Absence of association between frailty index and survival in elderly Brazilians : the FIBRA Study

In Brazil, the frailty index has not been evaluated previously for its capacity to predict mortality in community-dwelling elderly. The objective of the current study was to evaluate the association between frailty index and mortality in the elderly. This was a prospective study consisting of data from the FIBRA Network-2008-2009 in Campinas, São Paulo State, with information on community-dwelling older adults from the urban area and through the Mortality Information System. Comparisons and statistical associations were performed with the following tests: Mann-Whitney, Kruskal-Wallis, chi-square, and Cox regression with 95% confidence intervals. A total of 689 older adults participated (72.1 ± 5.3 years), of whom 68.8% were women. The prevalence rate for frailty was 38.8%, compared to 51.6% for pre-frailty and 9.6% for fit elders; overall mean frailty index was higher in women. There was no association between frailty index and chronological age. Cox regression showed that the variables age (HR: 1.10; 95%CI: 1.05-1.15) and gender (HR: 0.57; 95%CI: 0.33-0.99) were significantly associated with mortality. No association was found between frailty index and mortality (HR: 3.02; 95%CI: 0.24-37.64). Frailty index was not capable of predicting mortality in community-dwelling elderly Brazilians. Aged; Frail Elderly; Health of the Elderly; Mortality Correspondence F. S. A. Borim Faculdade de Ciências Médicas, Universidade Estadual de Campinas. Av. Tessalia Vieira de Camargo 126, Campinas, SP 13083-887, Brasil. flarbex@hotmail.com 1 Faculdade de Ciências Médicas, Universidade Estadual de Campinas, Campinas, Brasil. ARTIGO ARTICLE doi: 10.1590/0102-311X00194115 Cad. Saúde Pública 2017; 33(5):e00194115 This article is published in Open Access under the Creative Commons Attribution license, which allows use, distribution, and reproduction in any medium, without restrictions, as long as the original work is correctly cited.


Introduction
Frailty is defined as a clinical syndrome with increased vulnerability to various types of internal and external stressors.The syndrome reflects a decline in energy reserves, inherent to physiological aging, which can be aggravated by current and life-course biological and environmental variables 1,2,3,4 .It involves a set of characteristics and genetic and environmental determinants that distinguish individuals in a cohort 4,5,6 .There is a consensus among researchers concerning the notion of increased vulnerability, heterogeneity, and multidimensionality associated with frailty 4,7,8,9,10 .In clinical terms, it includes risks of adverse events such as falls, reduced mobility, loss of independence, hospitalization, disability, and death.Among these outcomes, mortality, functional disability, and institutionalization are the most common findings in the literature 6,11 .
There are different ways of operationalizing the frailty phenomenon 3,4,5,6,9 .According to a literature review by Boiullon et al. 11 , the most widely adopted model is the frailty phenotype proposed by Fried et al. 3 , followed by the frailty index (FI) described by Mitnitski et al. 1 and Rockwood & Mitnitski 7 .
FI is a mathematical model derived from data from the longitudinal Canadian Study of Health and Aging (CSHA) 1,12 .It does not provide for a specific set of clinical markers present in old age, as in the model by Fried et al. 3 , but rather a sum of observed deficits in different systems at the time of measurement (signs, symptoms, functional disability, diseases, laboratory results) 1,2,7 , based on the notion that age-related changes have a cumulative effect on health.It is based on the quantification of observed changes in a variety of physiological, psychological, and functional conditions and the search for relations between them and adverse outcomes in the elderly 12,13 .According to the Rockwood & Mitnitski 7 , frailty results from an accumulation of deficits and expresses a continuous scale ranging from 0 to 1, which reflects the relationship between the number of deficits the individual presents and the total number of possible deficits from a model corresponding to the study sample (for example, an individual with 4 deficits in 38 study variables has a frailty index of 0.10) 7,12 .
The FI does not require the inclusion of a specific number of deficits.Previous studies have used 30 to 70 variables 1,2,7,14,15 .However, to be part of the index, the variable must meet the following criteria: associated with age and negative health outcomes; present in at least 1% of the population; includes various organ systems; does not contain more than 5% missing data; and is not saturated, i.e., present in at least 80% of individuals below 90 years of age 12,16,17 .In community-dwelling elderly, the prevalence of frailty measured by the FI model is around 24% 14,18,19 .It is higher in women than in men and increases with age 13,14,15,19,20 .
In Song et al. 14 , FI consisted of 36 deficits that included health conditions, signs and symptoms, and functional disability.Frail individuals were defined as those that scored > 0.25.The results showed that mean FI was higher in elderly individuals that died during follow-up compared to those that survived.Having more deficits was associated with greater risk of adverse events.Frail elderly showed 15% greater risk of death than non-frail elders, independently of gender 19 .FI is a more robust predictor of mortality than chronological age 2,7,14 .
In Kulminski et al. 20 , FI showed greater accuracy than frailty phenotype in discriminating elderly with moderate to severe frailty 3 , since it assesses frailty as the product of the cumulative effect of deficits in multiple physiological systems on a continuous scale rather than by specific indicators present in old age 11,14,19,20,21 .Based on the concept of biological heterogeneity, a higher FI score is associated with greater risk of death, independently of chronological age 1 .In this context, the current study's hypothesis was that the frailty index is capable of predicting mortality, independently of chronological age.There are no Brazilian studies on frailty and mortality in the elderly using the cumulative deficits model.The study aimed to investigate the prevalence of frailty based on the frailty index and the association between this measure of frailty and death in community-dwelling elderly.

Material and methods
This study was based on two databases.One was the electronic database from the FIBRA Study (the Portuguese acronym for Frailty in Elderly Brazilians), conducted in Campinas, São Paulo State, Brazil, in the context of a multicenter, population-based cross-sectional study aimed at investigating frailty and its relations with socio-demographic, psychosocial, clinical, cognitive, anthropometric, functional capacity, and physical and mental health variables in community-dwelling elderly.The second database was from the Mortality Information System (SIM) of the city of Campinas for the years 2009, 2010, 2011, 2012, and 2013, accessed every six months.
The FIBRA Study project was submitted to the Ethics Committee for Research in Human Subjects, School of Medicine, State University in Campinas, and approved under case review n. 208/2007.The current study was submitted as an addendum to the main project and was approved under case review n. 736.943/2010.The ethical principles of the Declaration of Helsinki were followed, and all the participants signed a free and informed consent form.

Participants
Participants in the FIBRA Study were recruited in family or individual households located in 90 randomly selected urban census tracts in Campinas, in which previously specified quotas of men and women were recruited, 65 to 69, 70 to 74, 75 to 79, and 80 years and older.The quotas were proportional to their presence in the elderly population in each census tract.The following eligibility criteria were adopted: age 65 years or older, permanent resident in the household and in the census tract, and absence of severe cognitive, communicative, or sensory impairment or severely impaired mobility.The exclusion criteria were: problems with memory, attention, orientation in time and space, and communication, suggestive of dementia; bedridden elderly; elderly with severe stroke sequelas, with loss of strength and/or aphasia; advanced or unstable Parkinson's disease, with serious impairment of mobility, speech, or affect; seriously impaired hearing or vision, hindering communication; and terminal illness 22 .Of all the elderly recruited, 1,055 appeared at the data collection sites and 900 comprised the sample.The 155 exclusions were due to the following: age under 65 years, and nonresident in the census tract, and withdrawal.
The 900 selected elderly participated in the first phase of the data collection, consisting of a section on socio-demographic, anthropometric, clinical, and frailty measures 3 .In this phase, the score obtained on the Mini Mental State Examination (MMSE) was used as the criterion for participation in the second data collection phase, on self-rated physical and mental health, functional performance, and subjective well-being.The following cutoff scores were used for exclusion: 17 for illiterate individuals; 22 for elderly with 1 to 4 years of schooling; 24 for those with 5 to 8 years of schooling; and 26 for those with at least 9 years of schooling 23,24,25 .Six hundred and eighty-nine elderly without cognitive deficit suggestive of dementia selected according to this criterion constituted the current study's sample.Mean age was 72.1 ± 5.3 years; 68.8% were women 26 (Figure 1).

Variables and measures
The target variables were investigated according to the following conditions: (a) FI: composition of the index used 40 variables from different domains assessed by the FIBRA Study, according to the eligibility criteria in Searle et al. 12 and Song et al. 14 .Among the available anthropometric measures, the following were selected: body mass index (BMI) and waist to hip ratio (WHR) 27 .In self-rated health, the following were chosen: chronic diseases, signs and symptoms, difficulties in performing activities of daily living, falls, number of medication, smoking, alcohol consumption, self-perceived health, and leisure-time physical exercise and sports 3,12,28,29 .Physical performance measures included: gait speed and grip strength 3 .The selected psychosocial variables were: depressive symptoms and life satisfaction 30,31 .Frailty indices were calculated for all the participants, based on the selected variables.When the variables were dichotomous (e.g.hypertension -yes vs. no), the attribute's presence was scored as 1 and its absence as 0. For continuous variables, intermediate points were created (e.g., self-rated health as  very good = 0; good = 0.25; fair = 0.5; bad = 0.75; and very bad = 1.0).For cognitive status, the points corresponded to the quintiles obtained by the sample in the cognitive screening test (1st quintile = 1; 2nd quintile = 0.75; 3rd quintile = 0.5; 4th quintile = 0.25; and 5th quintile = 0).The study followed the rules recommended in the literature for assessing parameters, such as waist to hip ratio (men > 1 = 1.0; women > 0.85 = 1,0; men < 0.99 = 0; and women < 0.84 = 0) 27 , low grip strength (the lowest 20% of values in the distribution of the means on three attempts, adjusted by gender and BMI = 1.0) 3 , low gait speed (values above the 80th percentile in the distribution of the mean times in seconds that the individual took to walk 4.6 meters three times) 3 and low level of physical activity (the lowest 20% of values in the distribution of the sum of kcal spent in physical exercise, adjusted by gender = 1.0) 3 .For each elderly individual, FI was calculated, based on the sum of the scores for frailty divided by 40, which was the total number of selected items in the protocol.Unanswered items were excluded from the participant's score.The lowest denominator considered was 29 deficits.Based on Rockwood et al. 32 , on a scale from 0 to 1, individuals were classified as fit when they scored ≤ 0.11, as pre-frail when they scored from 0.12 to 0.24, and as frail when they scored ≥ 0.25 (Table 1) 14 .(b) Gender and age: two self-report items, (male vs. female; years of age; date of birth).(c) Mortality: in the database of the SIM for Campinas, we identified and counted the surviving and non-surviving elderly from 2009 to 2013 at each moment when the database was consulted.The database was obtained by probabilistic record linkage using the blocking strategy in multiple linked steps: first name, last name, year of birth, and home address.It was measured as the proportion of deaths in elderly in Campinas, and the variable was categorized as "yes" or "no".

Statistical analysis
The FIBRA Study forms were consecutively checked by two supervisors before keying-in.The electronic database was checked by two trained evaluators, with 100% agreement required.Frequency measures were performed for the scores obtained by the elderly in each of the health and psychosocial variables comprising the FI and in the score ranges corresponding to the three levels of frailty.Statistical comparisons between the frailty indices for the groups formed by men and women were performed using the Mann-Whitney test and between the age groups according to the Kruskal-Wallis test.

Results
Of the 689 elderly whose data were analyzed according to the FI protocol, 86.6% (588) answered 40 items; 9.75% answered 39; 2.81% answered 38; 0.15% answered 37; 0.30% answered 31; and 0.15% answered 29.Twelve elderly were excluded who failed to respond to more than 30% of the study items.Table 1 showed the absolute frequencies and percentages of the variables that comprised the FI according to the score adopted.The observed frailty indices were generally low or intermediate: the lowest observed index was 0.03 and the highest 0.62, with low dispersion around the mean (0.23 ± 0.10) and a median of 0.22.Prevalence of frail elderly was 38.8%, pre-frail elderly 51.6%, and fit elderly 9.6%.Women showed significantly higher mean values for subcomponents of the frailty index compared to men in BMI, WHR, fatigue, hypertension, arthritis, osteoporosis, urinary incontinence, falls, sleep problems, and need for help using transportation and shopping.The mean indices in men exceeded those in women for the variables cancer, smoking, alcohol consumption, and need for help preparing meals.Overall mean FI was higher in women (0.25 ± 0.10) than in men (0.20 ± 0.10) (Table 2).
As for mean values for each FI component and age bracket, the variables with significant differences were: cognitive status, WHR, level of physical activity, fatigue, grip strength, slow gait, falls, sleep problems, and need for help using transportation, shopping, and doing household chores.Differences between age brackets were verified with the multiple comparisons test and are represented by letters in Table 3. Overall mean FI did not differ statistically by age bracket (p = 0.063).
In the five years of follow-up, 8.2% of the elderly died.There was no significant difference between mortality and the variables gender and FI (Table 4).There was a significantly higher percentage of mortality in elderly 75 years and older (Table 4).
Cox multiple regression analysis adjusted by age, gender, and frailty index showed that the variables age and gender were significantly associated with mortality.Increased risk of death in the elderly was associated with older age (each additional year of age was associated with an increase of 10.2% in mortality) and male gender (73% greater risk of death than in women).No association was observed between FI and mortality in the overall sample (Table 5).

Discussion
Prevalence of frail elderly was 38.8%, higher than the mean prevalence in a review of 24 populationbased studies in elderly 65 years and older, showing FI prevalence of 24% (range: 18-44%) 19 .A survey by Collard et al. 18 , involving 21 studies of community-dwelling elderly (61,500 participants), found prevalence rates ranging from 4 to 59%.The total FIBRA Campinas sample (n = 900), which used the model by Fried et al. 3 , showed 7.7% frail, 52.3% pre-frail, and 40% fit or non-frail elderly.Elderly women and elders 80 years or older scored on more frailty criteria, when compared to elderly men and elders under 80 years of age 26,33 .
Cad. Saúde Pública 2017; 33(5):e00194115     This wide variation may be explained by the difference in the instrument for assessing the syndrome and differences in the sample's composition, especially relating to ethnicity and nationality.Despite the universal decline in the mechanisms for adaptation and biological regulation associated with aging, different trajectories can distinguish individuals from different cohorts and different contexts 8,20,34,35,36,37 .
Women showed higher frailty indices than men, corroborating findings in the literature, for example the National Population and Health Survey (NPHS) 14 , Survey of Health, Ageing, and Retirement in Europe (SHARE) 38 , and Beijing Longitudinal Study of Aging 15 .Unlike these studies, which also found higher frailty indices and higher frailty prevalence rates in the older elderly, the current study did not show differences between these variables (frailty index and age bracket).Individuals in the same age bracket may present different health profiles.However, the current sample did not show differences in relation to age groups as the frailty index proposes.
In this study, risk of death in men was 1.73 times higher than in women.In a study of elderly Chinese, women showed higher mean FI than men, while the incidence of death was higher for men than for women 39 .In a sample of elderly Brazilians, males were 2.7 times more likely to die than females 40 .
Cad. Saúde Pública 2017; 33(5):e00194115 According to the literature, men generally die more from acute illnesses, while women live longer and with more disabilities and more comorbidities 41,42 .Kulminski et al. 20 assessed the prevalence of morbidity in men and women in relation to mortality, using the cumulative deficits model.They found that morbidity and risk of mortality according to gender may vary as a result of the set of deficits used, the cohort, and environmental factors.These results show the paradox of morbidity and mortality: despite worse health conditions, women present higher survival rates than men.
The association between chronological age and mortality has been widely reported in the literature 40,41,42 .Physiological changes associated with aging and lower functional reserve are important factors influencing the relationship between age and mortality, especially in advanced old age 36,41,42 .
According to the theoretical formulation of the frailty index model, the index represents a measure of the individual's biological age, and FI should thus be a more robust predictor of mortality than chronological age itself 1,7 .Unlike findings from international studies, the current study found no association between FI and mortality 14,36,38 .As far as we know, this is the first study in which FI does not predict mortality in community-dwelling elderly.The main objective of the frailty index as a measure is to assess biological heterogeneity in order to identify the individuals most vulnerable to adverse health events 1,7 .However, this result was not observed in this sample.This may be related to the limitation of the index in terms of the variability in health in old age and its different influences on the occurrence of negative events 37,43 .That is, some deficits have a greater effect than others on mortality rates 40 .For example, elderly individuals with heart disease, diabetes, and hypertension score lower on the index when compared to elderly with osteoporosis and arthritis, which require help using transportation and performing household chores.Still, the former conditions are more lethal than the latter.Besides, there is no consensus at present on the cutoff point as the classificatory criterion for FI or for characterizing frail elderly in different contexts 34,35,36,37,42,43 .Despite this lack of consensus in the literature, the cutoff points used in the current study followed the recommendations made by the developers of the FI 12,14,32 .According to Martin et al. 44 and Walston & Bandeen-Roche 45 , although the developers of the frailty index suggest that the items comprising the measure are correlated, there is still no evidence on the internal validity of the set of items in the FI, since the number and nature of the variables differ in the studies that used the operational model.The original FI study used 70 variables to compose the measure, including diseases, signs and symptoms, functional disabilities in basic and instrumental activities of daily living, cognitive decline, and psychological disorders 32 .The FIBRA study used 40 variables to comprise the FI.According to the authors 7,12 , a composite index with at least 30 variables is capable of predicting adverse health outcomes.It is also necessary to understand the underlying mechanisms in the interaction of deficits in the pathophysiology of frailty 45 .
For convenience purposes, the data in the FIBRA Study were collected in a social setting in the community that is well-known and easy for participants to access.This decision may have partly selected elderly individuals in better physical condition.Likewise, when selecting the elderly for the second phase of data collection using the MMSE, FIBRA clearly opted for elderly with more intact cognition.Song et al. 14 did not use a cognitive screening test in their sample selection.In the Beijing Longitudinal Study of Aging, the MMSE score was not used as an exclusion criterion for participants 15,39 .Ours is the first Brazilian study on the prevalence of FI and its association with socio-demographic variables and mortality as the outcome.The data showed that women presented higher FI than men.For the study population, the frailty index did not serve as a good measure of frailty, since it was not associated with either mortality or chronological age.Future studies may assess and compare different measures of frailty with unfavorable outcomes, besides identifying factors that protect against negative events related to the frailty syndrome.

Figure 1
Figure 1Composition of sample of elderly selected in the FIBRA Study to construct the frailty index (FI).Campinas, São Paulo State, Brazil.

Table 1
Distribution of variables included in the frailty index, frailty categories according to frailty index (FI) score, and percentage of deaths.FIBRA Study, Campinas, São Paulo State, Brazil, 2008-2009.

Table 1 (continued)
Cad. Saúde Pública 2017; 33(5):e00194115 The Dunn test was used for post hoc comparison of the results from the Kruskal-Wallis test.The rates of surviving and non-surviving elderly by gender, age, and frailty levels were compared with the chi-square test.Associations between the independent variables gender, age, and frailty and mortality were studied by Cox regression.The analyses used SAS, version 9.2 (SAS Inst., Cary, USA).Statistical significance was set at 5%, or p < 0.05.

Table 2
Comparison of men and women on scores for variables included in frailty index (FI) and according to FI score.FIBRA Study, Campinas, São Paulo State, Brazil, 2008-2009 (n = 677).

Table 3
Comparison of age groups on scores for variables included in the frailty index (FI) and according to FI score.FIBRA Study, Campinas, São Paulo State, Brazil, 2008-2009 (n = 677).
BMI: body mass index; WHR: waist to rip ratio.