| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
| ||||||||||||||||||||||||
RESEARCH ARTICLE |
1 Centre for Mental Health Research, Australian National University, Canberra.
2 Centre for Ageing Studies, Flinders University, Adelaide, Australia.
Address correspondence to Kaarin J. Anstey, Centre for Mental Health Research, Australian National University, Canberra, Australia. E-mail: kaarin.anstey{at}anu.edu.au
 |
Abstract |
|---|
 TOP Abstract MethodsResultsDiscussionReferences |
|---|
Key Words: Falls • Well-being • Depression • Morale
FALLS constitute a major public health problem in later life and are the leading cause of unintentional injury and hospitalization in persons aged 65 and older (Dellinger & Stevens, 2006
). With age, the risk of falling rises steadily, and the 12-month risk of mortality after being hospitalized for a fall is approximately 50% (Rubenstein, 2006). Considerable research has identified risk factors for falling, and evaluation of interventions to address these risk factors is extensive. Most of this research has focused on understanding the physiological, sensory, and environmental factors leading to falls (Lord, Clark, & Webster, 1991; Lord & Fitzpatrick, 2001; Lord et al., 2003; Lord, Ward, Williams, & Anstey, 1993, 1994). There is also literature demonstrating that depressive symptoms may increase the risk of falling in later life for both community-based individuals (Kron, Loy, Sturm, Nikolaus, & Becker, 2003) and stroke survivors (Jorgensen, Engstad, & Jacobsen, 2002). The Osteoporotic Fractures Study found that, over a 6-year follow-up period, depression increased the risk of falling by 60% in women aged 65 and older (Whooley et al., 1999). Another multisite study of 311 community-dwelling elders followed for 36 weeks found that "depressive state of mind" predicted falling along with more traditional risk factors such as sway, fall history, and grip strength (Stalenhoef, Diederiks, Knottnerus, Kester, & Crebolder, 2002).
It remains unclear whether the association between depression and falling is a direct effect or whether it is mediated by psychotropic medication (Cumming, 1998; Leipzig, Cumming, & Tinetti, 1999; Lord, Anstey, Williams, & Ward, 1995), cognitive decline (Anstey, von Sanden, & Luszcz, 2006), physical decline (Penninx et al., 1998), or other factors. One study has shown that depression among hospitalized older adults is associated with motor disturbances that may predispose individuals to falling, thus providing one mechanistic link between depression and falling (Turcu et al., 2004). In the present study we seek to broaden the scope of factors that are considered to be potential predictors of falling in later life, by examining psychological well-being as a predictor and covariate of falling, in the Australian Longitudinal Study of Ageing.
It is possible that the observed associations between depressive symptoms and falling reflect a more general association between psychological well-being (Luszcz, 1998) and falling, justifying the investigation of other measures of psychological well-being for potential inclusion in fall-risk assessments or for incorporation in intervention strategies. The relationship between well-being and adaptive aging has received increasing attention (Ryff & Singer, 2000) as the literature recognizes that well-being is more than the absence of mental illness or negative affective states. Two approaches to well-being have been identified (Ryan & Deci, 2001). The hedonic approach focuses on the experience of positive or negative valence, with the relationship between these two states typically an excellent indicator of life satisfaction (Diener, 1984). In contrast, the eudaimonic approach focuses on those characteristics (e.g., mastery and control beliefs, self-growth and acceptance, self-referent beliefs, social support, and autonomy) that are related to adaptive functioning (Ryff & Keyes, 1995). However, increasing degrees of commonality among these approaches has been demonstrated at the higher-order factor level (Abbott et al., 2006). Consequently, the present study includes measures of well-being that are inclusive of these different approaches.
Another important question for understanding psychological risk factors for falls is whether changes in factors that are associated with falling in cross-sectional analyses or at the population level are also predictive within individuals in longitudinal analyses. Researchers have stressed that between-person differences do not necessarily reflect within-person change (Hofer & Sliwinski, 2001). Therefore, in the present study, we first assessed the key variables of interest (measures of well-being) by using population-average models that estimated whether individuals having poorer scores on measures of well-being were at increased risk of falling compared with individuals with higher scores on measures of well-being. We then used random effects models to evaluate whether decline in well-being within persons was also associated with an increasing fall risk over the follow-up period.
Finally, an important aspect of this study was the inclusion of a larger number of covariates than are typically incorporated in falls research. We adjusted for demographic factors, medical conditions, a measure of cognitive function, and health behaviors that have been associated with falling in previous research (Anstey et al., 2006; Lord et al., 1994; Whooley et al., 1999) to evaluate the extent to which well-being was independently associated with falling. Our specific hypotheses were as follows. First, lower morale, increased depressive symptoms, and lower expectation of control would be associated with an increased risk of falling in population-average models. Second, declines in indices of well-being would be associated with an increased rate of falling in random effects models. Third, given the association between falls and several of the covariates included in this study, we expected effect sizes to reduce progressively after adjustment for covariates.
|
METHODS |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
). Participants provided written consent and the Clinical Investigations Committee of the Flinders Medical Centre in South Australia granted ethics approval. Researchers used the South Australian Electoral Roll as a sampling frame to identify potential participants aged 70 years and older. They stratified the sample (N = 2,087) by age and sex into three 5-year cohorts, that is, 70–74, 75–79, and 80–84; there was a fourth cohort aged 85 years and older. In addition, partners of participants aged 65 and older were invited into the study. Our investigation drew from three waves of data collection including clinical assessments: Wave 1 was conducted between September 1992 and March 1993; Wave 3 was conducted in 1994; and Wave 6 was conducted in 2000. Waves 2, 4, and 5 comprised telephone interviews and did not include clinical or cognitive assessments. We included participants in the present study (N = 787) if they had answered falls questions at Wave 6. All of these participants had falls data available at Wave 1, and only 14 (1.78%) had missing data at Wave 3.
Education and Mini-Mental State Examination
We measured education as the total number of years of formal education (range = from 0 to 20 years). We used the clinical cutoff for the Mini-Mental State Examination (MMSE), that is, 23 points or fewer on the 30-item version or prorated equivalent on the 21-item version, to screen out those participants who had dementia. Researchers also used MMSE scores as a measure of cognitive functioning to predict falls among the nondemented participants. At Wave 1, researchers used a 21-item version of the test (Folstein, Robins, & Helzer, 1983). This test assesses orientation, registration, attention, calculation, and recall (Teri, Hughes, & Larson, 1990). We administered the 30-item version at Waves 3 and 6. We converted the 21-item version to a 30-item scale by using this formula, (MMSE score/21) x 30, to match scores from Waves 3 and 6 (Anstey et al., 2006; Luszcz, 1998; Luszcz, Bryan, & Kent, 1997).
Sensorimotor Measures
We tested corrected distance visual acuity at 3 m for each eye by using a well-illuminated Snellen chart. The criterion for distance visual acuity was the smallest line at which the participant could correctly read at least half of the characters. We measured grip strength in the dominant hand by using a dynamometer while the participant was standing. The average of two attempts at maximum force was recorded in kilograms. We measured balance with the semitandem stand and a measure of functional reach. For the tandem stand, participants were required to stand with the heel of one foot next to the ball of the opposite foot and to maintain their balance without moving their feet. Performance was rated according to whether the participant could hold the position for 10 seconds (coded as 1), was able to hold the position for less than 10 seconds (coded as 2), or tried but was unable to hold the position (coded as 3). Functional reach assessed the maximal reach a participant could achieve while maintaining a fixed-base standing posture. The average distance in centimeters from three trials was used as the score.
Health Measures
We obtained health measures during the home interview. We measured self-rated health (SRH) on a 5-point scale (1 = excellent, 2 = very good, 3 = good, 4 = fair, and 5 = poor). A self-report questionnaire of 61 medical conditions was based on measures used in the Duke Longitudinal Studies of Aging (Palmore, 1970) and the Established Populations for Epidemiologic Study of the Elderly (National Institute on Aging, 1986). We selected conditions known to affect fall risk for this study as independent variables, including heart condition or attack, hypertension, stroke or transient ischemic attack, and diabetes. We scored all of these as 1 = yes and 0 = no. Information was also obtained on psychotropic medication (1 = yes; 0 = no), current smoking (1 = yes; 0 = no) and frequency of alcohol consumption (1 
4 times per month; 2 = 2 to 3 times per week; 3 
4 times per week).
Psychological Well-Being
We measured depressive symptoms by using the Center for Epidemiologic Studies–Depression scale (CES-D), a 20-item questionnaire designed to measure depression in community-based epidemiological studies (Radloff, 1977). Item responses are in reference to the way the individual felt over the past week. A 4-point Likert-scale is utilized, with answers extending from rarely or none of the time (0) to most of the time (3), with scores ranging from 0 to 60. The CES-D has shown internal consistencies of.83 in 60- to 69-year-olds and.78 in adults aged 70 and older, and it has a sensitivity of 80% and a specificity of 73% for detecting clinically significant depression in previous research (Lewinsohn, Seeley, Roberts, & Allen, 1997). Cronbach's alpha was 
= 0.84 or 0.85 at each wave.
We measured morale with the five items loading most strongly on each of the subscales of agitation, lonely dissatisfaction, and attitude toward own aging from the Philadelphia Geriatric Centre Morale Scale (Lawton, 1983; Liang & Bollen, 1983; Luszcz, 1998). The factor structure of scales has been shown to be invariant across age and gender in older samples (Liang, Lawrence, & Bollen, 1986). In this 15-item test, with responses of agree or disagree, a higher score indicates higher morale. Cronbach's alpha was = 0.81 or 0.82 at each wave.
We measured control with the Expectancy of Control subscale from the Desired Control Measure (DCM; Reid & Ziegler, 1980), which asks respondents to indicate the extent to which they believe they actually have control in areas of involvement with others, engagement in activities, and health. The DCM comprises two subscales measuring desire and expectancy for control in areas found to be important to psychological well-being of older adults (Reid, Haas, & Hawkings, 1977). The expectancy subscale appears to be a more reliable and sensitive measure than either the desire subscale or a total score, based on summing the cross-products of the two separate subscales (Reid & Ziegler), possibly because of the importance of the expectancy of realizing desired outcomes rather than the desire for control per se. This 12-item scale with five levels of response (1= strongly agree, 2 = agree, 3 = neither agree nor disagree, 4 = disagree, and 5 = strongly disagree) was scored such that a higher score indicated lower control expectancies and hence poorer well-being. Cronbach's alpha was = 0.71, 0.75, and 0.77 at Waves 1, 3, and 6, respectively.
Questions on Falling
At Waves 1, 3, and 6 participants answered an injury questionnaire as part of the larger health and sociodemographic questionnaire, and this included questions on falls. For this study, the main dependent variable was falls per year, measured by this question: "Have you had any falls in the past year—including those falls that did not result in injury as well as those that did?" Although the gold standard method for reporting falls is the prospective calendar method, retrospective self-reports of falls over 6 months have shown an 84% agreement with falls reported by use of a calendar (Mackenzie, Byles, & D'Este, 2006), and a meta-analysis has shown that the recall of falls in the previous year was specific (specificity was 91% to 95%) but less sensitive (sensitivity was 80% to 89%) than a prospective method of recording falls (Ganz, Higashi, & Rubenstein, 2005).
Statistical Analyses
We compared sociodemographics, health, sensorimotor function, MMSE score, and depressive symptoms between three fall groups at Wave 1 (nonfallers, single fallers, and multiple fallers) by using a one-way analysis of variance in which variables were continuous and by using a Pearson chi-square (
2) test in which variables were categorical. In those instances in which the expected frequency of a cell was <5 for categorical variables, we used Fisher's Exact Test. We imputed missing item data by using the expectation maximization algorithm, which is an iterative optimization procedure (Graham, Hofer, & Piccinin, 1994; Schafer & Graham, 2002) available in SPSS using all available items.
We used marginal models (population-average models) to estimate between-subject effects of the relationships between measures of well-being at Wave 1 and rate of falls over the 8 years of follow-up. We used generalized estimating equations with the log link function and Poisson distribution (Fitzmaurice, Laird, & Ware, 2004). We assumed the within-subject association among the vector of repeated outcomes to have an exchangeable correlation structure (Twisk, 2003). This approach allows for the full distribution of the falls variable to be modeled and incorporates all falls data available over three occasions of measurement, thereby allowing for reliable longitudinal estimates and maximizing the use of available data.
We used a hierarchical approach to evaluate the relationship between well-being and falling, with three models being fit to the data. Model 1 adjusted for time in the study, and demographic variables (dichotomous: gender; continuous: age, years of education), to determine the observed associations independent of sampling effects. Model 2 added health variables (categorical: self-rated health, heart condition or attack, hypertension, stroke or transient ischemic attack, diabetes, psychotropic medication, current smoker, and alcohol frequency) to Model 1. It adjusted for medical conditions and health behaviors to evaluate the association between well-being and falling independent of potentially confounding physical health factors that may lead to poorer physical outcomes and mental health. Model 3 added sensorimotor variables and cognitive function (categorical: semitandem stand; continuous: visual acuity, functional reach, grip strength, and MMSE score) to Model 2. It tested whether well-being was associated with falling independent of the sensorimotor factors and cognitive function shown to affect balance and fall risk (Anstey et al., 2006; Lord et al., 1994). We used the lowest value as the reference category for all categorical variables. In those instances in which main effects of well-being measures were significant, we also tested interactions between time and these measures after we included all covariates in Model 3.
We examined the effects of within-person change in depressive symptoms, morale, or control on the rate of falls (i.e., subject-specific effects) by using random effects models (random intercept and slope specified for time) using the log link function and Poisson distribution (Fitzmaurice et al., 2004). We fitted Models 1, 2, and 3 as specified herein; the well-being variables were time varying rather than just being included at Wave 1.
To overcome potential recall bias caused by possible cognitive impairment, we reestimated all marginal models and random effects models by excluding participants who scored below the cutoff score for possible dementia (i.e., <24; see Folstein, Anthony, Parhad, Duffy, & Gruenberg, 1985) on the MMSE at any wave. There were 40 cases at Wave 1, 27 cases at Wave 2, and 31 cases at Wave 3 (total 4.2% of sample). We conducted descriptive analyses in SPSS 15.0. We conducted all longitudinal modeling of the fall rate in Intercooled Stata 9.0.
|
RESULTS |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
2(1) = 34.81, p <.01]. After we adjusted for differences in age and gender, our sample of individuals reported better SRH [F(4, 2076) = 97.363, p <.01], higher MMSE scores [F(4, 2076) = 42.033, p <.01], fewer depressive symptoms [F(4, 1988) = 45.496, p <.01], higher levels of morale [F(4, 1314) = 31.322, p <.01], and lower scores on the measure of control [F(4, 1323) = 9.321, p <.01].
Wave 1 Characteristics and Fall Rates
Table 1 displays Wave 1 characteristics of participants according to the number of falls they reported experiencing over the past 12 months at Wave 1. Compared with nonfallers, those experiencing one fall were more likely to be female and have poorer SRH and grip strength. Compared with nonfallers, multiple fallers were more likely to be older and female; to have hypertension, stroke, or transient ischemic attack; and poorer SRH, visual acuity, and grip strength. Multiple fallers were also more likely than the others to have higher depression, lower morale, and higher control scores, all indicating lower levels of well-being. At Wave 1, 24.90% of the sample reported experiencing at least one fall in the past 12 months, with this number increasing to 34.41% and 40.18% at Waves 3 and 6, respectively.
|
|
Higher scores on the Control scale at Wave 1 were associated with a higher fall rate. Specifically, a participant who scored one unit higher (i.e., worse) on the Control scale compared with another participant experienced 4% more falls. When examined as a time-varying covariate in the random effects model, control was not associated with fall rate. This was because there was no mean change in control from Wave 1 to Wave 6.
Higher scores on the CES-D scale at Wave 1 were associated with a higher fall rate. A participant who scored one unit higher on the CES-D scale compared with another participant experienced 4% fewer falls. In the random effects model, individuals whose depressive symptoms increased over the follow-up period had an increased risk of falling such that for every increase of 1 point on the CES-D score, an individual increased her or his risk of falling by 4%.
In the marginal models, there were significant Morale x Time interactions (β <.01, p <.01) and CES-D Score x Time interactions (β <.01, p <.01), indicating that those with poorer well-being at Wave 1 had a greater rate of increase in depressive symptoms over the 8-year follow-up. When we excluded those participants with MMSE scores of <24 from the analyses, we found that the pattern of results for Model 1 was unchanged in both marginal models and random effects models.
Adjustment for Demographic and Health Variables
Table 3 shows the results of models of well-being and falls as adjusted for demographic variables and health. There were no changes in the patterns of results for demographic variables after the addition of the health measures, except that education became a significant predictor of morale in the marginal models. Reporting a heart condition, taking psychotropic medication, and smoking were not associated with falling. Results for the marginal models showed that those with poorer SRH, hypertension, stroke, and diabetes at Wave 1 had higher rates of falling in the subsequent 8 years of follow-up. Results for alcohol consumption were mixed; those individuals consuming two to three drinks per week reported an increased fall rate, whereas those consuming four or more drinks per week had a reduced fall rate, compared with participants who consumed four or fewer drinks per month. For the marginal models, the Time x CES-D Score interaction remained significant (β <.01, p <.01).
|
Adjustment for Cognitive Status and Demographic, Health, and Sensorimotor Variables
Table 4 shows the results of models of well-being and falls with adjustment for demographic, health, and sensorimotor variables and MMSE score. Time, age, and gender continued to be significant, and education became significant in all marginal models. The health predictors showed similar patterns of significance as in the previous models. Visual acuity was not a significant predictor of falling, but functional reach was significant in all models, and grip strength was significant in all marginal models. Semitandem balance was significant in marginal models for control and CES-D score (those individuals able to complete the test had a reduced fall rate compared with those unable to attempt the test). MMSE score was significant in the random effects models that included control and CES-D score. The pattern of results for the well-being measures was the same as in the previous models that did not adjust for sensorimotor function and cognitive status; all three measures were associated with fall rate in the marginal models, and time-varying morale and CES-D score were associated with fall rate in the random effects models. Results were unchanged when we excluded participants with MMSE scores of <24 from the analyses. In the marginal models, the patterns of interactions with time remained unchanged after we adjusted for sensorimotor variables and MMSE score.
|
|
DISCUSSION |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
Our second hypothesis, that within-person change in well-being would be associated with change in fall rate, was supported in relation to depressive symptoms and morale but not control. Our examination of the estimated marginal means for control indicated that there was no change between Wave 1 and Wave 6 but that there was a decrease in scores at Wave 3. This explained the lack of significant effects seen in the random effects models when we used control as the time-varying covariate. It seems that the expectancy of control is relatively stable in late life (Luszcz, 1998) and that individual differences (rather than within-person change) in this variable are important for predicting falls. It is possible that domain-specific aspects of control relevant to physical function may show more fluctuation as declines in mobility occur. The discrepancy in results between analyses of between-person differences and within-person change in control highlights the importance of using both approaches when one is investigating the significance of predictors in longitudinal research.
A particular strength of this study was the control for a wide range of covariates, and hence the results demonstrate an independent effect of well-being on fall risk in community-dwelling older adults. As we expected, a number of the potentially confounding variables were significantly associated with an increase in fall rate, including SRH, hypertension, stroke, diabetes, functional reach, grip strength, and semitandem stand. The association between alcohol consumption and falling depended on the level of consumption, which is suggestive of the nonlinear association seen between alcohol consumption and cognition (Rodgers et al., 2005). Previous studies have shown that depressive symptoms are associated with some of the physical function measures used in the ALSA (Penninx et al., 1998). Depressive symptoms are associated with antidepressant use and with cognitive decline (Wilson, Mendes De Leon, Bennett, Bienias, & Evans, 2004), which has in turn been associated with fall risk (Anstey et al., 2006). Therefore, by controlling for these potential confounds, the present study presents a robust finding in relation to well-being and falls.
Our third hypothesis was that the effect of well-being would be reduced after we adjusted for the large number of covariates in our models. This hypothesis was weakly supported, in that effect sizes did reduce, but in several cases they remained quite substantial in clinical terms as well as statistical terms. For example, in the marginal models, even after we adjusted for all covariates, a reduction of 1 point on the Morale scale was associated with a 6% increase in fall rate over 8 years. This suggests that the effect of well-being on falling was not due to indirect effects in terms of its effect on the measures of health and sensorimotor function included in this study.
Study limitations include the lack of a clinical diagnosis of depression, and the retrospective, self-report nature of our falls measure. This was overcome somewhat by the exclusion of participants with possible dementia and the consistency of results across analyses. The strengths of the study include the large sample size, multiple follow-ups over a long interval, and, compared with previous studies, the adjustment for a wide range of potentially confounding variables. Moreover, to our knowledge this is the first study to include time-varying measures of well-being in predicting falls in a population-based sample over a prolonged period.
The results of the present study may be viewed within the broader context of the association between psychological and physical well-being in later life. The fact that morale predicted falling suggests that researchers should expand the conceptualization of risk factors for functional outcomes to consider the psychosocial context of the individual. Just as social support is protective against dementia (Fratiglioni, Paillard-Borg, & Winblad, 2004), so might strategies to enhance psychological well-being protect against falling in late life. The Philadelphia Geriatric Center Morale Scale has been associated with neighborhood and social characteristics (Breeze et al., 2005) and is used as a measure of quality of life (Wong, Woo, Hui, & Ho, 2004). Compared with morale, the influence of depressive symptoms on fall risk may be more driven by biologically based mechanisms, particularly those that contribute to poor concentration and reduction in energy levels. It is possible that deficits in executive function associated with depression are also associated with an increase in the risk of falling (Holtzer et al., 2007).
Potential mechanisms for the association between well-being and falls may include shared biological factors operationalized as allostatic load. Better self-monitoring and chronic disease management resulting from higher levels of psychological well-being may be protective against accidental falls. Other factors not examined in this study such as neighborhood, social support, and access to recreation and services may also contribute to the overall psychosocial context that predisposes individuals to falling.
An indication of the need for research on psychosocial factors associated with falls is seen by surveying the recent Cochrane review of interventions to prevent falling (Gillespie et al., 2003). That review identified studies of exercise, improvements in home safety, medication withdrawal, nutritional and hormone supplementation, vision assessments, cardiac pacing, and multifactorial interventions, but psychosocial interventions were not included.
The results of this study suggest that falls-risk assessments should include brief measures of well-being. Psychosocial interventions that focus on improving positive aspects of well-being, as well as reducing depressive symptoms, may also reduce the incidence and prevalence of falls in older adults.
|
Acknowledgments |
|---|
|
Footnotes |
|---|
Received for publication July 6, 2007. Accepted for publication January 14, 2008.
|
References |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
| ||||||||||||||||||||||||
| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
