| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
| ||||||||||||||||||||||||||||||||
Correspondence: Address correspondence to Robert John, PhD, University of Oklahoma Health Sciences Center, Department of Health Promotion Sciences, College of Public Health, PO Box 26901, Oklahoma City, OK 73190. E-mail: Robert-john{at}ouhsc.edu
 |
Abstract |
|---|
 TOP Abstract MethodsResultsDiscussionReferences |
|---|
Key Words: Chronic illness • Health-related quality of life • Functional impairment
Despite the prevalence of comorbid conditions, relatively few studies have addressed this issue. In one study, van den Akker, Buntinx, and Knottnerus (1996) investigated the use of the term in the health sciences literature from 1966 to 1994. Although they point out the lack of consensus regarding the use of the concept, currently, the simplest definition of comorbidity is the presence of two or more health problems in the same individual. Gijsen and colleagues (2001) have shown that most studies between 1993 and 1997 investigated comorbidity from the perspective of a specific or index disease, most commonly cardiovascular diseases (48%), cancers (23%), musculoskeletal diseases (13%), or diabetes (11%). Of the 89 outcomes tested in the 78 articles that investigated the consequences of comorbidity, most (56.2%) investigated the influence of comorbid pairs, an index (37.1%), or a count (12.4%) on the outcome of interest; only 6 studies (7.7%) used more than one measure of comorbidity.
According to Wallace and Lemke (1991), the ways that health researchers have measured comorbidity has advanced our understanding, but there is a need for new and better measures of the health status of individuals that summarize complex disorders. Gijsen and colleagues (2001) agree with this sentiment, specifically calling for more studies of what they label multimorbidity, which they defined—following van den Akker and colleagues (1996)—as the "co-occurrence of two or more diseases within one person, without defining an index-disease" (p. 670). Gijsen and colleagues (2001) explicitly acknowledge the importance of this perspective in describing the health of populations, and McGee and colleagues (1996) advocate better methods to summarize the total burden of disease within a population. According to Guralnik (1996), evaluation of current techniques of assessing comorbidity as well as development of new measurement approaches are needed to advance our understanding.
Studies of comorbidity or multimorbidity reveal that there is no consensus about how the co-occurrence of diseases should be measured (Guralnik, 1996). Researchers have used four basic approaches to the study of comorbidity. A common approach is to count the number of diseases (Guralnik et al., 1989; Verbrugge, Gates, & Ike, 1991). As Guralnik and colleagues (1989) pointed out, the prevalence of comorbidity estimated by this measure is influenced by the number of chronic conditions considered. Verbrugge and colleagues (1989) found an average of 2.7 conditions per person among a representative sample of the noninstitutionalized U.S. civilian population aged 55 years and older. Only 16.4% of the noninstitutionalized population did not have a chronic health problem; 20.5% had a single chronic condition; and the remainder (63.1%) reported two or more chronic illnesses. Fried, Bandeen-Roche, Kasper, and Guralnik (1999) assessed 14 chronic diseases in older women, finding an average of 3 chronic diseases per woman.
A common variant of the simple count is a conditional count (Berkanovic & Hurwicz, 1990; Schellevis, van der Velden, van der Lisdonk, van Eijk, & van Weel, 1993; Seeman et al., 1989; Stewart et al., 1989; Verbrugge et al., 1991). The conditional count is the number of chronic diseases given that the patient has a particular or index disease. Using arthritis as the index disease, Verbrugge and colleagues (1991) found that people over the age of 55 with arthritis experience, on average, 3.8 chronic conditions compared with 1.8 conditions among people without arthritis. Schellevis and colleagues (1993) studied 5 common chronic diseases (osteoarthritis, hypertension, chronic ischemic heart disease, diabetes mellitus, and chronic lung disease) in a sample of patients from general practices in the Netherlands. Among patients aged 65 and older, 32.8% of those with arthritis reported at least one of the other chronic diseases. This study found that people with arthritis reported an average of 1.4 of the 5 chronic diseases assessed. Stewart and colleagues (1989) surveyed adults aged 18 years and older for 9 chronic diseases, and 77.5% of those with arthritis reported at least 1 of the other chronic diseases. Berkanovic and Hurwicz (1990) investigated comorbidity among 288 adults with rheumatoid arthritis who were recruited through clinical referrals. They found that 54% had at least one additional chronic health problem and 20% of the entire sample rated at least one of their additional health problems as severe.
The use of a count or a conditional count has limitations. Although Charlson and colleagues (1987) found that a simple count of the number of comorbid diseases predicted 1-year mortality, they were critical of this approach because it does not consider severity, which they reasoned should be a better predictor of mortality. Verbrugge and colleagues (1989) documented that a count hides the actual condition or combination of conditions that are responsible for a particular outcome within a global measure that contains irrelevant or even moderating conditions.
The second approach considers the severity as well as the number of comorbid conditions. Most commonly, this approach creates a summative index weighted by the severity of the individual conditions or a comorbidity severity score based on the most severe condition (Guralnik, 1996). In a review of the comorbidity literature from 1993 to 1997, Gijsen and associates (2001) identified nine weighting schemes. Undoubtedly, the best known and most frequently used is the Charlson Comorbidity Index (Charlson et al., 1987). This index was created to enhance prediction of 1-year mortality, but it has been used to predict other health outcomes such as functional status. Another index, the Index of Co-Existent Diseases (ICED), was developed to predict in-hospital postoperative complications and 1-year health-related quality of life of patients who underwent total hip replacement surgery (Greenfield, Apolone, McNeil, & Cleary, 1993). A major limitation of the ICED is that it requires medical records and highly trained reviewers who can follow complex decision rules in creating the index. Crabtree and colleagues (2000) developed a comorbidity symptom scale (CmSS) based on a series of questions about 23 medical conditions or sensory or motor impairments and the severity of the perceived symptoms. The CmSS index was significantly correlated with activity of daily living (ADL) impairments, perceived health status, and anxiety and depression.
A third approach to studying comorbidity is to assess the proportion of people who have pairs of comorbid diseases. Typically, this approach studies comorbidity from the perspective of an index disease. If arthritis is the index disease, one assesses how many people have arthritis and vision problems, how many have arthritis and heart disease, and so on. Fried and colleagues (1999) used this approach in their study of older women. They found that the most common comorbid pair was arthritis and visual impairment, with 44% reporting both conditions. Verbrugge and associates (1989) also used this approach to assess the influence of 13 chronic conditions and sensory impairments on disability among community-resident individuals aged 55 years and older. They found that arthritis and high blood pressure was the most common comorbid pair, with 21.1% experiencing both conditions.
A fourth approach to studying comorbidity is to assess the relative association between diseases by using a measure of association such as gamma (Verbrugge et al., 1989) or the odds ratio. This approach is useful in assessing the degree to which comorbid conditions exceed a level expected by chance alone (Guralnik et al., 1989; Verbrugge et al., 1989).
To illustrate this approach, consider a sample of older adults in which both arthritis and diabetes occur at a rate of 50%. If the two diseases are unrelated, then one would expect that 25% of the adults would have both arthritis and diabetes, as a result of mere chance association. Although 25% have both diseases, the odds ratio would be 1, indicating that having one disease does not increase the odds of having the second disease. By contrast, consider two diseases that are relatively rare, with a prevalence of 5% apiece. If they are perfectly correlated, then individuals who have one disease would always have the other. In this case, 5% of the sample will have both diseases. Though the proportion is small, the association is perfect. Thus, in the first case 25% had both diseases, yet there was no association. In the second case, only 5% had both diseases, but the degree of association was perfect.
Verbrugge and colleagues (1989) addressed whether comorbidity occurred more frequently than chance alone. She and her colleagues found that arthritis was implicated in 6 of the top 10 comorbid pairs, including arthritis–high blood pressure (21.1%), arthritis–hearing impairment (14.7%), arthritis–vision disease (8.6%), arthritis–visual impairment (6.3%), arthritis–ischemic heart disease (6.2%), and arthritis–other circulatory system condition (6.2%). However, of these comorbid pairs, only the arthritis–high blood pressure combination occurred significantly more often than predicted by chance alone, and Verbrugge and associates (1989) found that only 9 of 78 comorbid pairs exceeded a level predicted by chance alone.
Together, these studies suggest that all four approaches are needed to get an accurate picture of comorbidity: a count of comorbid diseases, an index weighted by severity of the comorbid conditions, the proportion of those who have pairs of comorbid diseases of interest, and the relative association between diseases.
Unfortunately, we have little understanding of comorbidity or multimorbidity in aging populations. In part, our understanding is limited because few studies have used a variety of approaches (Gijsen et al., 2001), and even fewer have attempted to describe the overall pattern of comorbidity within a given population. For example, Fried and colleagues (1999) studied 14 diseases. Although they reported counts and proportions, they did not report the relative association between diseases. Moreover, they, like many others, defined comorbidity as the presence of two diseases, but a majority of the women in their study (57%) reported three or more diseases. Because most women reported three or more diseases, and because the study limited comorbidity to the occurrence of two diseases, the study did not assess the complex nature of the multiple forms of comorbidity present in most of the sample.
Verbrugge and associates (1989) have conducted the most thorough analysis of comorbidity and its impact on a variety of measures of disability. Despite the abiding value of their contribution to the understanding of comorbidity, Verbrugge and associates (1989) concluded their study by speculating that "there are probably certain triplets and larger clusters of conditions that have powerful effects on disability" (p. 478). Because they were concerned that these larger constructions would lack statistical power to detect reliable effects, they terminated their analysis of comorbidity by considering the effect of comorbid pairs.
Unfortunately, the study of comorbidity among aging American Indians has not progressed to the level achieved by Verbrugge and colleagues (1989). The only study to examine comorbidity among American Indians (Chapleski, Lichtenberg, Dwyer, Youngblade, & Tsai, 1997) investigated the importance of comorbidity on instrumental ADL (IADL) and ADL impairments among three strata of Indian elders over 55 years of age in Michigan (N = 309). The three strata (federally recognized reservations, the urban metropolitan area of Detroit, and state-historic tribal groups who were living in rural areas) were heterogeneous in terms of age, education, smoking habits, drinking habits, and number of chronic illnesses.
Self-reported data were collected on the presence of 21 chronic conditions. Chapleski and colleagues used two measures of comorbidity: a count of self-reported medical conditions and the Charlson Comorbidity Index (Charlson et al., 1987) that weights illnesses by risk of short-term mortality. Individuals in the sample averaged 3.9 chronic conditions (range 0–11) and scored 1.65 on the comorbidity index.
The Indian elders who participated in Chapleski's study experienced few impairments of ADLs. According to Chapleski and associates (1997), 78% of the sample had no limitation in the seven ADLs (eating, dressing, grooming, walking, getting in and out of bed, bathing, and toileting). Approximately 14% experienced a single limitation, 5% two or three limitations, and 3% four or more limitations. Because of their sample size, there were very few elders in the highest impairment categories.
Using hierarchical regression, Chapleski and colleagues found that the comorbidity index accounted for decrements in functioning better than a simple tally of self-reported medical problems. The hierarchical model consisted of four sets of variables: sociodemographic variables (age, gender, education, marriage status, and rural or urban residence), health risk behaviors (smoking and alcohol habits), number of diseases (comorbidity count), and the comorbidity index (CMI). Chapleski and colleagues defended the practice of including both a count of the number of chronic conditions and the CMI (which is a weighted index based on the same information used for the comorbidity count) in the regression equation because it did not represent "harmful multicolinearity" and because the CMI explained additional variance in ADL functioning after the comorbidity count had entered the equation. In the end, the entire saturated model explained 16% of the variance in ADL functioning. However, with all of the variables in the model, only age and CMI were significantly related to level of ADL impairment.
As these studies document, researchers have looked at comorbidity in limited ways for good reason: the combination of two or three diseases gets complex quickly. For example, suppose one wished to compare three diseases such as arthritis, heart disease, and diabetes. We might ask this simple question about relative association of three diseases: Are people with arthritis more likely to have heart disease or are they more likely to have diabetes? In a study with only 10 chronic conditions, when there are 45 comorbid pairs, there are 120 combinations of three diseases to evaluate. To summarize the pattern of associations in this many combinations is a challenging task. We must identify an approach that goes beyond the analysis of comorbid pairs to advance the study of comorbidity.
One way to address this challenge is to use cluster analysis. This method of analysis is a descriptive technique that considers how variables tend to occur in conjunction with each other. Cluster analysis may be useful for summarizing broad patterns of comorbidity in a convenient and interpretable way. With the use of this method, it is possible to go beyond simple comorbid pairs to obtain a general overall picture of the broad pattern of how diseases are associated in a particular population and where a particular disease of interest appears in the pattern.
Despite the common occurrence of multiple chronic health conditions, and the importance of comorbidity in predicting functional impairment, hospitalizations, mortality, and other health outcomes, little is known about the patterns of comorbidity. Even less is known about comorbid conditions among minority elders. To our knowledge, no study to date has reported the extent of chronic illness and patterns of comorbidity within an American Indian population. The purpose of this paper is to study the patterns of comorbidity among a representative sample of American Indian elders living in a rural, reservation environment. We describe the comorbid conditions, with arthritis as the index disease, by using all four standard approaches: a count of comorbid diseases, an index weighted by severity of the comorbid conditions, the proportion of those who have pairs of comorbid diseases of interest, and the relative association between diseases. In addition, we summarize the broad empirical pattern of comorbidity with cluster analysis of the diseases, and we examine the usefulness of this technique in understanding the effects of comorbidity on selected health outcomes relevant to community-resident elders' quality of life.
|
Methods |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
Because many elders are not fluent in English, the instrument was translated into the tribal language. The tribal-language version of the instrument was pretested on a small sample of elders, and the instrument was revised. The instrument was then backtranslated into English, and the questionnaire was evaluated for fidelity to the original intent or purpose of asking each question. Designed as a culturally appropriate, multidimensional functional assessment, the instrument is the first survey research protocol designed specifically to accomplish a comprehensive assessment of the status and characteristics of American Indian elders.
This project surveyed a random sample of 1,039 American Indian elders from one tribe who were aged 60 years and older. The sample was drawn from a complete list of potential subjects that included name, sex, and age supplied by IHS Service Unit records. This technique of subject identification has been independently used and recommended (Kunitz & Levy, 1991). Elders living in each of the IHS Service Units were sampled separately and each case was weighted so that the results would be representative of the entire population of the same age group in this tribe.
After completion of questionnaire revision and sample selection, the lead author conducted the interviewer training with the assistance of tribal personnel. The 1-day training included an overview of the project and general scientific survey research procedures, training in administration of the questionnaire, the administrative organization of the project, and the protection of human subjects (including issues of confidentiality). After completion of interviewer training, the Community Health Representatives (CHRs) and tribal aging services providers conducted a face-to-face interview with each elder selected for the study. Informed consent was obtained from each elder. The mean length of the interviews was 1 hr and 39 min, and the median was 1.5 hr.
Table 1 summarizes the characteristics of the sample weighted to represent the entire population of tribal elders. Several characteristics of the sample are worth noting. The vast majority of tribal elders lack a high school education; half have a monthly family income of less than $481; and few elders are fluent in English. Approximately half are currently married and, compared with the general elderly population, few live alone. Their self-assessed physical and mental health was generally good.
|
|
Results |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
|
|
|
|
|
Cluster Analysis
As a way to look at the broad pattern of comorbidity, a cluster analysis was conducted. The analysis began with a random division of the sample into two groups. For each group, a correlation matrix was computed between the 11 conditions, using phi as the measure of association and average linkage as the combination method. Inspection of the dendrogram for both groups suggested the same four clusters, supporting the reliability of the cluster structure. Next, the analysis was repeated for both groups using Yule's Q as the measure of association and complete linkage as the combination method. Again, inspection of the dendrogram suggested the same four-cluster structure, supporting the reliability of results across methods.
Another way to assess the cluster results is to use an internal criterion (Cormack, 1971; Hubert & Levin, 1976; Jardine & Sibson, 1971; Lorr, 1983). Milligan (1981) conducted a Monte Carlo study of 30 internal criterion measures, and Goodman and Kruskal's gamma emerged as the best measure. An internal criterion such as gamma is a goodness-of-fit measure, comparing how well the model fits the data. The logic of gamma as a goodness-of-fit measure is that variables in the same cluster should be more strongly related to one another than variables not in the cluster. In Milligan's study, the models with clear structure and "added noise" had gammas from.78 to.88. As a way to assess the goodness of fit for the four-cluster solution with the current data, gamma was computed for the total sample, and gamma was.86. Milligan (1981) argued that gamma and other measures of internal criterion could be interpreted as "valid indices of true cluster recovery" (p. 194). Thus, the value of gamma found here suggests that the data contain an underlying structure and that the four-cluster model has recovered the structure fairly well.
On the basis of the results depicted in the dendrogram (available upon request), a four-cluster structure was evident. The largest cluster consists of six chronic health conditions: (a) stroke, (b) heart disease, (c) diabetes, (d) tuberculosis, (e) urinary tract or bladder problems, and (f) high blood pressure. This may be interpreted as a cardiopulmonary cluster. The next largest cluster consists of three health conditions: (a) vision problems, (b) hearing problems, and (c) problems with teeth or gums. This can be labeled a sensory-motor cluster. The third cluster consists of a single condition: depression. The last condition to combine in the analysis is arthritis. The fusion coefficient for arthritis is.20, which means that the average correlation of arthritis with the other diseases is.20.
The cluster analysis identifies patterns of comorbidity beyond mere comorbid pairs. For example, consider diabetes, which occurs in a cluster that includes heart disease but not arthritis. This cluster pattern suggests that diabetes is more likely to occur with heart disease than with arthritis.
One can assess how well the cluster results describe the overall pattern by looking at 36 disease combinations suggested as important by the cluster analysis. The odds ratios (ORs) for the 36 disease combinations can be seen in Table 7. The consideration of diabetes can be used as an example of how to interpret this series of comparisons. Because diabetes occurs in the cluster with heart disease and not with arthritis, one would expect the OR for diabetes–heart disease to be higher than the OR of the combination of diabetes–arthritis. As expected, this analysis shows the pattern predicted by the cluster tree: diabetes has an OR of 5.35 with heart disease, but only 1.88 with arthritis. In general, this series of comparisons support the pattern uncovered by the cluster analysis.
|
In order to determine the utility of the competing approaches to multimorbidity, we tested the explanatory power of each approach on four health outcomes by using multiple regression. The health outcomes included an index of ADL impairment, self-assessed physical health, self-assessed mental health, and how much health troubles stand in the way of doing things. ADL impairment and self-assessed physical health were chosen because these health-related quality of life measures have been used in other studies. To our knowledge, the other outcome measures have not been used in previous studies of comorbidity.
For each of these outcomes, four separate models of multimorbidity were tested. All models included variables for gender, age group, educational attainment, and being married or not married. Model 1 tested the influence of the comorbidity count as the measure of multimorbidity. Model 2 substituted a comorbidity severity index for the count. Model 3 used comorbidity cluster scores (arthritis, sensory-motor, cardiopulmonary, and depression) as the measures of multimorbidity, and Model 4 added the six cluster score interaction terms to the equation used in Model 3. The results reported represent the most parsimonious model.
ADL Impairment
Level of ADL impairment was assessed by asking respondents if problems existed in the performance of six ADLs: bathing, walking, grooming, dressing, getting in and out of bed, and eating. The response options to each of these questions were without help, with some help, or unable. Most tribal elders were able to perform these activities without assistance: eating (98.6%), getting in and out of bed (95.8%), dressing (94.9%), grooming (89.9%), walking (88.8%), and bathing (88.0%). Because few elders experience any of these ADL impairments, a summative index was used as the ADL impairment outcome measure. When the six items were summed to form a scale ranging from 0 (unable to do any of the activities) to 12 (able to do all activities without help), the Cronbach's alpha of the scale was.86, suggesting that the scale is an adequate measure of overall ADL impairment.
Health-Related Quality of Life
Self-assessed physical health, self-assessed mental health, and how much health troubles stand in the way of doing things were single-item questions. The response options for self-assessed physical and mental health were poor, good, or excellent. The response options for how much health troubles stand in the way of doing things were a great deal, a little, or not at all. Higher values represent better self-assessment of health-related quality of life.
Multimorbidity
As can be seen in Table 8, inspection of Model 1 for each of the outcome measures reveals that the count of chronic health problems is routinely the most powerful predictor of the outcomes we investigated, with the exception of ADL impairment. However, the severity index (Model 2) is a better predictor of each of these outcomes than the count. For each outcome, Model 2 explains more variance than Model 1. Similarly, Model 3 (cluster scores) routinely improved the amount of variance explained for each outcome or added greater specificity to the model. For each outcome measure, explanatory power or greater specificity resulted from the inclusion of interaction terms.
|
ADL impairment also reveals how the cluster approach provides very different results than those found with the use of a count or a severity index. Until the interaction term for arthritis and sensory-motor impairments was included in the model, age was the best predictor of ADL impairment (Models 1–3). However, Model 4 reveals that the combination of severe arthritis and multiple sensory-motor impairments is the best predictor of ADL functioning. In fact, the combination of arthritis and sensory-motor troubles predicts ADL impairment almost as well as the entire comorbidity count model (Model 1). In Model 4 advancing age, multiple problems in the cardiopulmonary cluster in combination with severe depression, female gender, and arthritis severity also were related to worse ADL functioning. Overall, Model 4 improves explained variance by 4% over Model 3.
Similarly, for self-assessed mental health, Model 4 explains the most variance and is the most precise model revealing that the combination of sensory-motor troubles and severe depression and the combination of multiple problems in the cardiopulmonary cluster with multiple sensory-motor troubles are the best predictors of self-assessed mental health. The degree that health troubles stand in the way is best explained by Model 4. Multiple sensory-motor troubles, severe arthritis, multiple cardiopulmonary health problems, lower educational attainment, advancing age, and the combination of multiple sensory-motor impairments and severe depression predict the global perception of how much health influences elders' daily life.
|
Discussion |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
The results suggest the importance of measuring comorbidity in multiple ways. If comorbidity is defined as simple pairs, then arthritis has high comorbidity. For example, approximately one third of this sample report arthritis and vision problems. Alternately, if comorbidity is measured by a conditional count or as a relative association, then arthritis has low comorbidity. The cluster analysis revealed that arthritis is a relatively independent chronic health problem in this group of tribal elders. This suggests that arthritis is an important health care need that deserves special attention. Unfortunately, the IHS and other entities responsible for the health care of American Indian elders have not made any special effort to address arthritic conditions among American Indian elders.
Another striking finding of this study is that depression also constitutes a relatively independent health problem. Here, too, the American Indian elderly population is underserved by mental health professionals. Approximately 41% of this sample admitted having been told that they have depression. The findings suggest that depression is another health problem that deserves special attention.
A third result suggests that the sensory-motor cluster is a systematic health care need within this group of elders. Difficulty with vision, hearing, and eating or chewing because of gum disease or loss of teeth constitutes an interrelated health care need of this group of elders. Once again, this is a health care need that the IHS does not address adequately, although tribal assistance with glasses, hearing aids, and dentures moderates this need to some degree. Previous research has suggested that much more has to be done to address the oral health needs of older American Indians who experience high rates of periodontal disease (Skrepcinski & Niendorff, 2000) and among whom the prevalence of complete tooth loss has been estimated to be 42% for patients aged 65 and older (Presson, Niendorff, & Martin, 2000).
The only finding in this study that appears to be relatively expected is the cardiopulmonary cluster of diseases that are the leading causes of mortality generally or are recognized as special health problems in the Indian population (i.e., tuberculosis and diabetes).
Knowing the pattern of comorbidity in a given service population has important implications for health care planning and resource allocation as well as clinical practice. Looking beyond individual diseases or comorbid pairs, by including mental health and sensory and motor impairments and conditions associated with morbidity, pain, and functional impairment but not mortality, strengthens health care planning and delivery. Use of the techniques discussed herein is a better basis for health care planning and delivery than the anecdotally based generalizations that clinicians often make based on their impressions of individual patient encounters, because this method considers the entire community-resident population and a fuller range of health problems than presents itself in clinical settings.
This study has several limitations. First, the presence of the chronic conditions is based on self-reports. Elders in this study have low educational attainment and, presumably, less knowledge of the health information that is typically communicated by health care providers. Because most elders were interviewed in the tribal language, it is possible that "something was lost in the translation." Few elders were fluent in English, and a full account of the cultural differences in basic health concepts between the tribal language and English is not known. Every precaution was taken to reduce the likelihood of this occurrence, but the magnitude of this problem cannot be determined.
Despite these potential limitations, we believe that this research has advanced the conceptualization and measurement of comorbidity or multimorbidity. To some degree, the choice of measures will depend on the purpose of the study and the outcomes of interest. Multimorbidity measures also will depend on data available or the potential sources of information. However, we recommend that future studies of the health of aging populations measure comorbidity in multiple ways. Until the value of our approach is confirmed, use of multiple measures seems warranted.
We believe that several strengths recommend the approach we offer. First, the cluster approach appears to explain important outcomes better than the other measures. Second, the results are consistent across all four health outcomes. Third, cluster analysis makes it much easier to test for interactions, which we have shown are important to all of the health outcomes we investigated. Finally, this approach gives precise and parsimonious results that should be important in primary, secondary, and tertiary prevention.
Our comparison of cluster analysis with other measures reveals that cluster analysis identifies particular health problems that need to be addressed to alter each health outcome. The other approaches suggest that we need to reduce the number (count) or severity of chronic health problems to influence these outcomes. Knowing this leaves open the question of where to begin. We believe that cluster analysis appears to better target particular health problems for prevention or remediation.
Overall, this study clearly supports the importance of preventing comorbidity, and particular combinations of chronic conditions. Probably the most important health issue to address is the prevention of ADL impairments since the existence of these problems signals a host of undesirable outcomes including decreased independence, increased need for caregiving, growing frailty, and risk of institutionalization. Although little can be done to reduce the effects of age, gender, or educational attainment, our findings suggest the importance of addressing several disease clusters and combinations of disease clusters.
Presumably, the patterns of comorbidity found in our study could be expected to change if the results presented here were used to improve health care delivery to this group of elders. Clearly, any special effort to address the chronic conditions in the sensory-motor cluster (vision, hearing, and tooth and gum trouble), provide mental health services to reduce depression, or improve treatment of arthritic conditions could alter the pattern we document. Even more precisely, several combinations of chronic health clusters are salient health care needs within this population of elders. In particular, the combination of severe arthritis and multiple sensory-motor problems was the best predictor of ADL impairment. Similarly, the multiple sensory-motor problems and severe depression combination was a significant predictor of three health outcomes and was the best predictor of self-assessed physical health and mental health. Targeted efforts to prevent or reduce the prevalence of these problems should have a positive effect on the health-related quality of life of this group of American Indian elders.
|
Footnotes |
|---|
1 University of Oklahoma, Health Sciences Center, Oklahoma City.
2 Department of Sociological Studies, University of Sheffield, United Kingdom.
Decision Editor: Linda S. Noelker, PhD
Received for publication November 8, 2002. Accepted for publication March 18, 2003.
|
References |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
This article has been cited by other articles:
 |
|
 |
|
  T. R. Fitzpatrick, S. Aleman, and T. Van Tran Factors That Contribute to Levels of Independent Activity Functioning Among a Group of Navajo Elders Research on Aging, May 1, 2008; 30(3): 318 - 333. [Abstract] [PDF]
|
|
|
|
 |
|
 R. T. Goins, R. John, C. H. Hennessy, C. H. Denny, and D. Buchwald Determinants of Health-Related Quality of Life Among Older American Indians and Alaska Natives Journal of Applied Gerontology, February 1, 2006; 25(1_suppl): 73S - 88S. [Abstract] [PDF]
|
|
| ||||||||||||||||||||||||||||||||
| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
