Introduction

Frailty is a geriatric state of vulnerability or reduced functional reserve, characterized by slow walking speed, muscle weakness, low levels of physical activity and, in some cases, cognitive impairment not related to neurodegenerative disorders (1). This condition has become a major public health problem in older adults, mostly because of its association with increased risks for hospitalization, institutionalization, systemic diseases, and all-cause mortality (2, 3). On the other hand, accidental falls are a leading cause of fatal and non-fatal injuries among older adults, with direct medical costs exceeding $30 billion dollars per year in the US only (4).

There is evidence supporting a link between frailty and risk of falls. A recent meta-analysis of 10 studies showed that the risk of falls in frail community dwelling adults aged ≥65 years is higher than in their robust counterparts (5). Other meta-analyses also showed a significant relationship between frailty status and risk of falls (6, 7). However, all these meta-analyses suffered from limitations due to inconsistencies among the included studies, which were partly related to heterogeneity in study designs (mainly because of the use of different questionnaires to identify frailty, and the confounders used for adjustment in individual studies). Moreover, with the exception of a few studies (8), data came from people living in urban or industrialized centers, where risk factors and lifestyles are totally different from that of rural settings. More information is needed before concluding that frail individuals are at an increased risk of falls in these remote communities. In this study, we aimed to assess whether a frail or pre-frail status is associated with an increased risk of falls among older adults living in rural Ecuador.

Methods

Study population

Atahualpa is a rural Ecuadorian village where previous studies on frailty have been conducted (9, 10). As detailed elsewhere, inhabitants are homogeneous regarding ethnicity, diet, and lifestyles (11). The study population included Atahualpa residents aged ≥60 years — identified during door-to-door surveys — in whom the frail status and the risk of future falls were assessed. The I.R.B. of Hospital-Clínica Kennedy, Guayaquil, Ecuador (FWA 00006867) approved the study.

Frailty assessment

Frailty was evaluated by the use of the Edmonton Frail Scale (EFS), a reliable instrument that consists of 10 domains with 11 items including cognition (the clock drawing test, 2 points), general health status (number of hospitalizations during the past year, 2 points and self-reported health status, 2 points), functional independence (from eight independent activities of daily living, 2 points), social support (count on someone who is willing and able to meet the subject’s needs when the subject needs help, 2 points), medication use (five or more prescription medications, 1 point and forgetfulness for taking medications, 1 point), nutrition (weight loss, 1 point), mood (depression, 1 point), incontinence (1 point), and balance and motility (the “Timed Get Up and Go” test, 2 points) (12). The maximum total score is 17, and individuals were classified into three categories including robust (0 to 4 points), apparently vulnerable or pre-frail (5 to 6 points) and frail (≥7 points) (13). The stratification of the EFS score in the three aforementioned groups was according to the recommendations of the reproducibility of the EFS in a Latin American elderly population (which is closely related to the current study population) (13). Also, same cutoffs have been used in previous studies on the correlates of the EFS conducted by our group (9, 10).

Risk of future falls

The Downton Fall Risk Index (DFRI) was used to assess the risk of future falls. The DFRI is a five-item questionnaire inquiring about history of previous falls, use of specific medications (tranquilizers/sedatives, non-diuretic anti-hypertensives, diuretics, anti-parkinsonians, and antidepressants), sensory or motor deficits (visual impairment, hearing impairment, paresis), gait abnormalities (with or without aid), and confusion (14, 15). Using the DFRI, a score ≥3 is considered positive (high risk of future falls). In a previous study from the Atahualpa Project, we used the DFRI, and conducted sensitivity analysis to assess its reliability in its residents. In that previously published study, we found a reasonable reliability of the DFRI in our population (16).

Covariables investigated

Demographics (age, sex, scholarity), alcohol intake (dichotomized in <50 and ≥50 g per day), cardiovascular risk factors, sleep quality, symptoms of depression, and history of an overt stroke were selected as confounding variables. These relevant confounders were assessed by means of procedures and interviews previously described in the Atahualpa Project. We used the American Heart Association criteria to assess physical activity, the body mass index, blood pressure, and fasting glucose (17, 18). Sleep quality was assessed by the use of the Pittsburgh Sleep Quality Index, which is basically used to differentiate between “good” and “poor” sleepers. The instrument consists of 19 items grouped into seven components (assessment of sleep duration, sleep disturbances, sleep latency, daytime dysfunction due to sleepiness, sleep efficiency, overall sleep quality, and medications needed to sleep), each weighted on a 0 to 3 scale, for a total score of 21 points (19). Symptoms of depression were assessed by the depression axis of the depression-anxiety-stress-21 scale, a validated field instrument that measures dysphoria, hopelessness, devaluation of life, self-deprecation, lack of interest/involvement, anhedonia, and inertia (20). To recognize patients with overt strokes, rural doctors screened all participants with the use of a validated field instrument, and then, certified neurologists confirmed the diagnosis with the aid of MRI (21).

Statistical analyses

Data analyses are carried out by using STATA version 15 (College Station, TX, USA). In univariate analyses, continuous variables were compared by linear models and categorical variables by x2 or Fisher exact test as appropriate. Logistic regression models, adjusted for relevant confounders, were fitted to assess the association between the continuous EFS score and the frail status (as exposures) and the DFRI (dependent variable). Partial and semi-partial correlation coefficients were constructed to assess confounders affecting the aforementioned associations.

Results

Of 463 individuals aged ≥60 years enrolled in the Atahualpa Project (2012 – 2018), 342 were active at the time of this study (January 2019). Of these, 18 declined consent and were excluded. The mean age of 324 participants was 70.5±8 years (median age: 69 years; age range: 60 to 95 years), 186 (57%) were women, 250 (77%) had primary school education only (mean years of education: 6.5±2.9 years), and 50 (15%) reported high intake of alcohol (all men). Blood pressure levels ≥140/90 mmHg were present in 137 (42%) individuals, a body mass index ≥30 kg/m2 in 75 (23%), fasting glucose ≥126 mg/dL in 98 (30%), poor physical activity in 28 (9%), poor sleep quality in 106 (33%), symptoms of depression in 37 (11%), and an overt stroke in 24 (7%).

The mean score in the EFS was 4.4±2.5 points, with 180 (56%) participants classified as robust, 76 (23%) as pre-frail and 68 (21%) as frail. The DFRI was positive in 87 (27%) participants. The most common positive component of the DFRI was history of falls (172 cases, 53%), followed by use of medications (142 cases, 44%), sensory or motor deficits (139 cases, 43%), gait abnormalities (55 cases, 17%), and confusion (13 cases, 4%). Table 1 shows characteristics of participants across categories of the EFS and the DFRI. As noted, most of the covariables evaluated — with the exception of a body mass index ≥30 kg/m2, fasting glucose ≥126 mg/dL, and symptoms of depression — were significantly associated with a frail status. In contrast, high alcohol intake was associated with a robust status. This inverse relationship can be an artifact since this covariable was only recorded in men, and men were most often robust than women. Regarding covariables associated with an increased risk of future falls, a positive DFRI was significantly associated with age, blood pressure levels ≥140/90 mmHg, fasting glucose ≥126 mg/dL, poor physical activity, poor sleep quality, symptoms of depression, and an overt stroke.

Table 1 Characteristics of 324 Atahualpa residents aged ≥60 years across categories of frailty status and the Downton Fall Risk Index (univariate analyses)
Full size table

Also in univariate analysis, the mean EFS score was significantly higher among participants with a positive, than in those with a negative DFRI (5.8±2.9 versus 3.9±2.2; p<0.001). The percentage of frail individuals was higher (40% versus 14%; p<0.001), while that of robust individuals was lower (37% versus 62%; p<0.001) among those with a positive DFRI. However, there were no differences in the percentages of prefrail individuals according to a positive or negative DFRI status (23% versus 24%; p=0.791).

A logistic regression model, adjusted for all the aforementioned covariables, showed no association between the continuous EFS score and the DFRI (Table 2). Then, the most parsimonious model was fitted with variables having p<0.2 significance in the fully-adjusted model. Again, there was no association between the continuous EFS score and the DFRI (Table 3). We then investigated the possibility of having a positive DFRI according to the three categories of frailty (instead of the continuous EFS score) using covariables included in the most parsimonious model, and found no differences in the Odds of having a positive DFRI when prefrail and frail subjects where compared with robust ones (Table 4).

Table 2 Fully-adjusted logistic regression model showing lack of association between the Edmonton Frail Scale score and the Downton Fall Risk Index (dependent variable)
Full size table
Table 3 Most parsimonious logistic regression model, using covariables reaching a significance of p<0.2 in the fullyadjusted model, showing lack of association between the Edmonton Frail Scale score and the Downton Fall Risk Index (dependent variable)
Full size table
Table 4 Logistic regression model, using covariables reaching a significance of p<0.2 in the fully-adjusted model, showing lack of association between categories of the frailty status (robust versus pre-frail, and frail) and the Downton Fall Risk Index (dependent variable)
Full size table

Since age seemed to be a relevant confounder in the association between frailty and risk of falls, we fitted an interaction model with participants stratified according to their median age, but found no significant interaction of age in this association (OR: 0.85; 95% C.I.: 0.25 – 2.87; p=0.793). History of an overt stroke also appeared as a significant confounder in all previous models. Therefore, we fitted a logistic regression model that only included the 300 stroke-free participants. Nevertheless, in this model without stroke as a covariable, the association between the continuous EFS score and the DFRI was also non-significant (OR: 1.06; 95% C.I.: 0.92 – 1.21; p=0.428).

In addition, partial and semi-partial correlation coefficients showed that several confounders together (age, high glucose levels, and history of an overt stroke) took away the association between frailty and the risk of falls found in univariate analyses (Table 5).

Table 5 Partial and semi-partial correlation coefficients showing that age, high glucose levels, and history of an overt stroke took away the association between the continuous Edmonton Frail Scale and the frailty status, and the risk of falls (dependent variable)
Full size table

Discussion

The relationship between frailty and the risk of falls has not been investigated in remote rural settings, and little is known of the impact of frailty on the risk of falls in these populations. The present study shows a significant association between both variables, but only in univariate analyses. Of note, several of the chosen covariables were significantly associated with both the exposure (frail status) and the outcome (risk of falls). This might explain why the association between frailty and the risk of falls became non-significant when confounders were added to multivariate regression models. This was better demonstrated by the use of partial and semi-partial correlation coefficients, which showed that age, high glucose levels, and history of an overt stroke tempered the significance of the association between frailty and risk of falls.

It is possible that the effect of confounders explains the disparities in the results of some previous studies attempting to assess the relationship between frailty and risk of falls. Another possibility is that instruments used to assess frailty in many of these studies fail to take into account several of the items included in the EFS (22, 23).

There are several factors that increase the risk of falls in older adults, some of which are intrinsic to the individual while others are extrinsic (i.e., related to the environment) (24). In any case, many of the intrinsic factors that increase the risk of falls are components of the frail syndrome (cognitive decline, functional independence, use of medications, depression, and abnormal balance and motility). Therefore, it is not surprising that both conditions are associated. Indeed, results of univariate analyses conducted in the present study strongly suggest that frail individuals are at increased risk of future falls. The problem here are the confounders, that is, the presence of specific circumstances or conditions that make some subjects more prone to developing falls in the follow-up. In the present study, the non-significant association found in multivariate models does not rule out the actual association between frailty and risk of falls. Rather, it means that some frail individuals, particularly those with increasing age, diabetes mellitus, and history of an overt stroke have a greater risk of falls than other frail subjects due to these characteristics. This should be considered in further studies assessing the relationship between frailty and risk of falls in the population at large.

Major strengths of the present study are the unbiased selection of participant subjects and the methods used to assess frailty and the risk of future falls. A potential limitation is its cross-sectional design which does not allow for an assessment of the reliability of the DFRI to predict the actual risk of falls among inhabitants living in rural populations, where environmental factors might be an important cause of falls (in particular, walking throughout uneven non-paved streets). Future longitudinal studies using the Atahualpa Project cohort will be of value to answer this question.

In summary, this study shows that the prevalence of frailty among older adults living in a remote rural community is similar to that reported in industrialized urban centers. In contrast, the risk of falls is somewhat lower according to the DFRI. This discrepancy is probably related to the fact that the DFRI does not take into account distinct environmental conditions that may result in falls. More importantly, frailty is significantly associated with the risk of falls in univariate analyses, but the association disappeared in multivariate models, most likely because increasing age, diabetes mellitus, and history of an overt stroke tempered the statistically significant association between exposure (frailty) and outcome (risk of falls) variables.