Background

Frailty is defined as an age-associated, biological syndrome characterized by decreased physiological reserves, which puts an individual at risk when facing minor stressors. Because of its powerful association with adverse events (i.e. disability, mortality, and hospitalization)1 and its high prevalence in older adults,2 frailty has been placed in the geriatric medicine spotlight.1

Frailty is dynamic in nature and potentially reversible.3, 4 However, to date, only a limited number of observational studies have examined how changes in frailty status (i.e. trajectories of frailty) can predict usual adverse events. The categorical structure of the classical tools for assessing the frailty status5 may partially account for the short number of studies assessing such relationship. In fact, most of the previous researches have not considered frailty as a continuum4, 6 or are limited by their analytical approach, ignoring the variability or reversibility in frailty score trajectories.7, 8 Nevertheless, our group has recently developed the so-named Frailty Trait Scale 5 (FTS5),8 a Short Form of the 12 items one7 that allows the continuous evaluation of frailty levels, potentially overcoming several of previous frailty assessment pitfalls. This instrument refines risk profiling of the individual and outperforms the predictive ability of adverse events (i.e. mortality, hospitalization, and disability)8 even better than classical frailty tools, as the two most commonly used tools,9 the Frailty Phenotype5 and Frailty Index.10

The aim of this study is to explore the existence of different frailty trajectories and to investigate their associations with adverse outcomes (disability, hospitalization, and mortality) in a representative sample of older adults. Our hypothesis is that prospective decreasing frailty levels along ageing are associated with a lower risk of adverse events, compared with developing or maintaining higher frailty levels. A secondary objective is to compare cross-sectionally and longitudinally assessed frailty levels in terms of predictive ability.

Methods

This is a prospective analysis of the Toledo Study of Healthy Ageing (TSHA), a population-based longitudinal study whose main aim is to explore the prevalence, determinants, and impact of frailty in a Spanish cohort of older adults (>65 years).11

In the present analysis, we included subjects with frailty and covariates data at TSHA Waves 1 (2006–2009) and 2 (2011–2013) and outcomes data at Wave 3 (2015–2017) or time-censoring, depending on the outcome of interest. Wave 2 occurred 5.04 (range 2.32–6.84) median years after the first wave. Wave 3 visit was performed 2.99 (range 2–5.4) median years after the Wave 2.

TSHA protocol was approved by the Clinical Research Ethics Committee of Toledo Hospital Complex, which was conformed according to the ethical standards defined in the 1964 Declaration of Helsinki. As detailed elsewhere previously,11 participants signed an informed consent form prior to inclusion.

Frailty Trait Scale 5 Frailty was assessed through the FTS58 that evaluates five core aspects of frailty: Physical activity was determined using the Physical Activity Scale for the Elderly scale. Gait speed was defined using the 3 m walking test at their usual pace, according to the standard protocol. The best time of two performances was chosen. At least, 1 min of resting was given between attempts. Hand grip strength was measured using JAMAR Hydraulic Hand Dynamometer (Sammons Preston Rolyan, Bolingbrook, IL). The best peak strength of three performances was selected and gathered using international standard procedures. Between performances, at least 1 min of resting was permitted. Body mass index was measured according to the standard procedure (weight/height2). Progressive Romberg test was established according to the standing balance feet together, semi-tandem, and tandem position. Each domain score ranges between 0 and 10, being 0 the best score and 10 the worst. FTS5 items are summed up to obtain a final score between 0 and 50, being 0 the lowest and 50 the highest frailty score. FTS5 scoring process is displayed in Supporting Information, Appendix S1. Adverse outcomes Mortality

Vital status and death dates were ascertained through the Spanish National Death Index (Ministry of Health, Consumer Affairs and Social Welfare). Participants were followed-up to March 2019 or death. Median follow-up for mortality was 6.76 (range 0.26–7.50) years.

Hospitalization

Hospitalization was defined as the first admission according to the records of the Toledo Hospital Complex. Participants' hospitalizations were followed-up to December 2016 with a median follow-up of 4.18 (range 0.02–5.25) years.

Incident or worsening disability

Basic activities of daily living were evaluated by the Katz Index.12 Transitions of a score of 6 to 5 or less in the Katz Index at follow-up were considered as incident disability. Worsening disability was defined as a loss of 1 point or more during the same time-period. Median follow-up for incident and worsening disability was 2.99 (range 2–5.4) years.

Covariates

Covariates were selected according to previous research and biological plausibility for confounding effects. Age, gender, and educational level (no school, primary school incomplete, and primary school complete or superior) were registered during visits. Comorbidities were measured using the Charlson Index score.13 The number of prescription and non-prescription drugs within the Anatomical Therapeutic Chemical Classification System taken by the participant was calculated. Polypharmacy was defined as the use of five or more drugs per day. Cognitive status was evaluated using the Mini-Mental State Examination.14 Covariates data were assessed at Wave 2 visit.

Trajectory modelling

We used GBTM to identify distinct patterns of evolution in frailty levels according to the baseline FTS5 score and changes in the FTS5 score between Waves 1 and 2 visits. This mixture analysis uses semi-parametric empirical models and tries to identify relatively homogeneous clusters of participants following similar longitudinal patterns of evolution and estimate their shape and direction parameters within a continuous distribution. GBTM assumes that the sample is composed of distinct subpopulations that are not identifiable based on measured characteristics ex ante.15, 16

We compared models defining two with 10 trajectory groups to find the optimal number of trajectories, based on values from the Bayesian Information Criterion (BIC). BIC is an index used in Bayesian statistics to choose between different models. Two times the change in the BIC between models greater than 10 was used as indicative of better fit in order to compare more complex—with a greater number of trajectories—or more parsimonious—with a lower number of trajectories—models.17 Every subject was assigned to a trajectory depending on their FTS5 score and changes in FTS5 score. The reliability of the final model classification was evaluated using mean posterior probability of membership, which represents their probability of belonging to the group they was assigned to by previous grouping based on their individual features. Trajectories mean posterior probability of membership indicates its internal consistency. A higher value indicates a better classification quality.15 At last, we checked the number of subjects within each trajectory to confirm adequate sample size for analysing the ensuing risk of adverse outcomes.

Statistical analysis

Descriptive statistics were shown as mean (standard deviation) and frequency (%) for continuous and categorical variables, respectively. Descriptive features of subjects in the different frailty trajectories were compared through the analysis of variance test for continuous variables and χ2 tests for categorical variables.

Cox proportional hazard regression models were used to explore associations between frailty trajectories and time-censored adverse outcomes (mortality and hospitalization), and logistic regressions were used for incident and worsening disability.

We further explored associations between final FTS5 scores in Wave 2 and adverse events. We studied adverse events risks according to the FTS5 score as the change of one point (continuous) and categorized into five categories: ≤10, 11–15, 16–20, 21–25, and >25.

We used the following set of models: Model 1 was the raw model. Model 2 was adjusted by age and gender. In Model 3, we added the MMSE. In Model 4, we added the Charlson Index, polypharmacy, and Katz Index. Finally, we estimated Model 5 including the educational level.

All analyses were performed using the Statistical Package R Version 3.6.1 for Windows (Vienna, Austria). Statistical significance was set at P value < 0.05.

Results

Participant baseline characteristics are shown in Table 1, and 975 subjects (mean age 73.14 ± 4.69; 43.38% men) were included.

Table 1. Demographic characteristics of the sample according to the FTS5 trajectory Variable All WFN INF DF RF IF P value N (%) 975 226 (23.17) 353 (36.20) 127 (13.03) 203 (20.82) 66 (6.76) Age, mean (SD) 73.14 (4.69) 71.20 (3.52) 72.48 (4.50) 73.72 (4.75) 75.12 (4.92) 76.05 (4.64) <0.001 Gender, men (%) 423 (43.38) 139 (61.50) 169 (47.88) 55 (43.31) 50 (24.63) 10 (15.15) <0.001 FTS5 basal, mean (SD) 18.08 (6.79) 9.44 (2.48) 19.01 (4.34) 16.02 (2.06) 24.67 (4.23) 26.39 (5.12) <0.001 Change in FTS5 per year, mean (SD) −0.01 (1.56) 0.80 (1.34) −1.23 (0.81) 1.61 (1.80) −0.25 (0.82) 1.37 (0.86) <0.001 Dependent according Katz Index score ≤ 5, n (%) 124 (12.72) 11 (4.87) 38 (10.76) 12 (9.45) 41 (20.20) 22 (33.33) <0.001 MMSE, mean (SD) 24.44 (4.39) 26.02 (3.34) 24.68 (4.40) 24.51 (3.81) 22.85 (4.72) 22.22 (5.27) <0.001 Charlson Index, mean (SD) 0.94 (1.43) 0.78 (1.28) 0.83 (1.37) 1.27 (1.60) 1.00 (1.45) 1.29 (1.64) 0.003 Educational level No school, n (%) 647 (66.70) 113 (50.22) 251 (71.71) 91 (71.65) 139 (68.81) 53 (80.30) <0.001 Primary school incomplete, n (%) 162 (16.70) 54 (24.00) 36 (10.29) 24 (18.90) 37 (18.32) 11 (16.67) <0.001 Primary school complete or superior, n (%) 161 (16.60) 58 (25.78) 63 (18.00) 12 (9.45) 26 (12.87) 2 (3.03) <0.001 Number of drugs per day, mean (SD) 4.06 (2.72) 3.40 (2.59) 3.46 (2.55) 4.72 (2.61) 4.84 (2.62) 5.88 (2.97) <0.001 Polypharmacy, n (%) 391 (40.10) 78 (34.51) 104 (29.46) 61 (48.03) 104 (51.23) 44 (66.67) <0.001 Death, n (%) 211 (21.64) 31 (13.72) 61 (17.28) 38 (29.92) 56 (27.59) 25 (37.88) <0.001 Hospitalization, n (%) 393 (40.31) 83 (36.73) 124 (35.13) 48 (37.80) 100 (49.26) 38 (57.58) 0.007 Incident disability, n (%) 183 (29.66) 31 (18.90) 64 (26.23) 25 (33.33) 50 (44.64) 13 (59.09) <0.001 Worsening disability, n (%) 204 (26.56) 31 (16.76) 69 (23.71) 29 (29.90) 56 (37.33) 19 (42.22) <0.001 DF, developing frailty; FTS5: Frailty Trait Scale 5; IF, increasing frailty; INF, improving to non-frailty; MMSE, Mini-Mental State Examination; RF, remaining frail; WNF, worsening from non-frailty. In bold: P < 0.05. Mean (SD): continuous variables. N, %: categorical variable. Frailty Trait Scale 5 trajectories and adverse outcomes

The GBTM yielded a five-frailty trajectory model as the best fit to the data: worsening from non-frailty (WNF) (226, 23.17%), improving to non-frailty (INF) (353, 36.20%), developing frailty (DF) (127, 13.03%), remaining frail (203, 20.82%), and increasing frailty (IF) (66, 6.76%) (Figure 1). Mean posterior probabilities of membership, an index of quality classification, ranged from 0.74 (±0.14) to 0.77 (±0.18) indicating a moderate fit for this model.

Trajectories according Waves 1 and 2 FTS5 score and changes in FTS5 score. DF, developing frailty; FTS5, Frailty Trait Scale 5; IF, increasing frailty; INF, improving to non-frailty; RF, remaining frail; WNF, worsening from non-frailty.

In the fully adjusted models, when comparing with WNF, subjects pertaining to trajectories showing both an increase and maintenance in frailty scores showed an increased risk for developing adverse outcomes compared with those showing a decreasing evolution, except for hospitalization. In this latter outcome, only those maintaining or increasing their frailty score, but not those who developed it during the time of the trajectory, DF, showed an increased risk (Table 2). By opposite, improving the frailty score according to FTS5—INF—was associated with a low risk for developing any of the adverse outcomes, similar to the reference risk, WNF, which denoted the risk of people that remained robust along the follow-up. We did not find any particular additional risk in any of the categories of high risk, not allowing to ascribe a particular higher risk to any of the three trajectories with a poor prognosis.

Table 2. Multivariate regression models showing the association between FTS5 trajectories on different adverse events Reference (WFN) Model 1 Model 2 Model 3 Model 4 Model 5 HR (95% CI) P value HR (95% CI) P value HR (95% CI) P value HR (95% CI) P value HR (95% CI) P value 1.00 — 1.00 — 1.00 — 1.00 — 1.00 — Death INF 1.28 (0.83, 1.97) 0.267 1.13 (0.73, 1.75) 0.587 1.12 (0.72, 1.74) 0.606 1.08 (0.69, 1.68) 0.735 1.11 (0.71, 1.73) 0.654 DF 2.42 (1.51, 3.90) <0.001 2.04 (1.24, 3.33) 0.005 2.01 (1.23, 3.29) 0.006 1.94 (1.18, 3.19) 0.009 2.01 (1.21, 3.32) 0.007 RF 2.25 (1.45, 3.49) <0.001 1.99 (1.24, 3.20) 0.004 1.95 (1.21, 3.14) 0.006 1.88 (1.16, 3.05) 0.010 1.92 (1.18, 3.12) 0.008 IF 3.18 (1.88, 5.38) <0.001 2.86 (1.62, 5.04) <0.001 2.76 (1.55, 4.89) <0.001 2.55 (1.42, 4.58) 0.002 2.67 (1.48, 4.81) 0.001 Hospitalization INF 0.94 (0.73, 1.22) 0.663 0.95 (0.73, 1.23) 0.680 0.94 (0.72, 1.22) 0.657 0.95 (0.73, 1.24) 0.720 0.94 (0.72, 1.22) 0.637 DF 1.12 (0.81, 1.56) 0.496 1.11 (0.79, 1.55) 0.553 1.10 (0.79, 1.54) 0.576 1.06 (0.75, 1.49) 0.743 1.04 (0.74, 1.47) 0.801 RF 1.57 (1.20, 2.06) <0.001 1.65 (1.23, 2.22) <0.001 1.63 (1.21, 2.19) 0.001 1.61 (1.19, 2.16) 0.002 1.58 (1.17, 2.14) 0.003 IF 1.92 (1.35, 2.75) <0.001 2.03 (1.38, 2.98) <0.001 1.98 (1.34, 2.92) <0.001 1.92 (1.29, 2.85) 0.001 1.89 (1.27, 2.82) 0.002 Worsening disability INF 1.60 (1.03, 2.50) 0.038 1.41 (0.90, 2.22) 0.138 1.40 (0.89, 2.20) 0.149 1.38 (0.87, 2.18) 0.173 1.53 (0.96, 2.45) 0.075 DF 2.42 (1.41, 4.15) 0.001 1.94 (1.11, 3.39) 0.020 1.93 (1.10, 3.37) 0.021 1.87 (1.06, 3.29) 0.030 2.11 (1.19, 3.76) 0.011 RF 3.06 (1.90, 4.94) <0.001 2.27 (1.35, 3.80) 0.002 2.18 (1.30, 3.67) 0.003 2.01 (1.18, 3.40) 0.010 2.14 (1.26, 3.64) 0.005 IF 3.81 (1.97, 7.40) <0.001 2.75 (1.36, 5.54) 0.005 2.60 (1.28, 5.27) 0.008 1.96 (0.95, 4.06) 0.070 2.21 (1.06, 4.62) 0.035 Incident disability INF 1.58 (1.00, 2.50) 0.050 1.43 (0.90, 2.28) 0.130 1.41 (0.88, 2.24) 0.152 1.40 (0.87, 2.25) 0.163 1.50 (0.92, 2.43) 0.103 DF 2.35 (1.31, 4.19) 0.004 1.96 (1.09, 3.56) 0.026 1.92 (1.06, 3.48) 0.032 1.87 (1.02, 3.43) 0.043 2.06 (1.11, 3.82) 0.021 RF 3.51 (2.11, 5.83) <0.001 2.50 (1.45, 4.34) 0.001 2.40 (1.38, 4.17) 0.002 2.20 (1.25, 3.87) 0.006 2.29 (1.30, 4.03) 0.004 IF 6.51 (2.70, 15.68) <0.001 4.49 (1.80, 11.19) 0.001 4.19 (1.67, 10.51) 0.002 3.22 (1.25, 8.32) 0.015 3.55 (1.37, 9.24) 0.009 95% CI, 95% confidence interval; DF, developing frailty; FTS5, Frailty Trait Scale 5; HR, hazard ratio; IF, increasing frailty; INF, improving to non-frailty; OR, odds ratio; RF, remaining frail; WNF, worsening from non-frailty. In bold: P value < 0.05. Model 1: raw model. Model 2: Model 1 plus age and gender. Model 3: Model 2 plus MMSE. Model 4: Model 3 plus Charlson Index, Polypharmacy, and Katz Index. Model 5: Model 4 plus educational level. Intermediate trajectories comparison and adverse outcomes

To study if similar baseline scores but with different evolution were associated with different risk, we repeated the analysis taking as reference each different trajectory.

When comparing with INF, subjects who were grouped in the DF had an increased risk of mortality of 1.81 [95% confidence interval (CI): 1.20–2.75] in the fully adjusted model, although not for the other adverse events. MF and IF trajectories had also a significant higher risk for death and hospitalization when INF was used as the reference category.

When DF was the reference, we found an increased risk for hospitalization for those subjects if belonging to MF (hazard ratio = 1.51; 95% CI = 1.10–2.08) or IF (HR = 1.81; 95% CI = 1.21–2.72).

Cross-sectionally assessed frailty and adverse outcomes

When we cross-sectionally assessed the association between frailty through FTS5 scores and the adverse outcomes, FTS5 showed a continuously increasing risk for developing any of the adverse outcomes (participant baseline characteristics according FTS5 score are shown in Table S1). This finding was consistent irrespective of the study wave at which the FTS5 score was selected as the exposure. When we split the score in FTS5 in five categories, we found a significant increased risk of death, incident, and worsening disability beyond a FTS5 ≥ 20, compared with FTS5 ≤ 10. In the case of hospitalization, the incremental risk was significant for values higher than 10, compared with 10 or lower, even after adjusting for potential confounders (Table 3).

Table 3. Multivariate regression models showing the association between FTS5 score (continuous or categorized) at Wave 2 on different adverse events Model 1 Model 2 Model 3 Model 4 Model 5 Death HR (95% CI) P value HR (95% CI) P value HR (95% CI) P value HR (95% CI) P value HR (95% CI) P value FTS5 change in 1 point (continuous) 1.05 (1.03, 1.07) <0.001 1.05 (1.03, 1.07) <0.001 1.05 (1.03, 1.07) <0.001 1.05 (1.02, 1.07) <0.001 1.05 (1.03, 1.07) <0.001 Reference (FTS5 ≤ 10) 1.00 — 1.00 — 1.00 — 1.00 — 1.00 — FTS5 > 10 and ≤15 1.49 (0.90, 2.46) 0.121 1.25 (0.75, 2.08) 0.398 1.25 (0.75, 2.09) 0.389 1.22 (0.73, 2.04) 0.442