- Email: [email protected]
Association of Comorbidity with Disability in Older Women
J Clin Epidemiol Vol. 52, No. 1, pp. 27–37, 1999 Copyright © 1999 Elsevier Science Inc. All rights reserved.
0895-4356/99/ $–see front matter PII S0895-4356(98)00124-3
Association of Comorbidity with Disability in Older Women: The Women’s Health and Aging Study Linda P. Fried,1,* Karen Bandeen-Roche,1 Judith D. Kasper,1 and Jack M. Guralnik2 for the Women’s Health and Aging Study Collaborative Research Group 1The
Johns Hopkins Medical Institutions, Baltimore, Maryland; and 2Epidemiology, Demography and Biometry Program, National Institute on Aging, Bethesda, Maryland ABSTRACT. There is substantial evidence that physical disability results from chronic diseases and that the number of chronic diseases is associated with the presence and severity of disability. There is some evidence that interactions between specific diseases are of import in causing disability. Beyond arthritis, however, little is known of the disease pairs that may be important to focus on in future research. This study explores the associations between multiple disease pairs and different types of physical disability, with the objective of hypothesis development regarding the importance of disease interactions. The study population comprised a representative sample of 3841 women 65 years and older living in Baltimore, screened for participation in the Women’s Health and Aging Study. The study design was cross-sectional. An interviewer-administered screening questionnaire was administered regarding self-reported physical disability in 15 tasks of daily life, history of physician diagnosis of 14 chronic diseases, and MiniMental State examination. Task difficulty was empirically grouped into six subsets of minimally overlapping disabilities, with a comparison group consisting of those with no difficulty in any task subset. Multiple logistic regression models were fit assessing the relationship of major chronic diseases and of interactions of disease pairs with each disability subtype and with any disability, adjusting for confounders. Fourteen percent of the population reported mobility difficulty only; 5%, upper extremity difficulty only; 9%, both of these difficulties but no others; 7%, difficulty in higher function but not self-care tasks; 7%, self-care task difficulty but not higher function tasks; and 15%, difficulty in both higher function and self-care (weighted data). Almost all in the latter three groups had difficulty, as well, in mobility or upper extremity tasks. In regression models, specific disease pairs were synergistically associated with different types of disability. For example, important disease pairs that recurred in their associations with different disability types were the presence of arthritis and visual impairments, arthritis and high blood pressure, heart disease and cancer, lung disease and cancer, and stroke and high blood pressure. In addition, the type of disability that a disease was associated with varied, depending on the other disease that was present. Finally, when interactions were accounted for, many diseases were no longer, in themselves, independently associated with a given type of disability. Partitioning disability into six subtypes was more informative in terms of associations than was evaluating a summary category of “any disability.” These findings provide a basis for further hypothesis development and testing of synergistic relationships of specific diseases with disabilities. If testing confirms these observations, these findings could provide a basis for new strategies for prevention of disability by minimizing comorbid interactions. j clin epidemiol 52;1:27–37, 1999. © 1999 Elsevier Science Inc. KEY WORDS. Aging, comorbidity, disability, chronic disease, physical function, geriatrics
INTRODUCTION Physical disability results primarily from chronic disease [1– 3] and is highly prevalent in older adults. The presence of multiple diseases increases the likelihood of developing physical disability. As the numbers of diseases increase, *Address correspondence to: Linda P. Fried, M.D., M.P.H., Professor, Medicine and Epidemiology, Welch Center for Prevention, Epidemiology and Clinical Research, Suite 2-600, 2024 E. Monument Street, Baltimore, Maryland 21205. Accepted for publication on 22 July 1998.
there is increasing risk of difficulty with activities of daily living (ADLs) [3–5], instrumental ADLs (IADLs) [3,5,6] and mobility [3,7–10]. There is related evidence that specific diseases can cause specific types of disability [1,2,8,11], and some studies suggest that interactions among specific diseases may be of importance [3,8,12]. This has been explored for arthritis, with results indicating that when arthritis occurs in the presence of other specific diseases, there is substantially greater risk of mobility and transfer difficulty than with arthritis alone [12] and that arthritis and other diseases act synergistically on some functions [8]. Little has
L. P. Fried et al.
28
been reported in this regard, however, for diseases other than arthritis. It is also not known, for diseases other than arthritis, whether specific combinations of diseases cause different types of disability. If so, interventions to prevent or minimize disability would need to take into account both the specific comorbidities and the type of disability in order to be optimally effective. The Women’s Health and Aging Study (WHAS) has recently completed screening of a representative sample of community-dwelling women 65 and older, providing information on major chronic diseases, cognitive function, and disability. This article evaluates these data to provide a broad overview of the relationships of individual diseases and pairs of diseases to different types of disability in a representative population of women 65 years and older. Specifically, we sought to explore how combinations of diseases may be related to different types of disability, over and above the independent contributions of each disease to disability. Such information can provide insight into the clinical profiles of potentially important disease pairs and into how these differ from the association between a single disease and disability, and can also suggest avenues for future hypothesis-driven research that can specifically delineate opportunities for intervention.
METHODS Study Population The study population consists of a representative sample of women 65 years and older who were screened to determine eligibility for the WHAS. The WHAS is a study of the causes and course of disability in moderately to severely disabled older women living in the community. These analyses, however, use data on all women who were screened and are not limited to the subgroup selected for the full study. The sampling frame was obtained from the Health Care Financing Administration Medicare Enrollment files for the eastern half of Baltimore City and County, consisting of 12 zip code areas. A random sample of older women was drawn, stratified by age (65–74, 75–84 and 85 and older). Those sampled were contacted first by letter and then received a screening interview in their homes. Eighty-one percent of the 5316 women sampled consented to be screened, and the study population considered for analysis here consists of 3841 women who completed the screening interview. Study design and sampling and data collection have been previously described in detail [13]. Data Collected The interviewer-administered, standardized screening questionnaire ascertained, through self-report, the participants’ demographic characteristics and whether a physician had diagnosed any of the following 14 chronic diseases and conditions: myocardial infarction, angina, congestive heart
failure, other heart disease, high blood pressure, diabetes, arthritis, stroke, cancer, lung disease, vision problems, hearing problems, Parkinson’s disease, or a hip fracture. A MiniMental State examination was administered in standardized fashion to assess cognitive function [14]. Self-reported disability was ascertained by asking about difficulty in 15 tasks of daily life, which were selected to describe four defined domains of physical functioning [2]: mobility and exercise tolerance-demanding tasks (getting in or out of a bed or chair, climbing up 10 steps without resting, walking two to three blocks, doing heavy housework); upper-extremity tasks (raising arms overhead, grasping or handling, lifting and carrying 10 pounds); higher-function tasks essential to independent living (using a telephone, light housework, meal preparation, shopping); and basic self-care tasks (bathing/showering, dressing, eating, toileting). Data Analysis To achieve our analytic goal of describing the relationships of comorbid diseases (defined as two diseases jointly present in the same individual) with disability in our population, we proceeded in stages. First, we examined the prevalence of disease in our population and whether a priori grouping of related comorbid diseases would be useful. Second, we sought to characterize major distinct patterns of reported disability in our population. Third, we examined the bivariate associations of each disease with disability, overall and by subtype. Fourth, we calculated, in multivariate analyses, the independent associations of the diseases with disability, and we compared these with the separate associations calculated in the previous step. Fifth and centrally, we sought to describe relationships between comorbidity and disability by determining which pairwise combinations of diseases substantially modified the independent effects of the individual diseases on disability and how these mutual effects varied in association with different types of disability. Finally, statistical diagnosis was undertaken to ensure that our multivariate analyses adequately described the data. All prevalence analyses were performed both unweighted, to identify instances of sparse data, and weighted, to represent prevalence appropriately in the population from which the sample was drawn. In any specific table, cell, or calculation, from 4 to 27 women were excluded owing to missing information. DISEASE DESCRIPTION AND SUMMARY. Prevalences of disease were determined and expected coprevalences of diseases were calculated as (prevalence of disease A) 3 (prevalence of disease B). Associations among each pair of chronic conditions were assessed; the strongest observed associations were those between pairs of cardiac diseases, that is, myocardial infarction, angina, and congestive heart failure (odds ratios 5 5.6–8.0). To diminish collinearity, these three diseases were combined into one binary heart disease
Association of Comorbidity with Disability
29
variable for use in regression models. All other associations were moderate or weak, indicating limited utility of further summary. This relative lack of association indicated that interactions would be an appropriate way to model the effects of diseases in combination, as opposed to further summary. In subsequent analyses, we characterized comorbidity as the presence of disease pairs. Parkinson’s disease and hip fracture were excluded from subsequent analyses because of their very low prevalence. SUMMARY OF PHYSICAL FUNCTION. In previous factor analyses [2], it was found that tasks of daily life aggregated into four groups in which difficulty in one task was associated with difficulty in the others, suggesting a physiologic grouping. These overall groups involved difficulty with (1) mobility and exercise tolerance-demanding tasks, (2) upperextremity tasks, (3) higher-function tasks (a subset of IADLs), and (4) basic self-care tasks (a subset of ADLs). Use of these groupings to summarize function improved specificity in evaluating associations between disease and disability. These domains were therefore used as the initial basis for these analyses. Data exploration in this study showed substantial overlap between domains, however, with individuals frequently reporting difficulty in more than one of these four groups. As a result, we investigated further the frequency of overlaps for each of 16 possible combinations of difficulty. This process yielded six distinct groupings of disability: mobility difficulty only; upper-extremity difficulty only; both of these but no other difficulty; difficulty in higher function tasks, but not self-care tasks; difficulty in self-care tasks but not higher function tasks; and difficulty in both higher function and self-care tasks (Table 1). Among those in the latter three groups, 95% reported difficulty in mobility and/or upper-extremity function as well. The least over-
lap was seen between those with mobility difficulty and those with upper-extremity difficulty. To maximize specificity, the study population was stratified into the six disability groups just outlined and a seventh group that reported no difficulty in any of these tasks. ASSOCIATIONS OF EACH DISEASE WITH DISABILITY. Polytomous logistic regression was used to determine the associations between disability as the dependent variable (seven categories) and each of nine diseases separately: arthritis, any heart disease (including report of angina, myocardial infarction or congestive heart failure), stroke, high blood pressure, diabetes, lung disease, cancer, hearing impairment, and visual impairment. Each of these models also included age (65–74, 75–84, 85 and older), race (white or AfricanAmerican), educational level (less than high school graduate versus high school or greater), and MiniMental State Examination score (,24 versus 24 or greater) as potential confounders. To describe these associations, we estimated and contrasted prevalences of each disability type among women reporting each disease. To obtain these prevalences, we weighted and averaged over confounders in each polytomous model, then back-calculated the resulting log odds parameters to prevalences. This approach provides a broad overview of associations between individual diseases and disability.
INDEPENDENT AND JOINT ASSOCIATIONS OF DISEASE WITH DISABILITY.
We used multiple logistic regression analyses to describe how the nine diseases were independently associated with, and how disease co-occurrences were related to, selfreported disability. To facilitate model checking, we fit binary logistic regression models for each of the six disability outcomes; persons with no reported difficulty served as the
TABLE 1. Prevalencea of different types of physical disability Women’s Health and Aging Study
screenees (n 5 3841)
Domains of disability—difficulty in: Mobility/exercise tolerance-demanding tasks
Upper-extremity tasks No
No Yes No Yes Yes aUnweighted.
Higher-function tasksb Self-care tasks No Yes
No 40.0%
(1536)7
0.7% (25)5
No
4.2%
Yes
0.5% (20)5
No Yes
13.5%
(159)2 (518)1
2.2% (85)5
Yes 0.9% (33)4 0.0% (1)6 0.1% (3)4 0.1% (5)6 3.0% (114)4 2.3% (87)6
No
8.5%
(326)3
4.5% (172)4
Yes
4.1% (158)5
15.5% (595)6
b Groupings of disability: 1 5 mobility difficulty only; 2 5 upper-extremity difficulty only; 3 5 both mobility and upper extremity difficulty; 4 5 higher function task difficulty but no self-care difficulty (may have difficulty in mobility or upper extremity tasks); 5 5 Self-care task difficulty but no higher function difficulty (may have difficulty in mobility or upper extremity tasks); 6 5 Both higher-function and self-care task difficulty (may have difficulty in mobility or upper extremity tasks); 7 5 No difficulty with any task assessed.
30
reference group in each model. For comparison, and to understand the relationships with comorbidity that would be apparent if using the most nonspecific characterization of disability, models using the outcome of any self-reported difficulty in one or more of the 15 tasks of daily life (versus no difficulty) were also fit (“Any difficulty” model). Logistic regressions were based on 3800 women with nonmissing information for dependent and independent variables. Models were first analyzed for the nine disease main effects: heart disease, arthritis, stroke, lung disease, diabetes, high blood pressure, cancer, hearing impairment and visual impairment; these were adjusted for age, race, education and MiniMental Score. Then, two-way interactions were used to model the effects of pairwise disease co-occurrence, using multiple logistic regression models. Here, our goal was to explore which interactions were particularly predictive of disability. Potential two-way interactions were added to the main effects model by a stepwise process, using entry and retention alpha levels of 0.10. Odds ratios were calculated from these regression models. In reporting, we focus on the strong, positive interaction terms. These indicate synergistic associations in which having both diseases was associated with substantially higher risk of disability than was expected by multiplying disease-specific risks. To compare disability risk for different, significantly interacting disease pairs, plots were constructed. These illustrate the odds ratios for each disability type (versus no disability) for the joint presence of two diseases and for each disease separately, compared with the reference population reporting no disease(s). Plotted odds ratios were derived from our final, multivariate fitted models. RESULTS Description of Disease and Disability in Population Among the study population of 3841 community-dwelling women 65 to 101 years of age, 25% were African-American and 75% were white. Thirty-one percent were married, and 35% reported 8 or fewer years of education, and 19 percent, more than a high school education. The mean MiniMental score was 26, with 11% and 4%, respectively, having scores of 18 to 23 and less than 18. Self-assessed health was reported as excellent, good, or fair/poor by 37%, 34% and 29%, respectively. There was a high frequency of chronic disease, with an average of three diseases per woman and a range of 0 to 11. Figure 1 displays the frequency of co-occurring diseases in the population. Only 5% reported having none of the 14 conditions assessed, whereas 81% reported two or more chronic conditions, and 18% reported five or more. Table 2 lists the frequencies of the diseases and conditions assessed; more than half the population reported visual impairment, arthritis, and high blood pressure, and one-fifth reported heart disease (any type) and hearing impairment. Table 3 presents the most frequently co-occurring chronic condi-
L. P. Fried et al.
FIGURE 1. The distribution of chronic diseases and conditions in the study population of community-dwelling women 65 years and older, as indicated by the proportion of women reporting 0 or 1 to 11 chronic diseases. Women’s Health and Aging Study screenees.
tions, grouping all heart diseases within one category. The prevalence of these disease pairs was similar to that expected, under an assumption of independent disease reporting (see Methods). Fifty-six percent of these women reported difficulty in one or more tasks of daily life. Half (50%) reported difficulty in tasks that primarily required mobility and/or exercise tolerance; 35% reported difficulty in upper-extremity tasks, 22% reported difficulty with self-care tasks, and 22% with higher functioning (household) tasks. To enhance the specificity of identifiable associations, subsequent analyses used the seven mutually exclusive disability groups described in the Methods. Table 1 shows the numbers of people in each of these seven categories (unweighted data; weighted data similar, Table 4). Bivariate Associations of Each Disease with Disability Types To explore the association of individual diseases with disability, we first evaluated the frequency with which each of
1737 542 1273 333 205 259 569 124 171 125 63 221 695 259 288 543 256 425
843 2216 2869 313 367 526 512 1997 826
n 2.4*** 3.1*** 1.3** 3.5*** 3.1*** 1.9*** 1.7*** 1.3** 1.8***
2.1 1.7
6.2* 30.7* 6.4**
2.8*
4.3*
15.7 3.4**
1.9**
2.2
2.9**
3.0*** 1.3 0.9 0.5 2.2*** 1.6** 1.6** 1.1 1.4*
B
1.9*
2.1*** 2.0*** 1.2 1.8* 2.5** 1.6* 1.7** 1.5* 1.4
A
2.7**
2.3*** 1.8** 1.0 2.6** 4.2*** 1.9*** 1.2 1.0 2.0***
B
OR
OR A
Mobility only (n 5 2032)
Any difficulty (n 5 3791)
1.3 2.4*** 0.9 2.4 1.9 1.5 2.4*** 1.4 1.1
A
B
2.3
3.1
3.6 14.7*
3.3*
0.9 1.6* 1.0 0.5 0.9 1.4 4.8*** 0.8 1.0
OR
Upperextremity only (n 5 1679)
3.0*** 3.4** 1.2 2.5* 3.5*** 1.8* 1.4 1.6* 1.5*
A
B
3.0**
3.2 17.2** 5.1*
3.9*
3.7*
2.9*** 2.6** 1.0 0.9 3.1*** 2.1** 0.4 1.4 1.7**
OR
Mobility and upper extremity (n 5 1843)
2.7*** 3.0 1.5* 4.0*** 3.1** 1.8* 2.2* 1.1 2.7**
A
B
2.3
10.3** 56.4
3.5
1.7* 2.4
2.7
1.5 1.2 1.2 5.7*** 8.1*** 1.1 1.4 0.8 6.2***
OR
Higher function (not self-care) (n 5 1835)
2.1*** 5.1*** 1.4 2.3** 2.9*** 1.4 1.5* 1.0 1.5*
A
5.2
1.3 3.1** 1.1 0.5 2.4** 1.3 1.0 0.8 0.9
B
2.3**
5.1* 27.6**
6.6
15.0**
OR
Self-care (not higher function (n 5 1805)
Strength of association for difficulty in disability group
for main effects (each of the individual diseases/conditions listed on table) plus age, race, education, and MiniMental state examination. main effects were entered into every model. cMiniMental score. *P # 0.05 **P # 0.01 ***P # 0.0001 Abbreviations: OR 5 odds ratio.
bAll
aAdjusting
Main effectsb Heart disease Arthritis Visual impairment Stroke Lung disease Diabetes Cancer High blood pressure Hearing impairment Synergistic interactions* Arthritis*Visual impairment Arthritis*Hearing Arthritis*High blood pressure Arthritis*Diabetes Arthritis*Stroke Arthritis*Lung disease Heart disease*Arthritis Heart disease*Stroke Heart disease*Diabetes Heart disease*Cancer Lung disease*Cancer Stroke*High blood pressure Visual*Hearing impairment Visual impairment*Stroke Visual impairment*Lung disease Visual impairment*Cognitive impairmentc Cancer*High blood pressure High blood pressure*Hearing impairment
Health and Aging Study screenees
2.7** 4.8** 1.6** 8.8** 3.3*** 3.1*** 1.7** 1.4** 1.9***
A
B
2.3*
6.7 58.2** 10.4* 3.0 19.1
5.6* 7.5***
3.2*** 2.4*** 1.9** 3.9* 3.0*** 3.7*** 0.9 0.9 1.3
OR
Higher function and self-care (n 5 2197)
TABLE 2. Independent contributions of diseasesa to disability in main effects only (A columns) and main effects plus interaction models (B columns) in Women’s
Association of Comorbidity with Disability 31
L. P. Fried et al.
32
TABLE 3. Most frequently co-occurring chronic conditionsa among Women’s Health and
Aging Study screenees
Diseases present
Proportion of populationb with both diseases (%)
Arthritis, visual impairment Visual impairment, high blood pressure Arthritis, high blood pressure Any heart disease, visual impairment Visual impairment, hearing impairment Any heart disease, arthritis Any heart disease, high blood pressure Arthritis, hearing impairment Diabetes, visual impairment Cancer, visual impairment
44 40 34 17 15 14 13 12 12 10
Rank 1 2 3 4 5 6 7 8 9 10 aAmong
11 chronic conditions assessed (3 heart disease categories summed in 1 category). to reference population.
bWeighted
the disability types was reported by those with a disease, using estimated percentage distributions derived from polytomous models. For the nine diseases assessed, each row of Table 4 shows the prevalence of different types of disability among those with the disease specified. These proportions can be contrasted with the overall prevalence of each disability type (bottom row, Table 4) to assess whether people with a given disease reported higher or lower frequency of a given disability type than was found in the overall population. The prevalence of the most severe disability type, difficulty with both higher function and self-care, was increased among those with each disease compared with the overall population group, with more than a 5% increase associated with heart disease, diabetes, hearing impairment, lung disease, and stroke. The population with a history of stroke had a substantially higher frequency of disability in
higher function/self-care tasks than in other types of disability. Those with heart disease or lung disease, on the other hand, had more than 5% higher rates of difficulty in the mobility and upper-extremity disability group than did the overall population (8.5%). Visual impairment and high blood pressure were associated with relatively less overall disability than were other diseases. We examined the significance of association between each disease, singly, and disability types, using polytomous logistic regressions and adjusting for confounders. Every disease considered was significantly associated (P # 0.05) with the distribution of disability. In post hoc comparisons, every disease was significantly associated with each disability type (vis à vis no disability), with only two exceptions: visual impairment and hearing impairment were not associated with difficulty in mobility only, relative to “no difficulty.”
TABLE 4. Prevalencea of disability among individuals with different chronic diseases and conditions
Disability Prevalencea Given Presence of Disease
Disease present Heart diseaseb Arthritis Cancer Diabetes High blood pressure Hearing impairment Visual impairment Lung disease Stroke Overall prevalencea
No disability among those with disease (%) 23.7 31.7 33.7 26.4 37.8 26.7 40.4 20.6 15.3 43.8
(Cumulative Mobility Upperdisability among those Mobility extremity and upper extremity only only with disease) (%) (%) (%) (%) 76.3 68.3 66.3 73.6 62.2 73.3 59.6 79.4 84.7 56.2
15.4 14.8 16.0 14.7 15.8 13.9 14.4 17.3 10.9 14.2
aWeighted. bIncluding
myocardial infarction, angina, and congestive heart failure.
3.4 5.0 7.0 4.4 4.6 3.6 4.3 4.2 3.9 4.5
13.6 10.4 8.8 12.0 10.0 9.8 9.0 15.0 9.5 8.5
Higher function, not self-care (%) 11.0 8.8 10.2 10.4 7.7 12.2 8.1 9.3 11.8 7.4
Higher Self-care, function not higher function and self-care (%) (%) 9.1 10.3 8.0 6.4 7.0 8.7 7.7 11.3 7.2 7.3
23.7 19.1 16.3 25.8 17.1 25.1 16.2 22.4 41.3 14.5
Association of Comorbidity with Disability
All diseases were also significantly associated with having “any task difficulty” (data not shown).
Multivariate Analyses: Independent Associations of Each Disease with Disability Types We then evaluated the independent association of each of the nine diseases and conditions with disability, adjusting for the other diseases as well as confounders in a main effects model (results are shown in Table 2, A columns). Almost all of the nine diseases were associated (P , 0.05) with all of the disability outcomes. The exception was with upper-extremity difficulty only, for which only two of nine of the diseases were associated; here, power to detect associations was somewhat weaker than for other comparisons, so that the magnitude of the nonsignificant odds ratios for heart disease, diabetes, and high blood pressure exceeded some that were significant in other disability groups. Overall, in these main effects models, the presence of significant associations did not provide more specific information on associations of disease with disability than was found for the “any difficulty” group. Some sense of specificity can be derived, however, from the relative strengths of associations. For example, vision and hearing impairments were most strongly associated with difficulty in the higher-function task group; self-care difficulty (without higher function difficulty) was twice as strongly associated with arthritis as with any other disease. Generally, the findings in Tables 2 and 4 were consistent. The major differences were that the association of arthritis with many disability types appears stronger after adjustment (in Table 2, A columns) than in the unadjusted prevalences (Table 4).
Associations of Disease Pairs with Disability Types Finally, we evaluated the associations of pairs of diseases, as well as the individual diseases, with disability, using multiple logistic regression models adjusting for age, race, education and MiniMental State exam score (see Methods). Table 2, B columns, shows the odds ratios from these models for (1) the main effects, representing independent effects of single diseases, and (2) the pairs of diseases in which the odds of disability was synergistically increased in the presence of both diseases, compared with neither disease being present and adjusting for the effect of each disease alone. Several observations emerge from these analyses. First, selected disease pairs were significantly and synergistically associated with disability. Given that 36 disease pairs were evaluated in these exploratory analyses, approximately two significant interactions (synergistic and otherwise, at the 0.05 level) are to be expected by chance alone per model. Although the number of synergistic interactions alone exceeded this for the more-severe disability types, interpreta-
33
tion should consider scientific coherence of results and strengths of associations. An overwhelming tendency to synergy was not observed. There were, however, important disease pairs that recurred in their associations with different disability types. These included the joint presence of arthritis and visual impairment, arthritis and high blood pressure, heart disease and cancer, lung disease and cancer, and stroke and high blood pressure. Second, the type of disability with which a disease was associated varied, depending on the other disease that was present. For example, vision impairment in combination with arthritis was associated with mobility task difficulty, but paired with stroke they were jointly associated with difficulty in both higher-function and self-care tasks (Table 2, B columns). Third, when interactions were accounted for, many diseases no longer appeared, in themselves, to be independently associated with a given type of disability. Most notably, with interactions in the models, neither high blood pressure nor visual impairment remained associated with any disability type, by themselves, except in one instance. For almost half of the disease pairs involved in interactions, each single disease was no longer significantly associated, by itself, with that disability subtype (Table 2, B columns). Overall, the association of disease pairs with disability groups showed more specificity than did the main effects alone. Some disease interactions showed specificity of association with only a few disability groups; others (e.g., heart*cancer) showed consistent associations with a number of disability types. Figure 2 summarizes these findings graphically for five disease pairs, compared with the individual diseases. The triangles in each panel indicate the odds ratios for the association of the pair of diseases when there was a synergistic interaction (P # 0.10), and the square and diamond symbols indicate the odds ratios for the association of each disease, singly, with each disability type (odds ratios are drawn from data in Table 2, B columns). Figure 2A shows that the joint presence of both heart disease and cancer was more strongly associated with difficulty in five disability types (upper extremity, both mobility and upper extremity, higher function tasks without self-care difficulty, and selfcare difficulty without higher-function task difficulty, and higher function and self-care) than was expected by multiplying the single-disease associations; the joint presence of the two diseases was not synergistically associated with mobility difficulty alone. In contrast, Figure 2B through 2E shows disease pairs in which the synergistic association was only important in relation to one or two disability types. Arthritis and heart disease (2B) were jointly associated with difficulty in higher-function or self-care tasks. Arthritis and stroke (2C) were jointly associated with self-care task difficulty (not higher-function task difficulty). Arthritis and hearing (Figure 2D) and vision and stroke (2E), as pairs, were each synergistically associated with difficulty in both higher function and self-care tasks.
34
L. P. Fried et al.
FIGURE 2. Odds ratios for the association of important synergistic interactions between disease pairs and for the independent associations of each of the two diseases, with disability types. For each disability type, “no disability” serves as the reference comparison group. For disease types, reference is “neither disease.”
Association of Comorbidity with Disability
Associations of Individual Diseases and Confounders in Models with Interaction Terms In terms of main effects, report of any task difficulty showed disease associations that appeared most consistent with the findings in the group with mobility only or both mobility and upper-extremity difficulty (Table 2, B columns). The main effects for other disability outcomes showed substantially different patterns of association than for “any difficulty.” In addition to self-reported disease, cognitive function, as measured by the MiniMental State Exam score, was associated (P , 0.05) with “any difficulty” and with difficulty in the two combination groups (both mobility and upperextremity difficulty, and difficulty in both higher-function and self-care tasks), but with none of the other disability outcome measures. Age greater than 85 years was associated with all disability outcomes, and education was associated only with “any difficulty” and with the two groups with difficulty in the higher-function tasks (P , 0.05). Race was not independently associated with any disability outcome measure, except for those with difficulty in higher function but not self-care (data not shown).
DISCUSSION The goals of this study were to describe the prevalence and co-occurrence of multiple chronic diseases in a representative population of older women and to explore how pairs of comorbid diseases were related to self-reported disability and whether disease–disability relationships differed depending on the type of disability considered. Paralleling previous studies, we found that a number of chronic diseases were highly prevalent and that comorbidity was common [3,4,7,15], that disease and disability were highly associated [1–12], and that disease was more highly explanatory of disability than were age, race, or educational status [2]. Beyond these observations, this study performed exploratory analyses that indicated that specific pairs of chronic diseases were synergistically associated with different types of disability. Certain of these interactions appear importantly related to disability; that is, the associations were large and well beyond the multiplicative effects of the two diseases independently. It appears from the results of these multivariate analyses that a substantial amount of the contribution of disease to disability may derive from the interactions of specific comorbid diseases. If this is proved to be the case in future studies, then a new, potentially effective strategy for prevention or amelioration of disability would be to decrease targeted disease – disease interactions. Thus, clinicians would target prevention of a comorbid disease that could cause disability given the one already present or would target treatment to decrease severity of one or both comorbid diseases already present in order to eliminate the interaction.
35
Evaluation of disease interactions potentially sheds light on etiologic relationships in a way that looking only at independent associations does not. This is apparent when comparing the associations of individual diseases with disability types when interactions were (Table 2, B columns) or were not (Table 2, A columns) accounted for. In the main effects only models (Table 2, A columns), almost every disease was associated with almost every disability type. When interactions were considered, however, there was much more specificity apparent in the relationships. For example, in Table 2 (B columns), arthritis was interactively associated with mobility in the presence of visual impairment or high blood pressure but showed no independent association itself with mobility, whereas it was independently associated with upper-extremity difficulty. Given the high prevalence of arthritis, it is unlikely that the lack of association between arthritis alone and difficulty in mobility is a result of inadequate power. Thus, comorbidity may substantially heighten the risk of mobility difficulty associated with arthritis, as has been indicated by other studies [3,8,12]. Thus, when there is a significant synergistic interaction, there are at least two qualitative interaction effects that can be operating that could be both biologically plausible and important: first, that having both diseases is associated with substantially higher risk of disability than is expected by combining the individual disease effects; and, second, that one of the diseases is associated with disability only in the presence of the other disease. These effects have been previously reported by Verbrugge et al. [3] when looking at disability broadly as an outcome. Instances of both can be found in this study: The first scenario is exemplified by the interaction of visual impairment and stroke in association with higher-function and self-care difficulty; the second scenario is exemplified by heart disease and cancer in association with upper-extremity difficulty. The predominant pattern in the analyses evaluating interactions was of two diseases in combination being strongly associated with disability, while neither or only one was independently associated. Additionally, there were two types of disease combinations associated with excess disability risk—those that confer a great deal of excess risk and those that are prevalent and confer at least moderate excess risk. Inferentially, it is likely that those with more severe disability (i.e., IADL and/or ADL difficulty) have more severe underlying disease; perhaps a greater severity of a single disease is required for that disease to cause disability by itself. Conversely, in the presence of less-severe disease, interaction with other diseases may be necessary to produce disability. This could explain why some diseases were not associated with what is thought to be less-severe disability in themselves, but they were associated in specific disease pairs. In these data we were not able to assess disease severity to test this question. The role of disease severity will require further evaluation. The joint effect of two diseases equals that which would
L. P. Fried et al.
36
be expected by directly multiplying the individual effects, a multiplicative interaction. The absence of a synergistic interaction does not rule out an increased functional impact from the presence of two diseases, compared to one. The lack of an interaction could also result from inadequate power to rule out interactions. Although prior studies have looked at synergy between two diseases in causing disability in general [3] or in relation to mobility difficulty [3,7,8,12,17], this study is the first to show that there may be specificity in the effect of disease pairs on different types of disability. In fact, the patterns of interaction varied in their associations by disability type. These results also indicate that partitioning disability into mutually exclusive subtypes is potentially more informative than looking only at the presence of “any difficulty.” The specificity of associations that is apparent by comparing diseases associated with mobility difficulty with those associated with other disability types would not be revealed by a simple summary measure of disability. Such specificity could, potentially, guide clinical evaluation of the underlying causes of disability. That is, given a patient with difficulty in a self-care task but not in higher-functioning tasks, the clinician might be able to target identification of the individual diseases potentially associated with this disability and the combinations of diseases that confer high risk. We have characterized the main effects of the models with interaction terms (Table 2, B columns) as roughly reflecting effects of diseases in isolation. This is a substantial approximation—both because some triplets and higherorder combinations of diseases likely interact in even more complicated ways, and because some of the diseases considered rarely appear in isolation. While in-depth examination of these complexities is beyond the scope of these exploratory analyses, we did assess whether the approximation was reasonable using analyses that restricted to two diseases at a time (say, heart disease and cancer), as follows: We computed the probabilities of being in each disability category for each of the following: neither disease, heart disease only, cancer only, or heart disease and cancer. Then, we computed the odds ratios for each disease group relative to no disease, comparing odds of being in a specific disabled category to odds of being not disabled; in doing this, to approximate adjustment for other diseases that is made within our multivariate models, we stratified analysis by the number of other diseases reported—two or fewer other diseases versus two or more diseases. In fact, these simple analyses did agree quite well with our Figure 2. However, they also pointed out that: (a) there was evidence of higher-way interactions, so that interactions between disease pairs appeared amplified in women with a larger versus a smaller “other” disease burden; (b) for the rarer diseases that do not often occur in isolation (stroke, for example), the main effects may be underestimated and the interactions correspondingly overestimated; and (c) for the more common
diseases, the multivariate model information draws more heavily from persons with a lesser disease burden, and the opposite is true for the rarer diseases. In-depth examination of these complexities is beyond the scope of these exploratory analyses, but they should be kept in mind in designing confirmatory studies on the effects of comorbidity. The findings reported here provide a basis for hypothesis development and to guide additional research and testing of individual synergistic relationships in a more in-depth manner. Given that these data are cross-sectional, prospective studies are also needed to determine the causal nature of these relationships. If hypothesis testing, especially in prospective studies, confirms these observations, this would suggest several directions for both prevention and treatment of disability. Clinicians evaluating potential etiologies of disability may look to the pairs of diseases that should be treated if a particular disability is present, as well as an individual cause, with the goal of minimizing the severity of at least one disease so as to diminish the interaction. Similarly, for disability prevention, individuals with one high-risk chronic disease present should be carefully screened for prevention of other diseases that could interact to cause disability. Thus, confirmation of the results of this study could add a new perspective to the import of disease prevention, placing emphasis on the prevention of comorbidity as a goal in itself. Meaningful decreases in disability could potentially result from interventions to decrease the synergy between diseases. Supported by the National Institue on Aging (Contract No. NO1AG12112 and Grant # AG11703-05). Dr. Fried was a Henry J. Kaiser Family Foundation Faculty Scholar in General Internal Medicine during the conduct of this study. Dr. Bandeen-Roche is a Brookdale National Fellow in Aging Research. The authors thank Mr. Raymond Burchfield for his excellent preparation of this manuscript and Ms. Carol Han for preparation of the figures.
References 1. Fried LP, Guralnik JM. Disability in older adults: Evidence regarding significance, etiology, and risk. J Am Geriatr Soc 1997; 45: 92–100. 2. Fried LP, Ettinger WH, Lind B, Newman AB, Gardin J. Physical disability in older adults: A physiological approach. J Clin Epidemiol 1994; 47: 747–760. 3. Verbrugge LM, Lepkowski JM, Imanaka Y. Comorbidity and its impact on disability. Milbank Q 1989; 67: 450–484. 4. Guralnik JM, LaCroix AZ, Everett DF, Kovar MG. Aging in the Eighties: The Prevalence of Comorbidity and Its Association with Disability. Advance Data from Vital and Health Statistics; No. 170, Hyattsville, MD: National Center for Health Statistics; 1989. 5. Mor V, Murphy J, Masterson-Allen S, Willey C, Razmpour A, Jackson ME, et al. Risk of functional decline among well elders. J Clin Epidemiol 1989; 42: 895–904. 6. Fried LP. Older women: Health status, knowledge, and behavior. In: Falik M, Scott Collins K, Eds. Women’s Health: Commonwealth Fund Survey. Baltimore, MD: Johns Hopkins University Press; 1996, 175–204.
Association of Comorbidity with Disability
7. Guralnik JM, LaCroix AZ, Abbott RD, Berkman LF, Satterfield S, Evans DA, et al. Maintaining mobility in late life: Demographic characteristics and chronic conditions. Am J Epidemiol 1993; 137: 845–857. 8. Ettinger WH, Davis MA, Neuhaus JM, Mallon KP. Long-term physical functioning in persons with knee osteoarthritis from NHANES-I: Effects of comorbid medical conditions. J Clin Epidemiol 1994; 47: 809–815. 9. Campbell AJ, Busby WJ, Robertson MC, Lum CL, Langlois JA, Morgan FC. Disease, impairment, disability and social handicap: A community based study of people aged 70 years and over. Disabil Rehabil 1994; 16: 72–79. 10. Almy TP. Comprehensive functional assessment for elderly patients. Ann Intern Med 1988; 109: 70–72. 11. Guccione AA, Felson DT, Anderson JJ. Defining arthritis and measuring functional status in elders: Methodological issues in the study of disease and physical disability. Am J Public Health 1990; 80: 945–949. 12. Verbrugge LM, Lepkowski JM, Konkol LL. Levels of disability among U.S. adults with arthritis. J Geriatr Soc Sci 1991; 46: S71–83.
37
13. Fried LP, Kasper JD, Guralnik JM, Simonsick EM. The Women’s Health and Aging Study: An Introduction. In: Guralnik JM, Fried LP, Simonsick EM, Kasper JD, Lafferty ME, Eds. The Women’s Health and Aging Study: Health and Social Characteristics of Older Women with Disability. NIH Pub. No. 95-4009; Bethesda, MD: National Institutes of Health, National Institute on Aging; 1995: 1–8. 14. Folstein MF, Folstein SE, McHugh PR. Mini-mental state: A practical method for grading the cognitive state of patients for the clinician. J Psychiatr Res 1975; 12: 189–198. 15. U.S. Senate Special Committee on Aging, the American Association of Retired Persons, the Federal Council on the Aging, and the U.S. Administration on Aging. Aging America, Trends and Projections. 1991 Edition; 107–142. 16. LaPlante MP. Data on disability from the National Health Interview Survey 1983–85. Washington, DC: National Institute on Disability and Rehabilitation Research, U.S. Department of Education; 1988. 17. Verbrugge LM. Disability transitions for older persons with arthritis. J Aging Health 1992; 4: 212–243.
0895-4356/99/ $–see front matter PII S0895-4356(98)00124-3
Association of Comorbidity with Disability in Older Women: The Women’s Health and Aging Study Linda P. Fried,1,* Karen Bandeen-Roche,1 Judith D. Kasper,1 and Jack M. Guralnik2 for the Women’s Health and Aging Study Collaborative Research Group 1The
Johns Hopkins Medical Institutions, Baltimore, Maryland; and 2Epidemiology, Demography and Biometry Program, National Institute on Aging, Bethesda, Maryland ABSTRACT. There is substantial evidence that physical disability results from chronic diseases and that the number of chronic diseases is associated with the presence and severity of disability. There is some evidence that interactions between specific diseases are of import in causing disability. Beyond arthritis, however, little is known of the disease pairs that may be important to focus on in future research. This study explores the associations between multiple disease pairs and different types of physical disability, with the objective of hypothesis development regarding the importance of disease interactions. The study population comprised a representative sample of 3841 women 65 years and older living in Baltimore, screened for participation in the Women’s Health and Aging Study. The study design was cross-sectional. An interviewer-administered screening questionnaire was administered regarding self-reported physical disability in 15 tasks of daily life, history of physician diagnosis of 14 chronic diseases, and MiniMental State examination. Task difficulty was empirically grouped into six subsets of minimally overlapping disabilities, with a comparison group consisting of those with no difficulty in any task subset. Multiple logistic regression models were fit assessing the relationship of major chronic diseases and of interactions of disease pairs with each disability subtype and with any disability, adjusting for confounders. Fourteen percent of the population reported mobility difficulty only; 5%, upper extremity difficulty only; 9%, both of these difficulties but no others; 7%, difficulty in higher function but not self-care tasks; 7%, self-care task difficulty but not higher function tasks; and 15%, difficulty in both higher function and self-care (weighted data). Almost all in the latter three groups had difficulty, as well, in mobility or upper extremity tasks. In regression models, specific disease pairs were synergistically associated with different types of disability. For example, important disease pairs that recurred in their associations with different disability types were the presence of arthritis and visual impairments, arthritis and high blood pressure, heart disease and cancer, lung disease and cancer, and stroke and high blood pressure. In addition, the type of disability that a disease was associated with varied, depending on the other disease that was present. Finally, when interactions were accounted for, many diseases were no longer, in themselves, independently associated with a given type of disability. Partitioning disability into six subtypes was more informative in terms of associations than was evaluating a summary category of “any disability.” These findings provide a basis for further hypothesis development and testing of synergistic relationships of specific diseases with disabilities. If testing confirms these observations, these findings could provide a basis for new strategies for prevention of disability by minimizing comorbid interactions. j clin epidemiol 52;1:27–37, 1999. © 1999 Elsevier Science Inc. KEY WORDS. Aging, comorbidity, disability, chronic disease, physical function, geriatrics
INTRODUCTION Physical disability results primarily from chronic disease [1– 3] and is highly prevalent in older adults. The presence of multiple diseases increases the likelihood of developing physical disability. As the numbers of diseases increase, *Address correspondence to: Linda P. Fried, M.D., M.P.H., Professor, Medicine and Epidemiology, Welch Center for Prevention, Epidemiology and Clinical Research, Suite 2-600, 2024 E. Monument Street, Baltimore, Maryland 21205. Accepted for publication on 22 July 1998.
there is increasing risk of difficulty with activities of daily living (ADLs) [3–5], instrumental ADLs (IADLs) [3,5,6] and mobility [3,7–10]. There is related evidence that specific diseases can cause specific types of disability [1,2,8,11], and some studies suggest that interactions among specific diseases may be of importance [3,8,12]. This has been explored for arthritis, with results indicating that when arthritis occurs in the presence of other specific diseases, there is substantially greater risk of mobility and transfer difficulty than with arthritis alone [12] and that arthritis and other diseases act synergistically on some functions [8]. Little has
L. P. Fried et al.
28
been reported in this regard, however, for diseases other than arthritis. It is also not known, for diseases other than arthritis, whether specific combinations of diseases cause different types of disability. If so, interventions to prevent or minimize disability would need to take into account both the specific comorbidities and the type of disability in order to be optimally effective. The Women’s Health and Aging Study (WHAS) has recently completed screening of a representative sample of community-dwelling women 65 and older, providing information on major chronic diseases, cognitive function, and disability. This article evaluates these data to provide a broad overview of the relationships of individual diseases and pairs of diseases to different types of disability in a representative population of women 65 years and older. Specifically, we sought to explore how combinations of diseases may be related to different types of disability, over and above the independent contributions of each disease to disability. Such information can provide insight into the clinical profiles of potentially important disease pairs and into how these differ from the association between a single disease and disability, and can also suggest avenues for future hypothesis-driven research that can specifically delineate opportunities for intervention.
METHODS Study Population The study population consists of a representative sample of women 65 years and older who were screened to determine eligibility for the WHAS. The WHAS is a study of the causes and course of disability in moderately to severely disabled older women living in the community. These analyses, however, use data on all women who were screened and are not limited to the subgroup selected for the full study. The sampling frame was obtained from the Health Care Financing Administration Medicare Enrollment files for the eastern half of Baltimore City and County, consisting of 12 zip code areas. A random sample of older women was drawn, stratified by age (65–74, 75–84 and 85 and older). Those sampled were contacted first by letter and then received a screening interview in their homes. Eighty-one percent of the 5316 women sampled consented to be screened, and the study population considered for analysis here consists of 3841 women who completed the screening interview. Study design and sampling and data collection have been previously described in detail [13]. Data Collected The interviewer-administered, standardized screening questionnaire ascertained, through self-report, the participants’ demographic characteristics and whether a physician had diagnosed any of the following 14 chronic diseases and conditions: myocardial infarction, angina, congestive heart
failure, other heart disease, high blood pressure, diabetes, arthritis, stroke, cancer, lung disease, vision problems, hearing problems, Parkinson’s disease, or a hip fracture. A MiniMental State examination was administered in standardized fashion to assess cognitive function [14]. Self-reported disability was ascertained by asking about difficulty in 15 tasks of daily life, which were selected to describe four defined domains of physical functioning [2]: mobility and exercise tolerance-demanding tasks (getting in or out of a bed or chair, climbing up 10 steps without resting, walking two to three blocks, doing heavy housework); upper-extremity tasks (raising arms overhead, grasping or handling, lifting and carrying 10 pounds); higher-function tasks essential to independent living (using a telephone, light housework, meal preparation, shopping); and basic self-care tasks (bathing/showering, dressing, eating, toileting). Data Analysis To achieve our analytic goal of describing the relationships of comorbid diseases (defined as two diseases jointly present in the same individual) with disability in our population, we proceeded in stages. First, we examined the prevalence of disease in our population and whether a priori grouping of related comorbid diseases would be useful. Second, we sought to characterize major distinct patterns of reported disability in our population. Third, we examined the bivariate associations of each disease with disability, overall and by subtype. Fourth, we calculated, in multivariate analyses, the independent associations of the diseases with disability, and we compared these with the separate associations calculated in the previous step. Fifth and centrally, we sought to describe relationships between comorbidity and disability by determining which pairwise combinations of diseases substantially modified the independent effects of the individual diseases on disability and how these mutual effects varied in association with different types of disability. Finally, statistical diagnosis was undertaken to ensure that our multivariate analyses adequately described the data. All prevalence analyses were performed both unweighted, to identify instances of sparse data, and weighted, to represent prevalence appropriately in the population from which the sample was drawn. In any specific table, cell, or calculation, from 4 to 27 women were excluded owing to missing information. DISEASE DESCRIPTION AND SUMMARY. Prevalences of disease were determined and expected coprevalences of diseases were calculated as (prevalence of disease A) 3 (prevalence of disease B). Associations among each pair of chronic conditions were assessed; the strongest observed associations were those between pairs of cardiac diseases, that is, myocardial infarction, angina, and congestive heart failure (odds ratios 5 5.6–8.0). To diminish collinearity, these three diseases were combined into one binary heart disease
Association of Comorbidity with Disability
29
variable for use in regression models. All other associations were moderate or weak, indicating limited utility of further summary. This relative lack of association indicated that interactions would be an appropriate way to model the effects of diseases in combination, as opposed to further summary. In subsequent analyses, we characterized comorbidity as the presence of disease pairs. Parkinson’s disease and hip fracture were excluded from subsequent analyses because of their very low prevalence. SUMMARY OF PHYSICAL FUNCTION. In previous factor analyses [2], it was found that tasks of daily life aggregated into four groups in which difficulty in one task was associated with difficulty in the others, suggesting a physiologic grouping. These overall groups involved difficulty with (1) mobility and exercise tolerance-demanding tasks, (2) upperextremity tasks, (3) higher-function tasks (a subset of IADLs), and (4) basic self-care tasks (a subset of ADLs). Use of these groupings to summarize function improved specificity in evaluating associations between disease and disability. These domains were therefore used as the initial basis for these analyses. Data exploration in this study showed substantial overlap between domains, however, with individuals frequently reporting difficulty in more than one of these four groups. As a result, we investigated further the frequency of overlaps for each of 16 possible combinations of difficulty. This process yielded six distinct groupings of disability: mobility difficulty only; upper-extremity difficulty only; both of these but no other difficulty; difficulty in higher function tasks, but not self-care tasks; difficulty in self-care tasks but not higher function tasks; and difficulty in both higher function and self-care tasks (Table 1). Among those in the latter three groups, 95% reported difficulty in mobility and/or upper-extremity function as well. The least over-
lap was seen between those with mobility difficulty and those with upper-extremity difficulty. To maximize specificity, the study population was stratified into the six disability groups just outlined and a seventh group that reported no difficulty in any of these tasks. ASSOCIATIONS OF EACH DISEASE WITH DISABILITY. Polytomous logistic regression was used to determine the associations between disability as the dependent variable (seven categories) and each of nine diseases separately: arthritis, any heart disease (including report of angina, myocardial infarction or congestive heart failure), stroke, high blood pressure, diabetes, lung disease, cancer, hearing impairment, and visual impairment. Each of these models also included age (65–74, 75–84, 85 and older), race (white or AfricanAmerican), educational level (less than high school graduate versus high school or greater), and MiniMental State Examination score (,24 versus 24 or greater) as potential confounders. To describe these associations, we estimated and contrasted prevalences of each disability type among women reporting each disease. To obtain these prevalences, we weighted and averaged over confounders in each polytomous model, then back-calculated the resulting log odds parameters to prevalences. This approach provides a broad overview of associations between individual diseases and disability.
INDEPENDENT AND JOINT ASSOCIATIONS OF DISEASE WITH DISABILITY.
We used multiple logistic regression analyses to describe how the nine diseases were independently associated with, and how disease co-occurrences were related to, selfreported disability. To facilitate model checking, we fit binary logistic regression models for each of the six disability outcomes; persons with no reported difficulty served as the
TABLE 1. Prevalencea of different types of physical disability Women’s Health and Aging Study
screenees (n 5 3841)
Domains of disability—difficulty in: Mobility/exercise tolerance-demanding tasks
Upper-extremity tasks No
No Yes No Yes Yes aUnweighted.
Higher-function tasksb Self-care tasks No Yes
No 40.0%
(1536)7
0.7% (25)5
No
4.2%
Yes
0.5% (20)5
No Yes
13.5%
(159)2 (518)1
2.2% (85)5
Yes 0.9% (33)4 0.0% (1)6 0.1% (3)4 0.1% (5)6 3.0% (114)4 2.3% (87)6
No
8.5%
(326)3
4.5% (172)4
Yes
4.1% (158)5
15.5% (595)6
b Groupings of disability: 1 5 mobility difficulty only; 2 5 upper-extremity difficulty only; 3 5 both mobility and upper extremity difficulty; 4 5 higher function task difficulty but no self-care difficulty (may have difficulty in mobility or upper extremity tasks); 5 5 Self-care task difficulty but no higher function difficulty (may have difficulty in mobility or upper extremity tasks); 6 5 Both higher-function and self-care task difficulty (may have difficulty in mobility or upper extremity tasks); 7 5 No difficulty with any task assessed.
30
reference group in each model. For comparison, and to understand the relationships with comorbidity that would be apparent if using the most nonspecific characterization of disability, models using the outcome of any self-reported difficulty in one or more of the 15 tasks of daily life (versus no difficulty) were also fit (“Any difficulty” model). Logistic regressions were based on 3800 women with nonmissing information for dependent and independent variables. Models were first analyzed for the nine disease main effects: heart disease, arthritis, stroke, lung disease, diabetes, high blood pressure, cancer, hearing impairment and visual impairment; these were adjusted for age, race, education and MiniMental Score. Then, two-way interactions were used to model the effects of pairwise disease co-occurrence, using multiple logistic regression models. Here, our goal was to explore which interactions were particularly predictive of disability. Potential two-way interactions were added to the main effects model by a stepwise process, using entry and retention alpha levels of 0.10. Odds ratios were calculated from these regression models. In reporting, we focus on the strong, positive interaction terms. These indicate synergistic associations in which having both diseases was associated with substantially higher risk of disability than was expected by multiplying disease-specific risks. To compare disability risk for different, significantly interacting disease pairs, plots were constructed. These illustrate the odds ratios for each disability type (versus no disability) for the joint presence of two diseases and for each disease separately, compared with the reference population reporting no disease(s). Plotted odds ratios were derived from our final, multivariate fitted models. RESULTS Description of Disease and Disability in Population Among the study population of 3841 community-dwelling women 65 to 101 years of age, 25% were African-American and 75% were white. Thirty-one percent were married, and 35% reported 8 or fewer years of education, and 19 percent, more than a high school education. The mean MiniMental score was 26, with 11% and 4%, respectively, having scores of 18 to 23 and less than 18. Self-assessed health was reported as excellent, good, or fair/poor by 37%, 34% and 29%, respectively. There was a high frequency of chronic disease, with an average of three diseases per woman and a range of 0 to 11. Figure 1 displays the frequency of co-occurring diseases in the population. Only 5% reported having none of the 14 conditions assessed, whereas 81% reported two or more chronic conditions, and 18% reported five or more. Table 2 lists the frequencies of the diseases and conditions assessed; more than half the population reported visual impairment, arthritis, and high blood pressure, and one-fifth reported heart disease (any type) and hearing impairment. Table 3 presents the most frequently co-occurring chronic condi-
L. P. Fried et al.
FIGURE 1. The distribution of chronic diseases and conditions in the study population of community-dwelling women 65 years and older, as indicated by the proportion of women reporting 0 or 1 to 11 chronic diseases. Women’s Health and Aging Study screenees.
tions, grouping all heart diseases within one category. The prevalence of these disease pairs was similar to that expected, under an assumption of independent disease reporting (see Methods). Fifty-six percent of these women reported difficulty in one or more tasks of daily life. Half (50%) reported difficulty in tasks that primarily required mobility and/or exercise tolerance; 35% reported difficulty in upper-extremity tasks, 22% reported difficulty with self-care tasks, and 22% with higher functioning (household) tasks. To enhance the specificity of identifiable associations, subsequent analyses used the seven mutually exclusive disability groups described in the Methods. Table 1 shows the numbers of people in each of these seven categories (unweighted data; weighted data similar, Table 4). Bivariate Associations of Each Disease with Disability Types To explore the association of individual diseases with disability, we first evaluated the frequency with which each of
1737 542 1273 333 205 259 569 124 171 125 63 221 695 259 288 543 256 425
843 2216 2869 313 367 526 512 1997 826
n 2.4*** 3.1*** 1.3** 3.5*** 3.1*** 1.9*** 1.7*** 1.3** 1.8***
2.1 1.7
6.2* 30.7* 6.4**
2.8*
4.3*
15.7 3.4**
1.9**
2.2
2.9**
3.0*** 1.3 0.9 0.5 2.2*** 1.6** 1.6** 1.1 1.4*
B
1.9*
2.1*** 2.0*** 1.2 1.8* 2.5** 1.6* 1.7** 1.5* 1.4
A
2.7**
2.3*** 1.8** 1.0 2.6** 4.2*** 1.9*** 1.2 1.0 2.0***
B
OR
OR A
Mobility only (n 5 2032)
Any difficulty (n 5 3791)
1.3 2.4*** 0.9 2.4 1.9 1.5 2.4*** 1.4 1.1
A
B
2.3
3.1
3.6 14.7*
3.3*
0.9 1.6* 1.0 0.5 0.9 1.4 4.8*** 0.8 1.0
OR
Upperextremity only (n 5 1679)
3.0*** 3.4** 1.2 2.5* 3.5*** 1.8* 1.4 1.6* 1.5*
A
B
3.0**
3.2 17.2** 5.1*
3.9*
3.7*
2.9*** 2.6** 1.0 0.9 3.1*** 2.1** 0.4 1.4 1.7**
OR
Mobility and upper extremity (n 5 1843)
2.7*** 3.0 1.5* 4.0*** 3.1** 1.8* 2.2* 1.1 2.7**
A
B
2.3
10.3** 56.4
3.5
1.7* 2.4
2.7
1.5 1.2 1.2 5.7*** 8.1*** 1.1 1.4 0.8 6.2***
OR
Higher function (not self-care) (n 5 1835)
2.1*** 5.1*** 1.4 2.3** 2.9*** 1.4 1.5* 1.0 1.5*
A
5.2
1.3 3.1** 1.1 0.5 2.4** 1.3 1.0 0.8 0.9
B
2.3**
5.1* 27.6**
6.6
15.0**
OR
Self-care (not higher function (n 5 1805)
Strength of association for difficulty in disability group
for main effects (each of the individual diseases/conditions listed on table) plus age, race, education, and MiniMental state examination. main effects were entered into every model. cMiniMental score. *P # 0.05 **P # 0.01 ***P # 0.0001 Abbreviations: OR 5 odds ratio.
bAll
aAdjusting
Main effectsb Heart disease Arthritis Visual impairment Stroke Lung disease Diabetes Cancer High blood pressure Hearing impairment Synergistic interactions* Arthritis*Visual impairment Arthritis*Hearing Arthritis*High blood pressure Arthritis*Diabetes Arthritis*Stroke Arthritis*Lung disease Heart disease*Arthritis Heart disease*Stroke Heart disease*Diabetes Heart disease*Cancer Lung disease*Cancer Stroke*High blood pressure Visual*Hearing impairment Visual impairment*Stroke Visual impairment*Lung disease Visual impairment*Cognitive impairmentc Cancer*High blood pressure High blood pressure*Hearing impairment
Health and Aging Study screenees
2.7** 4.8** 1.6** 8.8** 3.3*** 3.1*** 1.7** 1.4** 1.9***
A
B
2.3*
6.7 58.2** 10.4* 3.0 19.1
5.6* 7.5***
3.2*** 2.4*** 1.9** 3.9* 3.0*** 3.7*** 0.9 0.9 1.3
OR
Higher function and self-care (n 5 2197)
TABLE 2. Independent contributions of diseasesa to disability in main effects only (A columns) and main effects plus interaction models (B columns) in Women’s
Association of Comorbidity with Disability 31
L. P. Fried et al.
32
TABLE 3. Most frequently co-occurring chronic conditionsa among Women’s Health and
Aging Study screenees
Diseases present
Proportion of populationb with both diseases (%)
Arthritis, visual impairment Visual impairment, high blood pressure Arthritis, high blood pressure Any heart disease, visual impairment Visual impairment, hearing impairment Any heart disease, arthritis Any heart disease, high blood pressure Arthritis, hearing impairment Diabetes, visual impairment Cancer, visual impairment
44 40 34 17 15 14 13 12 12 10
Rank 1 2 3 4 5 6 7 8 9 10 aAmong
11 chronic conditions assessed (3 heart disease categories summed in 1 category). to reference population.
bWeighted
the disability types was reported by those with a disease, using estimated percentage distributions derived from polytomous models. For the nine diseases assessed, each row of Table 4 shows the prevalence of different types of disability among those with the disease specified. These proportions can be contrasted with the overall prevalence of each disability type (bottom row, Table 4) to assess whether people with a given disease reported higher or lower frequency of a given disability type than was found in the overall population. The prevalence of the most severe disability type, difficulty with both higher function and self-care, was increased among those with each disease compared with the overall population group, with more than a 5% increase associated with heart disease, diabetes, hearing impairment, lung disease, and stroke. The population with a history of stroke had a substantially higher frequency of disability in
higher function/self-care tasks than in other types of disability. Those with heart disease or lung disease, on the other hand, had more than 5% higher rates of difficulty in the mobility and upper-extremity disability group than did the overall population (8.5%). Visual impairment and high blood pressure were associated with relatively less overall disability than were other diseases. We examined the significance of association between each disease, singly, and disability types, using polytomous logistic regressions and adjusting for confounders. Every disease considered was significantly associated (P # 0.05) with the distribution of disability. In post hoc comparisons, every disease was significantly associated with each disability type (vis à vis no disability), with only two exceptions: visual impairment and hearing impairment were not associated with difficulty in mobility only, relative to “no difficulty.”
TABLE 4. Prevalencea of disability among individuals with different chronic diseases and conditions
Disability Prevalencea Given Presence of Disease
Disease present Heart diseaseb Arthritis Cancer Diabetes High blood pressure Hearing impairment Visual impairment Lung disease Stroke Overall prevalencea
No disability among those with disease (%) 23.7 31.7 33.7 26.4 37.8 26.7 40.4 20.6 15.3 43.8
(Cumulative Mobility Upperdisability among those Mobility extremity and upper extremity only only with disease) (%) (%) (%) (%) 76.3 68.3 66.3 73.6 62.2 73.3 59.6 79.4 84.7 56.2
15.4 14.8 16.0 14.7 15.8 13.9 14.4 17.3 10.9 14.2
aWeighted. bIncluding
myocardial infarction, angina, and congestive heart failure.
3.4 5.0 7.0 4.4 4.6 3.6 4.3 4.2 3.9 4.5
13.6 10.4 8.8 12.0 10.0 9.8 9.0 15.0 9.5 8.5
Higher function, not self-care (%) 11.0 8.8 10.2 10.4 7.7 12.2 8.1 9.3 11.8 7.4
Higher Self-care, function not higher function and self-care (%) (%) 9.1 10.3 8.0 6.4 7.0 8.7 7.7 11.3 7.2 7.3
23.7 19.1 16.3 25.8 17.1 25.1 16.2 22.4 41.3 14.5
Association of Comorbidity with Disability
All diseases were also significantly associated with having “any task difficulty” (data not shown).
Multivariate Analyses: Independent Associations of Each Disease with Disability Types We then evaluated the independent association of each of the nine diseases and conditions with disability, adjusting for the other diseases as well as confounders in a main effects model (results are shown in Table 2, A columns). Almost all of the nine diseases were associated (P , 0.05) with all of the disability outcomes. The exception was with upper-extremity difficulty only, for which only two of nine of the diseases were associated; here, power to detect associations was somewhat weaker than for other comparisons, so that the magnitude of the nonsignificant odds ratios for heart disease, diabetes, and high blood pressure exceeded some that were significant in other disability groups. Overall, in these main effects models, the presence of significant associations did not provide more specific information on associations of disease with disability than was found for the “any difficulty” group. Some sense of specificity can be derived, however, from the relative strengths of associations. For example, vision and hearing impairments were most strongly associated with difficulty in the higher-function task group; self-care difficulty (without higher function difficulty) was twice as strongly associated with arthritis as with any other disease. Generally, the findings in Tables 2 and 4 were consistent. The major differences were that the association of arthritis with many disability types appears stronger after adjustment (in Table 2, A columns) than in the unadjusted prevalences (Table 4).
Associations of Disease Pairs with Disability Types Finally, we evaluated the associations of pairs of diseases, as well as the individual diseases, with disability, using multiple logistic regression models adjusting for age, race, education and MiniMental State exam score (see Methods). Table 2, B columns, shows the odds ratios from these models for (1) the main effects, representing independent effects of single diseases, and (2) the pairs of diseases in which the odds of disability was synergistically increased in the presence of both diseases, compared with neither disease being present and adjusting for the effect of each disease alone. Several observations emerge from these analyses. First, selected disease pairs were significantly and synergistically associated with disability. Given that 36 disease pairs were evaluated in these exploratory analyses, approximately two significant interactions (synergistic and otherwise, at the 0.05 level) are to be expected by chance alone per model. Although the number of synergistic interactions alone exceeded this for the more-severe disability types, interpreta-
33
tion should consider scientific coherence of results and strengths of associations. An overwhelming tendency to synergy was not observed. There were, however, important disease pairs that recurred in their associations with different disability types. These included the joint presence of arthritis and visual impairment, arthritis and high blood pressure, heart disease and cancer, lung disease and cancer, and stroke and high blood pressure. Second, the type of disability with which a disease was associated varied, depending on the other disease that was present. For example, vision impairment in combination with arthritis was associated with mobility task difficulty, but paired with stroke they were jointly associated with difficulty in both higher-function and self-care tasks (Table 2, B columns). Third, when interactions were accounted for, many diseases no longer appeared, in themselves, to be independently associated with a given type of disability. Most notably, with interactions in the models, neither high blood pressure nor visual impairment remained associated with any disability type, by themselves, except in one instance. For almost half of the disease pairs involved in interactions, each single disease was no longer significantly associated, by itself, with that disability subtype (Table 2, B columns). Overall, the association of disease pairs with disability groups showed more specificity than did the main effects alone. Some disease interactions showed specificity of association with only a few disability groups; others (e.g., heart*cancer) showed consistent associations with a number of disability types. Figure 2 summarizes these findings graphically for five disease pairs, compared with the individual diseases. The triangles in each panel indicate the odds ratios for the association of the pair of diseases when there was a synergistic interaction (P # 0.10), and the square and diamond symbols indicate the odds ratios for the association of each disease, singly, with each disability type (odds ratios are drawn from data in Table 2, B columns). Figure 2A shows that the joint presence of both heart disease and cancer was more strongly associated with difficulty in five disability types (upper extremity, both mobility and upper extremity, higher function tasks without self-care difficulty, and selfcare difficulty without higher-function task difficulty, and higher function and self-care) than was expected by multiplying the single-disease associations; the joint presence of the two diseases was not synergistically associated with mobility difficulty alone. In contrast, Figure 2B through 2E shows disease pairs in which the synergistic association was only important in relation to one or two disability types. Arthritis and heart disease (2B) were jointly associated with difficulty in higher-function or self-care tasks. Arthritis and stroke (2C) were jointly associated with self-care task difficulty (not higher-function task difficulty). Arthritis and hearing (Figure 2D) and vision and stroke (2E), as pairs, were each synergistically associated with difficulty in both higher function and self-care tasks.
34
L. P. Fried et al.
FIGURE 2. Odds ratios for the association of important synergistic interactions between disease pairs and for the independent associations of each of the two diseases, with disability types. For each disability type, “no disability” serves as the reference comparison group. For disease types, reference is “neither disease.”
Association of Comorbidity with Disability
Associations of Individual Diseases and Confounders in Models with Interaction Terms In terms of main effects, report of any task difficulty showed disease associations that appeared most consistent with the findings in the group with mobility only or both mobility and upper-extremity difficulty (Table 2, B columns). The main effects for other disability outcomes showed substantially different patterns of association than for “any difficulty.” In addition to self-reported disease, cognitive function, as measured by the MiniMental State Exam score, was associated (P , 0.05) with “any difficulty” and with difficulty in the two combination groups (both mobility and upperextremity difficulty, and difficulty in both higher-function and self-care tasks), but with none of the other disability outcome measures. Age greater than 85 years was associated with all disability outcomes, and education was associated only with “any difficulty” and with the two groups with difficulty in the higher-function tasks (P , 0.05). Race was not independently associated with any disability outcome measure, except for those with difficulty in higher function but not self-care (data not shown).
DISCUSSION The goals of this study were to describe the prevalence and co-occurrence of multiple chronic diseases in a representative population of older women and to explore how pairs of comorbid diseases were related to self-reported disability and whether disease–disability relationships differed depending on the type of disability considered. Paralleling previous studies, we found that a number of chronic diseases were highly prevalent and that comorbidity was common [3,4,7,15], that disease and disability were highly associated [1–12], and that disease was more highly explanatory of disability than were age, race, or educational status [2]. Beyond these observations, this study performed exploratory analyses that indicated that specific pairs of chronic diseases were synergistically associated with different types of disability. Certain of these interactions appear importantly related to disability; that is, the associations were large and well beyond the multiplicative effects of the two diseases independently. It appears from the results of these multivariate analyses that a substantial amount of the contribution of disease to disability may derive from the interactions of specific comorbid diseases. If this is proved to be the case in future studies, then a new, potentially effective strategy for prevention or amelioration of disability would be to decrease targeted disease – disease interactions. Thus, clinicians would target prevention of a comorbid disease that could cause disability given the one already present or would target treatment to decrease severity of one or both comorbid diseases already present in order to eliminate the interaction.
35
Evaluation of disease interactions potentially sheds light on etiologic relationships in a way that looking only at independent associations does not. This is apparent when comparing the associations of individual diseases with disability types when interactions were (Table 2, B columns) or were not (Table 2, A columns) accounted for. In the main effects only models (Table 2, A columns), almost every disease was associated with almost every disability type. When interactions were considered, however, there was much more specificity apparent in the relationships. For example, in Table 2 (B columns), arthritis was interactively associated with mobility in the presence of visual impairment or high blood pressure but showed no independent association itself with mobility, whereas it was independently associated with upper-extremity difficulty. Given the high prevalence of arthritis, it is unlikely that the lack of association between arthritis alone and difficulty in mobility is a result of inadequate power. Thus, comorbidity may substantially heighten the risk of mobility difficulty associated with arthritis, as has been indicated by other studies [3,8,12]. Thus, when there is a significant synergistic interaction, there are at least two qualitative interaction effects that can be operating that could be both biologically plausible and important: first, that having both diseases is associated with substantially higher risk of disability than is expected by combining the individual disease effects; and, second, that one of the diseases is associated with disability only in the presence of the other disease. These effects have been previously reported by Verbrugge et al. [3] when looking at disability broadly as an outcome. Instances of both can be found in this study: The first scenario is exemplified by the interaction of visual impairment and stroke in association with higher-function and self-care difficulty; the second scenario is exemplified by heart disease and cancer in association with upper-extremity difficulty. The predominant pattern in the analyses evaluating interactions was of two diseases in combination being strongly associated with disability, while neither or only one was independently associated. Additionally, there were two types of disease combinations associated with excess disability risk—those that confer a great deal of excess risk and those that are prevalent and confer at least moderate excess risk. Inferentially, it is likely that those with more severe disability (i.e., IADL and/or ADL difficulty) have more severe underlying disease; perhaps a greater severity of a single disease is required for that disease to cause disability by itself. Conversely, in the presence of less-severe disease, interaction with other diseases may be necessary to produce disability. This could explain why some diseases were not associated with what is thought to be less-severe disability in themselves, but they were associated in specific disease pairs. In these data we were not able to assess disease severity to test this question. The role of disease severity will require further evaluation. The joint effect of two diseases equals that which would
L. P. Fried et al.
36
be expected by directly multiplying the individual effects, a multiplicative interaction. The absence of a synergistic interaction does not rule out an increased functional impact from the presence of two diseases, compared to one. The lack of an interaction could also result from inadequate power to rule out interactions. Although prior studies have looked at synergy between two diseases in causing disability in general [3] or in relation to mobility difficulty [3,7,8,12,17], this study is the first to show that there may be specificity in the effect of disease pairs on different types of disability. In fact, the patterns of interaction varied in their associations by disability type. These results also indicate that partitioning disability into mutually exclusive subtypes is potentially more informative than looking only at the presence of “any difficulty.” The specificity of associations that is apparent by comparing diseases associated with mobility difficulty with those associated with other disability types would not be revealed by a simple summary measure of disability. Such specificity could, potentially, guide clinical evaluation of the underlying causes of disability. That is, given a patient with difficulty in a self-care task but not in higher-functioning tasks, the clinician might be able to target identification of the individual diseases potentially associated with this disability and the combinations of diseases that confer high risk. We have characterized the main effects of the models with interaction terms (Table 2, B columns) as roughly reflecting effects of diseases in isolation. This is a substantial approximation—both because some triplets and higherorder combinations of diseases likely interact in even more complicated ways, and because some of the diseases considered rarely appear in isolation. While in-depth examination of these complexities is beyond the scope of these exploratory analyses, we did assess whether the approximation was reasonable using analyses that restricted to two diseases at a time (say, heart disease and cancer), as follows: We computed the probabilities of being in each disability category for each of the following: neither disease, heart disease only, cancer only, or heart disease and cancer. Then, we computed the odds ratios for each disease group relative to no disease, comparing odds of being in a specific disabled category to odds of being not disabled; in doing this, to approximate adjustment for other diseases that is made within our multivariate models, we stratified analysis by the number of other diseases reported—two or fewer other diseases versus two or more diseases. In fact, these simple analyses did agree quite well with our Figure 2. However, they also pointed out that: (a) there was evidence of higher-way interactions, so that interactions between disease pairs appeared amplified in women with a larger versus a smaller “other” disease burden; (b) for the rarer diseases that do not often occur in isolation (stroke, for example), the main effects may be underestimated and the interactions correspondingly overestimated; and (c) for the more common
diseases, the multivariate model information draws more heavily from persons with a lesser disease burden, and the opposite is true for the rarer diseases. In-depth examination of these complexities is beyond the scope of these exploratory analyses, but they should be kept in mind in designing confirmatory studies on the effects of comorbidity. The findings reported here provide a basis for hypothesis development and to guide additional research and testing of individual synergistic relationships in a more in-depth manner. Given that these data are cross-sectional, prospective studies are also needed to determine the causal nature of these relationships. If hypothesis testing, especially in prospective studies, confirms these observations, this would suggest several directions for both prevention and treatment of disability. Clinicians evaluating potential etiologies of disability may look to the pairs of diseases that should be treated if a particular disability is present, as well as an individual cause, with the goal of minimizing the severity of at least one disease so as to diminish the interaction. Similarly, for disability prevention, individuals with one high-risk chronic disease present should be carefully screened for prevention of other diseases that could interact to cause disability. Thus, confirmation of the results of this study could add a new perspective to the import of disease prevention, placing emphasis on the prevention of comorbidity as a goal in itself. Meaningful decreases in disability could potentially result from interventions to decrease the synergy between diseases. Supported by the National Institue on Aging (Contract No. NO1AG12112 and Grant # AG11703-05). Dr. Fried was a Henry J. Kaiser Family Foundation Faculty Scholar in General Internal Medicine during the conduct of this study. Dr. Bandeen-Roche is a Brookdale National Fellow in Aging Research. The authors thank Mr. Raymond Burchfield for his excellent preparation of this manuscript and Ms. Carol Han for preparation of the figures.
References 1. Fried LP, Guralnik JM. Disability in older adults: Evidence regarding significance, etiology, and risk. J Am Geriatr Soc 1997; 45: 92–100. 2. Fried LP, Ettinger WH, Lind B, Newman AB, Gardin J. Physical disability in older adults: A physiological approach. J Clin Epidemiol 1994; 47: 747–760. 3. Verbrugge LM, Lepkowski JM, Imanaka Y. Comorbidity and its impact on disability. Milbank Q 1989; 67: 450–484. 4. Guralnik JM, LaCroix AZ, Everett DF, Kovar MG. Aging in the Eighties: The Prevalence of Comorbidity and Its Association with Disability. Advance Data from Vital and Health Statistics; No. 170, Hyattsville, MD: National Center for Health Statistics; 1989. 5. Mor V, Murphy J, Masterson-Allen S, Willey C, Razmpour A, Jackson ME, et al. Risk of functional decline among well elders. J Clin Epidemiol 1989; 42: 895–904. 6. Fried LP. Older women: Health status, knowledge, and behavior. In: Falik M, Scott Collins K, Eds. Women’s Health: Commonwealth Fund Survey. Baltimore, MD: Johns Hopkins University Press; 1996, 175–204.
Association of Comorbidity with Disability
7. Guralnik JM, LaCroix AZ, Abbott RD, Berkman LF, Satterfield S, Evans DA, et al. Maintaining mobility in late life: Demographic characteristics and chronic conditions. Am J Epidemiol 1993; 137: 845–857. 8. Ettinger WH, Davis MA, Neuhaus JM, Mallon KP. Long-term physical functioning in persons with knee osteoarthritis from NHANES-I: Effects of comorbid medical conditions. J Clin Epidemiol 1994; 47: 809–815. 9. Campbell AJ, Busby WJ, Robertson MC, Lum CL, Langlois JA, Morgan FC. Disease, impairment, disability and social handicap: A community based study of people aged 70 years and over. Disabil Rehabil 1994; 16: 72–79. 10. Almy TP. Comprehensive functional assessment for elderly patients. Ann Intern Med 1988; 109: 70–72. 11. Guccione AA, Felson DT, Anderson JJ. Defining arthritis and measuring functional status in elders: Methodological issues in the study of disease and physical disability. Am J Public Health 1990; 80: 945–949. 12. Verbrugge LM, Lepkowski JM, Konkol LL. Levels of disability among U.S. adults with arthritis. J Geriatr Soc Sci 1991; 46: S71–83.
37
13. Fried LP, Kasper JD, Guralnik JM, Simonsick EM. The Women’s Health and Aging Study: An Introduction. In: Guralnik JM, Fried LP, Simonsick EM, Kasper JD, Lafferty ME, Eds. The Women’s Health and Aging Study: Health and Social Characteristics of Older Women with Disability. NIH Pub. No. 95-4009; Bethesda, MD: National Institutes of Health, National Institute on Aging; 1995: 1–8. 14. Folstein MF, Folstein SE, McHugh PR. Mini-mental state: A practical method for grading the cognitive state of patients for the clinician. J Psychiatr Res 1975; 12: 189–198. 15. U.S. Senate Special Committee on Aging, the American Association of Retired Persons, the Federal Council on the Aging, and the U.S. Administration on Aging. Aging America, Trends and Projections. 1991 Edition; 107–142. 16. LaPlante MP. Data on disability from the National Health Interview Survey 1983–85. Washington, DC: National Institute on Disability and Rehabilitation Research, U.S. Department of Education; 1988. 17. Verbrugge LM. Disability transitions for older persons with arthritis. J Aging Health 1992; 4: 212–243.