Transitions in robust and prefrail octogenarians after 1 year: the influence of activities of daily living, social participation, and psychological resilience on the frailty state

Since 2015, 494 participants were recruited. The final longitudinal baseline sample (T0) consisted of 403 eligible participants (see Fig. 1). After 6 months (T1) and one year of follow-up (T2), participants were screened again for their frailty status. In the end, data of 322 participants could be used of this follow-up sample. Due to dropout between T0-T1 (n = 31), dropout between T1-T2 (n = 18), missing T2 assessments (n = 31), and missing data of the FFP (n = 1), ADLs, social participation, and psychological resilience (n = 28), 109 participants were excluded for this analysis. Also, an estimation of the FFP of 28 subjects, who were not tested but who had enough indirect information, could be made in consultation with the team of three persons and the study’s personal investigator.

The frailty states on T0 and T2 were identified based on the adapted version of the FFP, in accordance with previous results from the BUTTERFLY study using exhaustion, gait speed, grip strength and weight loss [2, 10]. A score of 0 indicated ‘robustness’, 1 or 2 signified ‘pre-frailty’, and a score of 3 or 4 identified ‘frailty’ [25]. Specifically, for longitudinal analyses, weight loss was evaluated in accordance with Stenholm, Ferrucci [26] and Theou, Cann [27] and defined as a weight loss of greater than 5% or ≥ 4.5kg compared to baseline.

As for any variable, the score on baseline (T0) and the delta score with the measurement after 6 months (T1) was used. Delta-values were calculated for all exploratory variables: values of 6 months minus the values of baseline (T1-T0). For this purpose, the difference between T0 and T1 (after 6 months) was examined to see if the variable can predict a change in the variable.

Firstly, descriptive statistics were presented by percentages and frequencies for categorical variables and means with standard deviation for continuous variables. The sample characteristics and differences within each group (improved and worsened group) were tested using Chi-squared test for categorical variables and independent samples T test for continuous variables. Fisher’s exact tests were performed for categorical variables when there were not enough variables in the cells. P-values were adjusted with the Benjamini–Hochberg procedure for False Discovery Rate detection.

Then, for observing relationships with the outcome frailty status of the participant, the multivariate modelling phase existed out of two different model: (a) one modelling the improved frailty transition, and (b) one modelling the worsened frailty transition. Both models contained an outcome variable that was either positive or negative for respectively model (a) and (b). In both analyses, we applied a binary logistic regression modelling based on the likelihood of the models by means of the Akaike Information Criterion (AIC) using a stepwise approach into both directions. The modelling started by defining an upper full model and a lower partial model. The lower partial model only contained an intercept and the three baseline ADLs (b-ADLs, i-ADLs and a-ADLs). The upper model contained the demographic variables (age and sex), all ADLs parameters (b-ADLs, i-ADLs and a-ADLs), all social participation variables (total participation score, being a member, total number of memberships, level of social participation, being a board member, volunteering, and formal participation), and psychological resilience. Other co-factors (education, MMSE and living circumstances) will be added to the model if significant differences between groups are found at baseline. For the ADLs, social participation variables, and psychological resilience in the upper model both the measure at time T0 and the delta score (T1-T0) were entered into the model. In this way, the association between the delta score of the exposure on the outcome, adjusted for its baseline value is evaluated. By using this difference, we partially removed the collinearity between the two variables. In the stepwise variable selection procedure only variables were retained that resulted in a better log-likelihood of the model by starting from the lower model and working towards the direction of the full model by combining both forward and backward regression modelling steps. A stepwise approach in both directions (forward and backwards modelling) has the benefit that variables that enter a model in the initial phase can later be considered non-significant since they correlate with variables that have entered the model in a later phase. After the model building was performed, we tested for overdispersion and if necessary, applied the quasibinomial family of distributions to the model to obtain more accurate estimates of the variance. Eventually, also multicollinearity was tested by calculating variance inflation factors (VIFs) and if necessary, leave highly correlated variables out of the final model. P-values were obtained by performing the Wald test for significance. The significance level was set at α = 0.05. Statistical analyses were performed using the statistical software RStudio version 1.1.463 running on R version 3.5.3.

All methods were carried out in accordance with relevant guidelines and regulations. The BUTTERFLY-study was approved by the ethical committee of UZ Brussel (B.U.N. 143201421976). All methods were performed in accordance with the Declarations of Helsinki and all subjects provided informed consent.

As Fig. 1 shows, from the total sample of 322 participants, two frailty transitions emerged: a worsened (n = 61) and an improved frailty transition (n = 28). In the worsened frailty transition, persons could change to a prefrail (n = 52) or frail (n = 9) state. This group will be compared to persons who remained robust after 1 year (n = 111). For the improved frailty transition, changes occurred from prefrail participants at baseline to robustness after 1 year (n = 28), they were compared with persons who remained prefrail (n = 122).

The sample consisted of 58% men and 42% women. In this sample a mean age of 83.04 ± 2.78 years and a mean score of 27.82 ± 1.72 on the MMSE was observed. No differences between groups were found for baseline characteristics (see Table 1). Also no differences were found between the included participants, dropouts/persons who missed a testing and the excluded persons with missing data for the baseline characteristics (all p > 0.05).

Table 2 shows the difference within the worsened and improved frailty transitions regarding ADLs, social participation, and psychological resilience. Overall, in terms of ADLs, social participations, and psychological resilience, no differences were found. Figure 2 shows the comparison between the worsened and improved frailty transition of the b-, i-, and a-ADL-DI of baseline and after 6 months. Overall, more limitations occurred in a-ADL-DI, followed by i-ADL-DI, and b-ADL-DI. However, the changes were statistically insignificant.

For the worsened frailty transition, binary logistic regression indicated that the baseline a-ADL-DI (odds ratio [OR] = 1.048;95% CI = [1.010,1.090]) and the difference in a-ADLs after 6 months (OR = 1.044;95% CI = [1.007,1.085]) are predictors for a worsened frailty transition after 1 year. Or in other words, having more deficits at baseline and an increase in deficits within 6 months in a-ADLs respectively increases the odds of physical prefrailty or frailty with 4.8% and 4.4% per additional point (see Table 3).

Regarding the improved frailty transition, being a woman (OR = 3.682;95% CI = [1.379,10.139]) and becoming a board member (OR = 4.343;95% CI = [1.082,16.437]) significantly was related to a greater odds of recovering to robustness after 1 year (see Table 3).

To our knowledge, the a-ADLs are relatively understudied in research and clinical practice. Therefore, it is quite innovating that limitations in a-ADLs and changes over time in those complex ADLs are found as potential predictors of prefrailty or frailty. Fortunately, this gives new opportunities for the development of interventions to postpone and/or reverse a frailty state and its negative health outcomes.

The results could be anticipated, as they are in line with the concept of the ‘functional continuum’. This concept states that there is a hierarchical decline in a person’s functional ability over time in a certain order. The b-ADLs and i-ADLs exist on a continuum [33], where the i-ADLs [15], decline first, and are then followed by limitations in b-ADLs, with the latter mostly performed on routine [14]. So, the more complex activities are, the more skills, including motor, cognitive and physical abilities, are needed to perform them. Therefore, the a-ADLs, the activities needing the most skills, could be the very first ones to decline [34]. Fieo, Zahodne [35] already referred to this as ‘an area of the functional continuum beyond i-ADLs competencies’. In this community-dwelling octogenarians sample, most limitations were found for a-ADLs, followed by i-, and b-ADLs (see Fig. 1), which might confirm the functional continuum. At first sight, more limitations evolved for b- and i-ADLs in 6 months. For a-ADLs, an overall improvement developed in the total sample, however when looking more closely, solely in the worsened frailty score group limitations in a-ADLs deteriorated. The relationship between a-ADLs and frailty has been examined to a limited extent. In our previous studies, an association between frailty and limitations in a-ADLs was found [10, 11]. Besides our own studies, Pegorari and Tavares [36], found that a better performance of a-ADLs resulted in a protective effect against frailty. The opposite (a decline in a-ADLs being an early indicator of frailty worsening) could also be stated. However, their definition of a-ADLs was mostly related to activities with social interactions, such as work, participation in community groups, and meetings, which comes close to our definition of social participation. Our definition of daily activities was broad, extensive, and based on all activities a person could perform [32, 37] and thus included more activities such as personal hobbies and gardening.

Up to now, most studies examined the negative health outcomes of frailty by including frail participants to follow up [5]. We excluded frail individuals at T0. Our sample of worsened frailty is therefore unique to investigate the early predictors of frailty. Most participants emerged to a prefrail state and longer follow-up could be needed to potentially develop frailty. This could be the reason why limitations in b-, and i-ADLs were not significant in the results. This in contrast to Rodriguez-Laso, Garcia-Garcia [17] who showed that transitions from robustness to prefrailty and vice-versa were associated with less/more autonomy in i-ADLs. Also, it was already found that the separate components of frailty, such as decreasing walking speed, increased the odds of more than 2 limitations in i-ADLs (OR:4.2). In two recent functional concepts, the frailty state and limitations in ADLs in older adults were combined, though limited to b- and i-ADLs. In both concepts, limitations in b-ADLs appeared as negative health outcomes of frailty and thus developed after the occurrence of frailty. Regarding the i-ADLs, Hoogendijk, Romero [38] categorised impairment in i-ADLs solely without frailty and as a characteristic of prefrailty. Though, Zamudio-Rodriguez, Letenneur [39] made no differentiation between prefrailty and frailty, but assigned limitations in i-ADLs (without frailty) before a frailty state with limitation in i-ADLs. Our study results could contribute to this intertwined relationship of ADLs and frailty by adding the a-ADLs. However, the causality pathway is still not certain, and this should be further clarified. Previous limited research, already declared that ADLs could be useful to monitor response to potential disease-modifying therapies [35] and a useful tool in the diagnostics of mild cognitive impairment [30].

Being a woman and becoming a board member after six months led to an improved frailty state from prefrailty to robustness.

Quite some discussion about the sex differences in frailty exist. There is a difference between the associated variables of frailty and factors that influence the frailty transitions. A systematic review of Ho, Cheung [6], showed that being a woman also was a protective factor in short follow-up intervals of 2–3 years, which was in line with our results. However, in intermediate follow-up interval of 4 to 6 years, either being a man [40] or being a woman [4], were identified as risk factors (M_age73-74 years). Moreover, it is recognized that women have longer lifespans with greater levels of comorbidity and thus are frailer [41] but tolerate the condition better than men. Women can therefore be seen as both frailer but also less frail compared to men, respectively due to their poorer health status and longer lifespan, also known as the ‘male–female health-survival paradox’ by Gordon, Peel [42].

Next, social participation, especially being a board member, was a potential protective factor of frailty (with the remark that the group changes were small); however, we expected to find more significant explaining variables in our investigation. Mehrabi and Beland [43] highlighted that few studies and evidence exist regarding the relationship between social network, social participation, social support, its adverse outcomes, and its effect on frailty. Two longitudinal studies already found similar results and identified social participation as a protective factor to frailty [44, 45]. Though, it is clear that the psychological and social factor associations are understudied in frailty transitions [6].

As mentioned before, it could be presumed that both social participation and psychological resilience might provide added value in targeting frailty, both in a worsened or improved transition. This was of no avail. Social participation came solely forward as a potential protector, although, other research already found that a diminishment of social participation was related to frailty worsening [44, 45]. Participation was already related to the frailty state [19]. Our results are not conclusive and could be explained by how social participation was measured. Social participation was identified using participation indicators, which could be far more investigated in depth by including all the environments and the frequency of participation [46].

In addition, it was unexpected that no results came forward for psychological resilience since other limited studies showed otherwise. In hemodialysis patients, psychological resilience had a protective effect on frailty [21] and low resilience was strongly associated with frailty in patients with cirrhosis [22]. Moreover, in nursing homes, resilience and social support showed to have a protective role against frailty [47]. Though, the aforementioned studies were not using samples of participants still living at home (opposed to our study sample), psychological resilience might still be promising in frailty. Our results could be explained by the follow-up period of both frailty and psychological resilience. A change of frailty status after one year is short compared to other already found follow-up intervals [6]. Also, most changes were found from robust to prefrailty, which also might lead to less pronounced results. Moreover, a change of psychological resilience after six months might not be responsive enough. Our results barely showed a difference in CD-RISC scores between T0 and T1. The group that remained robust was the only group that improved on the CD-RISC, while the other groups deteriorated, however not significant. Next, the two trends in the operationalization of psychological resilience, a dynamic versus trait-oriented approach, might also give reason for the found results. Psychological resilience was assessed in this study by the trait-oriented approach using the most reliable measurement tool, the CD-RISC, according to Windle, Bennett [48]. However, in further longitudinal research it is advisable to consider the stressfulness of the event, using the dynamic process approach, when investigating psychological resilience [49].

This study was unique due to the investigated group of octogenarians because being 80 years and over is an important predictor for a prefrail or frail state [36]. The octogenarians are mostly understudied as a separate group [50]. Moreover, this paper describes one of the first prospective studies about the relationship between b-, i-, and specifically a-ADLs with the frailty transitions. In addition, limitations in a-ADLs, social participation, and psychological resilience were longitudinally researched by not only including baseline data but also the changes after six months (including delta-values). Finally, by choosing physically frailty as longitudinal outcome circular reasoning was avoided. Especially since before, psychosocial and functional factors have been studied as predictors, characteristics and outcomes of frailty. In this study, there was little overlap between exposure (functional, social and psychological health) and outcome (physical frailty status), allowing to make some prudent statements about predictive and protective effects. However, possibly, when focusing on multidimensional frailty more significant results might have been found but this might also have led to unclarity regarding dependent and independent variables. Nonetheless, some remarks should be made about the limitations of this study. Since this study explicitly aimed to include robust octogenarians, the exclusion of participants at baseline due to frailty (18%) was relatively lower than the estimated prevalence of frailty in a meta-analysis of population-level studies. They found a prevalence of 31% in octogenarians [51]. It also has to be noted that the choice of the frailty instrument also influences the prevalence as well as the characteristics of the selected population [52]. On top, this study used the adapted version of the Fried frailty scale and left out the physical activity component. This approach could have underestimated the prevalence of pre-frailty, however research showed that low physical activity was one of the last items on which participants score positive in the frailty index [26]. Although this longitudinal study gives some indication of the temporal relationship between exposures and outcomes, it must be acknowledged that frailty is a dynamic state and results must be confirmed with longer follow up periods. It is important to monitor further frailty transitions, in the future it will be possible to extend our analysis to 2 year of follow up. However, the complex and bi-way relationship between the exposures and frailty state must be investigated further. As mentioned before, social participation could be investigated more thoroughly, not only using participation indicators. Also, psychological resilience was investigated by the trait-oriented approach and therefore a change after six months in resilience might not be responsive enough.