| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
| ||||||||||||||||||||||||||||||||
a Department of Preventive Medicine and Biometrics, University of Colorado School of Medicine, Denver
Correspondence: Lucinda L. Bryant, PhD,MSHA,MBA, Center for Health Services Research, Division of Health Care Policy and Research, University of Colorado Health Sciences Center, 1355 S. Colorado Boulevard, Suite 306, Denver, CO 80222. E-mail: lucinda.bryant{at}uchsc.edu.
Decision Editor: Laurence G. Branch, PhD
 |
Abstract |
|---|
 TOP Abstract MethodsResultDiscussionReferences |
|---|
Key Words: Cohort study • Ethnic disparities • Attributable risk • Aging • Functional disability
The aging of the United States population and its increasing ethnic diversity have stimulated interest in understanding the processes that lead to poorer and sometimes ethnically disparate health outcomes among older people. Because Hispanic persons now comprise the second largest and fastest growing minority in the United States, with an increase from 10.3% of the U.S. population in 1993 (U.S. Bureau of the Census 1995
) to 12.5% in 2000 (U.S. Bureau of the Census 2001), there is special reason to attend to the needs of this segment of the population.
Functional disability predicts negative outcomes of aging such as mortality and institutionalization (Manton 1988; Mor, Wilcox, Rakowski, and Hiris 1994; Reuben, Siu, and Kimpau 1992). Although the presence of any limitation suggests possible disability, the onset of limitations may even more specifically signal the beginning of loss of independence (Mor et al. 1994). Effective interventions to delay disability require identification of risks that can be reduced.
Previous studies of functional decline, generally including both incident and increased dependence in basic and instrumental activities of daily living (ADLs and IADLs), have identified a number of risk factors: older age; female sex; less education and income; greater disease burden including cognitive impairment and depression; poorer self-rated health; reduced physical performance; poorer health behaviors including alcohol consumption, smoking, nutrition, and physical activity; and psychological and social factors such as less self-efficacy and lower levels of social support (see Stuck et al. 1999, for a comprehensive review of the literature). Not all risks can be reduced or eliminated, but interventions at both individual and community levels may help older people improve their health behaviors and associated characteristics such as strength, body mass, and pulmonary function (King, Rejeski, and Buchner 1998; Nigg et al. 1999; Stewart et al. 1998).
We have previously reported somewhat greater prevalent and incident functional dependence among Hispanic than among non-Hispanic White (NHW) elders—age- and gender-adjusted ADL prevalence odds ratio (OR) 1.39, 95% confidence interval (95% CI) 1.01–1.92 (Hamman et al. 1999); age- and gender-adjusted IADL prevalence OR 1.49, 95% CI 1.16–1.93 (Shetterly, Baxter, Morgenstern, Grigsby, and Hamman 1998); incidence difference not statistically significant (Bryant, Shetterly, Baxter, and Hamman 2002)—in the entire San Luis Valley Health and Aging Study (SLVHAS) population. Although the ethnic differences in disability outcomes are not large and may not reflect the large disparities in observed risk factors (the "Hispanic epidemiologic paradox" identified by Markides and Coreil 1986), it is important to determine which factors contribute to increasing disability in order to design effective interventions, in NHW as well as Hispanic communities. In the analysis reported here, we have identified modifiable risk factors associated with incident functional dependence, compared their effects in multivariable models, and estimated the percent risk attributable to each individual factor, by ethnicity.
|
Methods |
|---|
|
TOPAbstractMethodsResultDiscussionReferences |
|---|
), the community-based SLVHAS examined health and disability among older Hispanic and NHW residents of rural Alamosa and Conejos counties in southern Colorado. These two counties include about 2000 square miles (3200 square km) with a 1990 population of 21,070 (U.S. Bureau of the Census 1992). Unlike much of the U.S. Hispanic population, the Hispanic residents of the San Luis Valley (SLV), who make up 48% of the population, report only 31% Mexican American origin; the remaining majority self-identify themselves as "other Spanish or Hispanic" (U.S. Bureau of the Census 2001). These lifelong SLV residents are descendants of the second oldest inhabitants of the area, after Native Americans, and there is little new in-migration. The SLV pattern of subgroup identification is typical for the central southwestern states (especially New Mexico, Arizona, Utah, and Nevada). These rural Hispanic populations have received little attention from researchers, yet almost 20% of all Hispanics live outside of standard metropolitan statistical areas (Bean and Tienda 1987). The SLVHAS is one of a very few studies of rural Hispanic elders and the only one with a non-Hispanic comparison group.
All occupied households were enumerated in 1992 and 1993, with a 97.2% response rate. Eligibility requirements for the study included age of at least 60 years old, residence in either county, and Hispanic or NHW ethnicity. The 1980 U.S. Census question, "Are you of Spanish or Hispanic origin or descent?," defined Hispanic ethnicity (U.S. Bureau of the Census 1982). Differential sampling within age and ethnic strata ensured appropriate numbers of subjects for planned ethnic contrasts. Bilingual interviewers collected responses from 1,358 community-dwelling participants (81.1% of sampled persons). Interviewers sought proxy informants for 188 participants unable or unwilling to respond, primarily individuals identified as cognitively impaired based on Folstein Mini-Mental State Examination (MMSE) scores of less than 18 (Folstein, Folstein, and McHugh 1975). The analysis here excludes those who required proxy assistance at baseline. Approximately 22 months later, interviewers revisited the participants, with responses from 93.6% of the 1103 surviving baseline self-responders. There were 429 Hispanic and 336 NHW elders, 75.4% and 72.7% of surviving self-responders in each ethnic group, respectively, who had reported no dependence in either ADLs or IADLs at baseline. An additional 37 of the 858 baseline dependence-free respondents had died by follow-up, and 56 others refused the follow-up visit. Fourteen individuals had missing ADL or IADL data at follow-up, leaving 751 individuals who were free of dependence at baseline and had follow-up information for this analysis.
Measures
Outcome.
We used ADL and IADL questions from the 1984 National Health Interview Supplement on Aging (Fitti and Kovar 1987) that measured functional status using the syntax, "Because of a health or physical problem, do you have any difficulty with" ADL tasks (eating, bathing, dressing, toileting, transferring between bed and chair, walking across a room, and getting outside) and IADL tasks (shopping, transportation, preparing meals, using the telephone, taking medication, managing money, and doing light and heavy housework). If the respondent reported any difficulty, the interviewer then asked, "By yourself, how much difficulty do you have: some, a lot, or are you unable to do it?" An "unable" response elicited a further question: "Can you do it with help from a person or equipment?" We defined disability for this analysis as dependence, either the inability to perform an ADL or IADL task or the need for assistance to do so.
Base-Model Covariates.
A base model of baseline covariates included demographic information (age, sex, ethnicity, years of education), cognition coded as poor (<24) versus good (
24) on the MMSE, and the number of reported disease conditions (arthritis, cancer, heart attack, mild or severe stroke, angina, diabetes, Parkinson's disease, high blood pressure, heart failure, pulmonary disease, cirrhosis, kidney failure, osteoporosis, seizure, migraine, depression, angioplasty or blood vessel surgery, and difficulty with hearing or vision). We chose this summed measure rather than condition-specific indicators for two reasons. First, the choice allows for comparisons with existing studies of risk factors for functional decline; all 78 of the studies reviewed by Stuck and colleagues 1999 used this methodology. Second, the purpose of this study is to assess the importance of modifiable behavioral risk factors, adjusted for a variety of covariates including disease burden. It would also be interesting and useful to determine the degree to which treatment of chronic conditions such as arthritis, diabetes, or depression might decrease incident dependence, but that is not the focus of this study.
Modifiable Risks.
To the base model we added modifiable risk factors: smoking, a nutritional risk score, a measure of physical activity, and the report of any falls within the previous 12 months. We did not include heavy drinking because alcohol consumption in this population is very low. SLV investigators previously developed a nutritional risk score (Marshall et al. 1999) that is similar to the Nutrition Screening Initiative checklist (Nutrition Screening Manual 1991). The score sums the weighted values of 10 components: low numbers of servings of fruit, vegetables, or milk; diet change due to illness; fewer than two meals per day; more than two alcoholic drinks per day; dental problems; lack of money for food; eating alone; more than two medications; unwanted weight gain or loss; and difficulty shopping, preparing meals, or eating. Because ADLs and IADLs are the outcome for this study, we modified the published index by removing the last item, reducing the maximum possible risk score from 21 to 19. The physical activity measure sums self-reported 1-year recollections of an extensive list of home-, work-, and leisure-based activities translated into metabolic equivalents (METS), the number of kilocalories per kilogram of body weight expended per hour. The measure was adapted from the Minnesota Heart Health Program's Cardia (Jacobs, Hahn, Haskell, Pirie, and Sidney 1989) to capture additional information on tasks common in this rural community. We categorized the data to compare the lower-performing three quartiles with the highest, using a cut point of 4780 METS expended during the year.
Analysis
Bivariate comparisons, with Student's t, 
2, and Mann-Whitney U tests as appropriate, assessed baseline ethnic differences in the variables. Logistic regression models constructed with SAS statistical software version 8.2 (SAS Institute, Inc., Cary, NC) provided estimates of the contributions of the independent risk variables to incident ADL and IADL dependence, adjusted for a base model of demographic and comorbidity variables. We first assessed each modifiable risk factor's individual relationship with the outcome. We then constructed two multivariable models to combine the factors' effects and tested for ethnic interactions. The first multivariable model contained the base model and modifiable risk factors. The second added an indicator of baseline difficulty with any ADL or IADL task. The first model underestimates the effect of pre-existing disability on incident dependence. The second, an attempt to correct the underestimate, errs in the opposite direction because the baseline difficulty category includes some prior effects of the studied risk factors on the disablement process and consequently "subtracts" them from the model. Together the two models provide a range of estimates of the effects of the risk factors on the outcome. Finally, in an analysis not shown, we examined the 532 individuals with no difficulty at baseline, 78 of whom developed incident dependence, to be sure that no dramatic differences existed between them and those with preexisting ADL or IADL difficulty.
We then computed the population attributable risk percent (PAR%) by ethnicity for each risk factor to assess the potential impact of interventions designed to reduce those risks, using the following computation (Gordis 1996; Rothman and Greenland 1998):
PAR% = [Pe(RR – 1)] / [1 + Pe(RR – 1)],
where Pe measures the percent of the population exposed to the particular risk (e.g., smoking) and RR represents the relative risk of the outcome associated with the risk factor in the entire population. Zhang and Yu 1998 recommend adjusting OR measures of relative risk in cohort studies of high-prevalence outcomes only if unadjusted ORs exceed 2.5. No effect sizes in our analyses exceeded 2.5, so we used unadjusted ORs from models with and without the baseline difficulty indicator as estimates of relative risk.
We computed the PAR% for each modifiable risk factor separately, adjusted for base model variables but not for possible interaction or confounding among the risk factors. Assigning attributable risk in multifactor situations requires generally unavailable knowledge of the risks associated with each possible combination of risk exposures and the joint distribution of numbers of persons so exposed in the population. Methods have been suggested to provide multifactor estimates (Walter 1983), but our purpose in providing PAR%s here was to estimate the degree to which risk-specific focused interventions might slow functional decline in older persons, so we have chosen to present single- rather than multifactor PAR%s.
|
Result |
|---|
|
TOPAbstractMethodsResultDiscussionReferences |
|---|
).
|
|
; Seeman et al. 1995). Older age, greater number of comorbidities, and female sex had statistically significant associations with an increased likelihood of incident dependence. Each modifiable risk factor except physical inactivity (p = .07) predicted incident dependence at a 95% significance level when added alone to the base model. Adding the measure of baseline difficulty to the base model reduced the size of the risk factors' effects modestly (by 0.15–0.40 OR, not shown in table); the CIs for physical inactivity and falls included 1.0.
|
Ethnicity did not contribute significantly to any model, and we found no significant interactions between ethnicity and the modifiable risk factors. Ethnicity modified the effect of the number of comorbid conditions (p = .05), with a pattern that suggested a somewhat greater association with incident disability in NHW than Hispanic elders (OR 2.5 vs 1.7, for 2 comorbid conditions vs <2). There were no other significant interactions.
The nutritional risk scale combined a number of factors that may affect the adequacy of elders' nutrition. Separate analyses (not shown) identified the most important predictors among the nutritional risk scale items: oral health problems (p < .0001, adjusted for base model variables), usually eating alone (p = .02), and multiple medications (p = .001).
Population Attributable Risk
PAR%, also called attributable or etiologic fraction, identifies the percentage of a population that might have avoided an outcome had there been no exposure to a specified risk related to that outcome. It depends not only on the risk factor's relationship with the outcome but also on the prevalence of the risk factor in the population. Ethnicity did not contribute significantly to the models of incident dependence, that is, there were no ethnic differences in relative risks. The ethnic differences in PAR%s seen here (Table 4 ) derive from differential exposure to risks, not to disparate relative risks. PAR%s derived from risk estimates adjusted for baseline ADL or IADL difficulty are somewhat lower than those derived from difficulty-unadjusted models but still suggest substantial potential benefit from ethnic-sensitive focused interventions. The largest percentages among the factors studied here were associated with low levels of physical activity (PAR% 25–32 for Hispanic elders, 22–28 for NHW elders) and greater nutritional risk (PAR% 14–23 for Hispanic elders, 8–22 for NHW elders), followed by smoking (PAR% 15–17 for Hispanic elders, 8–10 for NHW elders) and falls (PAR% 9–12 for Hispanic elders, 10–14 for NHW elders). The perhaps counterintuitive difference between high and moderate nutritional risk—PAR%s higher for moderate than for high risk—occurs because of the substantially greater prevalence of moderate than high risk (see Table 1 ). Attributable risk percents were larger for Hispanic than NHW elders in all categories except falls.
|
|
Discussion |
|---|
|
TOPAbstractMethodsResultDiscussionReferences |
|---|
; Shetterly et al. 1998) but not significantly different incidence rates (Bryant et al. 2002). Participants included in the analyses here, that is, those who were free of dependence at baseline, similarly reported identical levels of incident dependence at follow-up (23%, unadjusted for age or gender), even though Hispanic elders had greater exposure to most of the factors that predicted incidence, including all but one of the modifiable risks (falls) that are the focus of this report. We cannot from the analyses presented here say that any of these factors work differently for Hispanic elders. There are a number of possible explanations for this apparent "Hispanic epidemiologic paradox," some of which we have suggested previously (Bryant et al. 2002): Function may decline in increments too small for ethnic differences to be detected in less than 2 years; Hispanic adults become disabled at earlier ages (higher baseline prevalence), so effects of these risk factors may already have occurred, removing the most "susceptible" individuals from the study sample; detecting the effects of ethnic interactions may require more power than available for these analyses; or other factors not included in this model, possibly related to the manner in which biological systems deal with risk factors, may be at the root of the paradox. Answers will require further research. Older age, female sex, less education, and a larger number of comorbid conditions significantly predicted incident ADL or IADL dependence after 22 months. Addition of the proposed modifiable risk factors, either individually or combined in the multivariate model, increased the predictive power of the model but not substantially. ORs associated with each risk factor did not vary greatly between single-risk and multivariate models, which suggests that the factors act independently and cumulatively. Baseline difficulty with at least one ADL or IADL task nearly quadrupled the risk of incident dependence compared to absence of difficulty in this dependence-free group and somewhat reduced the size and significance of the effects of the modifiable risk factors. As suggested earlier, difficulty status likely incorporates earlier effects of these same risks. We cannot accurately assess the effect of that earlier exposure, because we cannot know what other characteristics not included in this model may also have contributed. We can, however, estimate that the effects of the modifiable risk factors lie in a range between the difficulty-adjusted and the unadjusted rates and PAR%s.
Study participants whose reported levels of physical activity at baseline lay in the lowest three quartiles were 1.5–1.7 times more likely to report new dependence in ADLs or IADLs after 22 months than those with higher activity levels. Other observational studies have also found that physical activity predicts the maintenance of daily living activity abilities, in populations such as the Longitudinal Study on Aging (Mor et al. 1989; Wolinsky, Stump, and Clark 1995) and the National Health and Nutrition Examination Survey (Hubert, Bloch, and Fries 1993). It is possible, however, that lower activity levels reflect early difficulties in functioning rather than a choice to be sedentary. When we adjusted our models for those persons who reported any difficulty in ADLs or IADLs at baseline, the protective association of physical activity diminished somewhat and became no longer significant, confirming the entanglement of these elements in observational studies. Intervention trials that have found that increased physical activity maintains or improves functional ability support a predictive rather than a reflective interpretation (Cress et al. 1999; Hunter et al. 1995; Teixeira-Salmela, Olney, Nadeau, and Brouwer 1999), but not all trials have found a positive effect of physical activity on disability (Chandler, Duncan, Kochersberger, and Studenski 1998). The Frailty and Injury: Cooperative Studies of Intervention Trials (FICSIT) may help to untangle these relationships (Judge, Schechtman, and Cress 1996).
Persons in this rural biethnic population with high levels of nutritional risk were 1.8–2.1 times more likely than those with low risk to become newly ADL or IADL dependent; even moderate levels of risk substantially increased the likelihood. These results confirm previous reports that poor nutritional status leads to declining health (Bidlack 1990), with greatest risk among minority and rural populations (Dwyer 1991; U.S. Public Health Service 1988; Windham, Wyse, Hansen, and Hurst 1983). Two of the three most important scale items suggest problems with medical management of multiple medications and access to oral health care. The other item concerns a primarily social deficit, not having companionship when eating.
Being a current smoker nearly doubled the risk of incident ADL or IADL dependence. Although smoking in this population is less prevalent than the other studied modifiable risks, it is more prevalent than among the general 1994–1995 U.S. population of people aged 65 and older (18% here vs 12% in the general population; National Center for Chronic Disease Prevention and Health Promotion 2001). Older smokers sustain proportionately more disease and disability than younger ones because of the duration of cumulative injury or change related to tobacco use, but cessation of smoking decreases the risks even among older adults (Burns 2000).
Many studies have reported the association between falls and daily living activity disability or functional decline (Ory et al. 1993; Tinetti and Williams 1998). This study found that any fall in the 12 months prior to baseline increased the risk of new ADL or IADL dependence by 37–51%. The prevalence of falls in this population (24% among Hispanics, 29% among NHW) mirrors that in the overall U.S. population aged 65 and older (Tinetti, Speechley, and Ginter 1988).
These four modifiable risks meet at least the first two of three criteria for prevention that Tinetti and Williams 1998 have noted about falls: frequency, morbidity risk, and evidence of effective interventions. Estimates of PAR% quantify the joint burden of frequency and risk. Because so many elders reported each of these risks, the amount of incident functional dependence that might be prevented by reducing risk-related situations and behaviors is large. Proposing to increase physical activity among all elders to the levels now reported by the most active quartile sets an unreachable goal, but the PAR% of 26–36% suggests a great opportunity to affect disability incidence. The greater Hispanic burdens from nutritional risk and smoking identify pathways that may help to reduce ethnic disparities in morbidity as well as disability.
There are limitations to this study. We have included many factors in our models that other studies have found important, but it is possible that additional factors not included may have important relationships with incident disability. Our models do not attempt to identify the psychological and social supports and barriers that affect different health behaviors, nor do they address the larger social structure factors that have an impact on individual behavior (e.g., tobacco advertising or the availability of high-fat fast food). Our goal was to assess the importance in two ethnic groups of individual risk factors that appropriate interventions might affect.
The results of this study indicate that problems related to inadequate nutrition, low levels of physical activity, smoking, and preventable falls offer opportunities to reduce the incidence of ADL and IADL disability, especially among Hispanic elders, who report a greater prevalence of these risks. The observed negative effects of these factors appear to be cumulative, leading to impairment in multiple domains that compromises compensatory ability, as Tinetti, Inouye, Gill, and Doucette 1995 suggested. Evidence from randomized intervention trials that targeted inactivity and smoking suggests that appropriate interventions can improve functional status. Characteristics of successful programs have included assessments of risk and clinical problems, personal attention and tailoring to individual needs, appropriate choice of provider determined by the individual's condition, supervision, health education, and sustained follow-up (Wagner 1997). Our study reinforces the importance of developing such interventions, with special attention to elders at greater risk due to socioeconomic and cultural factors or lower levels of education.
|
Acknowledgments |
|---|
Received for publication June 7, 2001. Accepted for publication March 28, 2002.
|
References |
|---|
|
TOPAbstractMethodsResultDiscussionReferences |
|---|
This article has been cited by other articles:
 |
|
 |
|
  M. K. Peek, J. P. Stimpson, A. L. Townsend, and K. S. Markides Well-Being in Older Mexican American Spouses Gerontologist, April 1, 2006; 46(2): 258 - 265. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||
| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
