in each of the profiles and invited them to rank the profiles from least to most frail. The focus group discussions were recorded and transcribed verbatim. We began the analysis by carefully rereading the transcript of the focus group several times and subsequently applied a data-driven approach to our thematic analysis per profile. We were looking for the interpretation of each of the profiles to understand the similarities and differences between the perceptions of the eight participants. Focus group quotes are presented with reference to respondents 1–8.

Fifth, we used the input of the focus group participants for additional (quantitative) analysis and further explored the class division quantitatively by looking into the distribution of demographic background variables (gender, living arrangement, marital status, ethnicity, educational level and age). We tested the relation between profiles and demographic variables and between the profiles and the frailty index with multinomial regression analysis (see Table 3). And we determined the scores of the frailty index, distribution of morbidities and functional limitations across the six subpopulations (see Tables 2, 4 and 5).

The Results section presents the first and second steps of the analysis separately. The results of the third, fourth and fifth steps are combined and reported by profile.

Latent class analysis with various class solutions identified six subpopulations within the population of older people. Additional file 1, Table S1  presents the model fit statistics of the various class solutions. In these different class solutions, two to three relatively big classes remained stable and the other classes became increasingly dispersed. In the eight-class solution, for example, two classes accounted for 50% of the study population and the remaining six classes were relatively small. We chose the six-class solution, based on the highest entropy score (0.81) which indicated the best quality classification.

Profile A (‘relatively healthy’) fundamentally differs from the other five profiles. Older people in this profile were relatively healthy; they indicated having good (mental) health and had very few problems across all the domains. They were not co-morbid; on average, they generally reported fewer than two morbidities and almost no functional limitations. When a functional limitation was reported, this was mostly related to household activities. The clear distinction between the relatively healthy and the other profiles is also demonstrated by differences in the background characteristics. The relatively healthy respondents are more likely to be male, younger, live independently and be married than the respondents in the other five profiles. Older people in the focus group clearly identified them as the least frail of the six profiles, “They are not frail compared to the rest, of course” (respondent 8). This is also confirmed by their mean score on the frailty index (0.10).

Profile B (‘mild physically frail’) features suffering from mild problems in the physical domain, and the focus group reinforced this characterization: “They need a lot of help. Needing help with four to five activities is quite a lot” (respondent 5). This profile reflected an initial loss of independence, particularly with regard to IADL activities. Almost all individuals required help in the household, and the vast majority needed help with travelling and shopping. Most still lived independently at home, but typically had no partner to help them with these activities. Moreover, mild physically frail people had multi-morbidity; joint damage and hearing problems were reported most frequently in this profile. Despite their functional limitations, their self-reported health and mental health were considerably good, underscoring the definition as “mild problems”. One focus group participant described the older people in this profile as follows: “The limitations are simply because of their age. But they’re not bothered by them and just go their own way” (respondent 4).

The types of problems in profile C (‘psychologically frail’) were rather different from the mild physically frail profile. Their reported health and mental health were relatively poor and social functioning was worse than in the relative healthy, mild and severe physically frail profiles. However, this profile reported only sporadic functional limitations; mostly related to problems in the household. Participants in the focus group still agreed that, despite their independence, the psychologically frail profile was rather frail. Their problems could partly be explained by their psychological condition, a relatively high percentage of people reported anxiety disorders and depression. However, the focus group also attributed the problems of this profile to their coping behaviour: “They treat every (minor) inconvenience as a major limitation or severe disease” (respondent 4). The participants of the focus group perceived the psychologically frail profile more frail than the mild physically frail profile. “People in this group are sensitive and will interpret things negatively which could lead to a self-fulfilling prophecy” (respondent 5). However, the mean score on the frailty index of this profile was lower than of the mild physically frail (0.22 respectively 0.26).

Profile D (‘severe physically frail’) was comparable to mild physically frail profile but here the physical problems were more severe and problems also expanded across other domains. On average, they had eight functional limitations, twice as many as found in the mild physically frail profile. Almost all people in the severe physically frail profile were hampered in IADL, such as taking care of the home, shopping and travelling. They needed considerably less help with the less physical IADL activities such as taking medication and handling finances. Older people in this profile also began encountering problems with ADL activities. For example, 80% said they needed help with showering. In this oldest profile, initial problems with cognitive and social functioning were prevalent. Despite their advanced age and severe limitations, they regarded their health as quite good. The mean score on the frailty index within was fairly high with 0.40. One focus group participant stated that the situation was delicate, “The moment anything goes wrong, they are in deep trouble but they’re not experiencing this yet” (respondent 7). The focus group agreed that people in the severe physically frail profile might be in denial of their frailty: “Their perception is positive even though the situation is serious” (respondent 5).

In the preceding four profiles, problems were mostly limited to one domain. However, in profile E (‘medically frail’) people accumulated problems in three domains –the physical, psychological and social – that seemed to be originated in their morbidities. People in this group mostly experienced fair or poor health in combination with a deterioration in their health compared to a year ago. Their social activities were frequently hampered by their physical condition and/or emotional problems. They experienced the worst mental health and the most morbidities of all profiles. These morbidities were psychological conditions such as depression and anxiety disorders but also physical conditions such as joint damage, dizziness with falling and heart failure. The score on the frailty index was similar to the severe physically frail profile. The older people in the focus group agreed that the older people in the medically frail profile were more frail. One of the focus group participants imagined that people with these kinds of morbidities “have physical problems that hamper them, for example in social activities, in particular compared to before” (respondent 4).

As for profile F (‘multi-frail’), in addition to problems in the physical, psychological and social domains, here people also had cognitive problems. They had the highest score on the frailty index (0.55) and also the focus group also characterized the multi-frail profile as the most frail, especially because of the cognitive problems combined with severe functional limitations. In the multi-frail profile, people had moderate to extreme cognitive problems and reported the highest percentage of dementia. On average, people in this profile needed help with 12 activities. The focus group described this state as “totally dependent” (respondent 1) and “needing help from morning to night” (respondent 8). They need help with almost all IADL and most ADL activities and almost half needed help with eating. Focus group participants felt that these older people should be institutionalized. Still, half of multi-frail profile lived independently, most often without a partner. Social functioning was hampered most frequently in this profile: “When you have these kinds of cognitive problems, I can image that you won’t undertake things yourself. I have 21 years of experience of dealing with my wife’s dementia. They don’t take the initiative, they become withdrawn” (respondent 6). The reported mental health in this profile was remarkable. As one respondent observed: “They still feel relatively good” (respondent 3). Another explained: “They’re not hampered by a sense of reality because of their cognitive problems” (respondent 5).

The main strength of this study is its strong empirical base for frailty profiles. We were able to use data from TOPICS-MDS, a large data-sharing initiative that contains data on older people from all around the Netherlands. The large sample, combined with considerations of several domains of functioning makes the current research valuable. The quantitative results were complemented with a qualitative approach, which also adds value. The focus group enabled us to further validate the profiles and to understand the older person’s perspective on frailty.

The first limitation is that even though the 54 TOPICS-MDS projects generally focused on older people at risk or frail older people, their sampling frame and inclusion criteria substantially differed. Older people were included based on functional limitations or were screened frailty instruments questionnaires such as Groningen Frailty Indicator, whereas other projects adopted an age criterion. Our study included all 54 projects and our only selection criterion was age; people 60 years and older were included. As the focus group also indicated, there is still disagreement on what is regarded as ‘old’ and 60 years might be relatively young. The literature recommends including people of 70 years and older for frailty interventions [38] but a systematic search of the literature revealed that different age criteria are adopted [18]. We decided to include everyone 60 years and older in order to also include older people with a migrant background in whom ageing begins at a younger age and who often experience worse physical and emotional health than people born in the Netherlands [39]. We expect that the relatively young sample may not have influenced our results since the effect of age on the frailty profiles was negligible. Including all people of 60 years and older might have also resulted in the rather large group of older people in the relatively healthy profile and it remains unclear whether these older people could actually be considered as being frail. They were not perceived as frail according to the participants of the focus group and their score on the frailty index was also below the general cut-off point of 0.20 [30]. Nevertheless, some older people in this relatively healthy profile were identified as being frail by the Groningen Frailty Indicator.

The second limitation was the formulation of the social functioning item that might possible have affected our results. The item was phrased as: “During the past four weeks, to what extent has your physical health or emotional problems interfered with your normal social activities with family, friends, neighbours, or groups (like visiting friends or close relatives)?” This phrasing related social functioning directly to both physical and psychological functioning and might have contributed to the absence of a separate social frailty profile.

The most important implication of our study is that we should take the heterogeneity of frailty into consideration in research, policy and practice. Future research should endeavour to further validate our six profiles. The focus group with older people was a good starting point but the profiles could also be validated by professionals and policy makers. Our identification effort should also be replicated on other databases of frail older people and with other measurement instruments, for example for social functioning. Testing the validity of these profiles would also involve examining frailty trajectories. In other words, our cross-sectional latent class analysis could be complemented with a latent transition analysis [34] which could account for the dynamic and progressive character of frailty [9]. In this regard, it would be beneficial to explore whether mild physically frail profile eventually transfer to the severe physically frail profile or how the trajectory of relatively healthy people progresses.

Our research also has implications for selecting the appropriate target groups for care interventions. The psycho-social domains of frailty are deemed important and focusing on the physical domain of frailty and functional limitations by professionals, researchers and policy makers could be too restricted. Our research showed that a relatively large group suffers from problems in the psychological domain without having problems in the physical domains in terms of functional limitations. It is important to target this profile for care interventions. Instruments with a broad perspective including the psycho-social domains such as frailty indexes [16, 30] and the Groningen Frailty Indicator [40] sum the total numbers of health problems which implies that they do not differentiate between the types of underlying problems or weigh the different domains. Older people with the same score on the frailty index or Groningen Frailty Indicator could have different underlying problems and frailty profiles [17]. The severe physically frail had a similar frailty index as the medically frail but their psychological and social functioning was not hampered. The medically frail profile showed that their problems related to morbidities extended to severe problems in the psychological and social domains. These problems should be also be acknowledged by health care professionals who are originally trained to have a disease-specific approach (Lette et al., 2015). In other words, the balance between deficits and assets in relation to frailty should be further explored in practice, policy and research. Assets could be coping style, resilience [41] or resources such as older people’s social network [3], which should be considered in the conceptualization and measurement of frailty.

Lastly, the six frailty profiles could be used to develop tailor-made care interventions for each profile rather than producing one-size-fits-all care. The heterogeneity of frailty should be incorporated in the evaluation of these interventions. Currently, traditional evaluation research is not acknowledging this heterogeneity in, for example, integrated care, which is misaligned with its goal to provide person-centred care with a holistic view of the individual [42]. So far, the effects of integrated care on health outcomes is limited [18]. Concurrently, the more heterogeneous a population is, the harder it is to achieve effectiveness [38, 43]. A possible explanation for the limited effectiveness of integrated care could be that the care professionals involved – and particularly researchers conducting the evaluation research – generally perceive frail older people as a homogeneous group. Evaluation research on integrated care could be replicated by incorporating the frailty profiles to gain deeper insight into the effectiveness of integrated care interventions. It would be beneficial to explore whether integrated is (more) effective on specific outcomes for each of the six profiles separately. Future research should explore, for example, whether integrated care is more effective in terms of mental health for the psychological frail than for the mild physical profile.