This Article Abstract Full Text (PDF) Services Download to citation manager Citing Articles Citing Articles via HighWire PubMed PubMed Citation

Cognitive Function and Acute Care Utilization

^a Center for Health Economics Research, Waltham, Massachusetts
^b Department of Health and Social Behavior and the Department of Epidemiology, Harvard School of Public Health, Boston, Massachusetts

Edith G. Walsh, Research Triangle Institute, 411 Waverley Oaks Road, Suite 330, Waltham, MA 02452-8414 E-mail: ewalsh{at}rti.org.

Decision Editor: Fredric D. Wolinsky, PhD

	Abstract

TOP Abstract Conceptual Model Hypotheses Methods Results Discussion References

 
Objectives. Little is known about variation in cognitive function across the aged population, or how use and costs of health care vary with cognitive impairment. This study was designed to create a typology of cognitive function in a nationally representative sample, and evaluate acute care use in relation to cognitive function, holding constant confounding factors. By including proxy assessments of cognitive function, this is the first study to include individuals unable to respond themselves.

Methods. We analyzed the baseline year of the Asset and Health Dynamics Among the Oldest Old (AHEAD) survey, sponsored by the National Institute on Aging, to create three levels of cognitive function, using direct measures for self-respondents (n = 6,651) and proxy evaluations for the others (n = 792). We used a two-part model to predict the likelihood of using various health services and to evaluate intensity of care among users.

Results. Sixteen percent, 64%, and 20% of the sample fell into the low, moderate, and high cognitive function groups, respectively, that differed significantly on almost all demographic and health status measures, and some utilization measures. Controlling for other health and functional status measures, lower cognitive function had a significant and negative effect on outpatient services, but did not affect hospital use directly.

Discussion. Lower cognitive function may be a barrier to outpatient care, but these analyses should be repeated using administrative use and cost data.

ALTHOUGH 18% of those 75–84, and 45% of those 85 and over are estimated to be cognitively impaired (Evans 1990), and researchers estimate close to 2 million Americans have Alzheimer's disease (AD; Hy and Keller 2000), we have limited knowledge of variation in cognitive function in the aged population, or of how health care use and costs vary in relation to cognitive function. Such basic knowledge is essential for cost-effectiveness studies regarding treatments for diseases that cause cognitive impairment, and valuable for policy and program development in the face of a large and rapidly increasing aged population (Albert and Drachman 2000). This study is the first step in estimating health care utilization and costs for people with AD and other dementias. We have used survey data from the National Institute on Aging's (NIA's) Asset and Health Dynamics Among the Oldest Old (AHEAD) survey that includes direct measures of cognitive function in a nationally representative sample of community-residing persons, permitting us to create a typology of cognitive function and link cognitive function to acute care utilization. We report levels of cognitive function in the community-residing population age 70 and over, and the relation of cognitive function to selected measures of acute care utilization, including hospital, physician services, medication, and dental care. In addition, by including proxy assessments of cognitive function, this is the first study to include, rather than ignore, those individuals who are not able to respond themselves.

Previous studies of resource use and cognitive impairment include cost studies using prevalence rate estimates and utilization data to estimate average total costs associated with AD (Ernst and Hay 1994; Huang, Cartwright, and Hu 1988) and primary data collection, usually involving small samples of persons with AD or diagnosed memory disorders (Andersen, Andersen, and Krach-Sorensen 2000; Ernst, Hay, Fenn, Tinklenberg, and Yesavage 1997; Gutterman, Markowitz, Lewis, and Fillit 1999; Rice et al. 1993; Stommel, Collins, and Given 1994; Weinberger et al. 1993), or otherwise nonrepresentative samples (Binder and Robins 1990; Callahan, Hendrie, and Tierney 1995; Hanlon et al. 1996). These studies report high utilization of community-based and institutional long-term care services, high average costs for supportive care, large out-of-pocket costs, substantial contributions of in-kind services, and high total economic costs. There are also large, population-based studies that include extensive measures of cognitive function, like the Established Populations for Epidemiologic Studies of the Elderly. The National Health and Nutrition Survey has a national, representative sample and some measures of cognitive function. These epidemiologic studies provide information about variation in cognitive function in the community elderly population, trajectories of cognitive decline, and factors associated with cognitive function, but do not examine resource use. These studies find impaired cognitive function is associated with increased age, lower levels of educational attainment, and current and future impairments in activities of daily living (ADLs; Burt and Harris 1994; Colsher and Wallace 1991; Gill, Williams, Richardson, Berkman, and Tinetti 1997; Moritz, Kasl, and Berkman 1995; Scherr et al. 1988), and that family members often fail to recognize memory problems (e.g., Ross et al. 1997).

Several large studies do evaluate the relationship between cognitive function and acute care utilization, although their samples are not representative, and each study includes a limited selection of covariates. In these studies, cognitive impairment is associated with increased hospital use, increased emergency room use, and decreased medication use (Binder and Robins 1990; Callahan et al. 1995; Hanlon et al. 1996).

Identifying the population with dementia is one of several challenges inherent in developing estimates of health care utilization and costs. Many health care utilization and cost studies use claims-based approaches to stratifying a sample by diagnosis, which works well for acute conditions, but dramatically underestimate the prevalence of persons with AD or other dementias, as they do for other chronic conditions (Gutterman et al., 1999; Newcomer, Clay, Luxenberg, and Miller 1999). Even where available, reliance on diagnosis of AD or other dementias would limit a sample to those who have sought care for dementia or are at a relatively advanced state of illness leading to a formal diagnosis. Yet, population-based studies indicate that there are many older men and women with decreased cognitive function who have not been diagnosed with dementia (e.g., Andersen et al. 1997; Di Carlo et al., 2000) and Callahan and colleagues 1995 found less than 25% of those with moderate-severe impairment had a dementia diagnosis in their medical record.

This study goes beyond the limitations of previous research by analyzing data from a large, representative national sample for whom there are detailed measures of cognitive function for the full sample, including the most impaired. The cognitive function measures are extensive and validated, and there are detailed measures of important covariates, including ADL and instrumental ADL (IADL) function, financial status, education, and living arrangements. There are sufficient numbers of minority group members, and of those 85 years old and over, to allow subanalyses and between-group comparisons. In this study, we used self- or proxy-reported information about acute care utilization. Future studies will link individual characteristics from the survey to administrative data for direct measures of health care utilization and costs. We address the following research questions:

1. How does level of cognitive function vary in the community population age 70 and over?
   
2. How does utilization of acute care services differ according to level of cognitive function, holding constant confounding factors?
   

	Conceptual Model

TOP Abstract Conceptual Model Hypotheses Methods Results Discussion References

 
Our conceptual model places cognitive function in a dynamic interaction with demographic characteristics (including socioeconomic status), health status, and functional status in predicting resource use. Cognitive function can be affected by each of the covariates, and has both direct and indirect effects on resource use. Direct effects of cognitive function on resource use include diagnostic testing and prescription medications directly related to the diagnosis or treatment of cognitive impairment. Increasing cognitive impairment also has indirect effects on resource utilization when it results in decreased ADL and IADL function and the need for assistance with those functions. Impairments in ADL and IADL function, whether related to cognitive impairment or not, also carry additional risks for acute care utilization, including rehabilitative services and utilization resulting from complications of inadequately compensated ADL or IADL impairment, such as decubitus ulcers or dehydration (Pope, Adamache, Walsh, & Khandker, 1999; Walsh 1998). Demographic characteristics may also have direct effects on utilization (e.g., insurance coverage) or indirect through their effects on health status, cognitive function, and functional status. Similarly, health status has a direct effect on resource use, but can also have indirect effects, for example, if poor health status leads to job loss and hence loss of insurance coverage. Supply factors also affect resource use, for example, if few dentists accept Medicaid reimbursement or provide care for impaired individuals.

	Hypotheses

TOP Abstract Conceptual Model Hypotheses Methods Results Discussion References

 

1. Increasing levels of cognitive impairment will be associated with decreased use of ambulatory care. This hypothesis is consistent with findings reported by Callahan and colleagues 1995 and Hanlon and colleagues 1996. Individuals with cognitive impairment (and their caregivers) may not recognize or be able to communicate health problems requiring outpatient care, they may be difficult to transport either because of ADL impairments or inability to understand or cooperate, or they may be unwelcome in some settings.
   
2. Increasing levels of cognitive impairment will be associated with increased inpatient utilization, including both admissions and length of stay. Also consistent with previous research, this hypothesis is based on several assumptions. Those sick enough to require hospitalization will not face the same barriers we hypothesize for outpatient care. Caregivers will notice they are ill, and an ambulance can be called if necessary. To the extent cognitive impairment is correlated with physical impairments, there will be an increased risk of health problems severe enough to warrant hospitalization; once hospitalized, those with cognitive impairment require more extensive discharge planning, thus resulting in longer stays. Finally, if there are barriers to ambulatory care, lack of outpatient treatment could lead to more severe illness that results in hospitalizations for ambulatory care-sensitive conditions, such as pneumonia or urinary tract infections.
   

	Methods

TOP Abstract Conceptual Model Hypotheses Methods Results Discussion References

 
Based on our conceptual model, we developed a typology of cognitive function and linked that typology to utilization holding relevant covariates constant in multivariate analyses. We created base models, including only cognitive function, demographic and insurance variables, and expanded models incorporating additional measures of health and functional status expected to be related to health care utilization and available in the data set.

Data and Sample
We used the first wave of the AHEAD survey, which is sponsored by the NIA. The AHEAD survey is a biennial panel study designed to allow analyses of the impacts and interrelationships of changes and transitions for older Americans in three domains: health, financial, and family (Soldo, Hurd, Rodgers, and Wallace 1997). It is designed to be a nationally representative sample of the community-residing population. In its baseline year, 1993, the AHEAD sample includes data from interviews with 8,222 individuals, including 7,443 individuals age 70 and over, the remainder being spouses under age 70. We used the baseline year (1993) and selected the 7,443 respondents who were age 70 and over at that time. Approximately 11% (792) of this sample responded by proxy. Although there are some sample members for whom there are financial proxies (i.e., for whom proxies responded to the questions about income and assets while the sample member or another caregiver responded to the rest of the survey), this distinction is not relevant to our study.

Measures
Dependent variables.
The AHEAD survey provides limited measures of acute care utilization, all of which are self- or proxy-reported. We constructed several measures of acute care utilization in the previous year. Hospital use, physician use, medication use, and dental use are defined as dichotomous variables, and used to determine the probability of service use. To examine intensity of use, we also constructed continuous variables for the number of nights hospitalized, the number of hospital stays, the number of physician contacts, and the number of medications taken each month.

Primary independent variable: cognitive function.
Our primary independent variable is cognitive function. The AHEAD survey includes direct measures of cognitive function for self-respondents, and proxy measures in cases where the subject refused or was unable to respond. The items included and baseline results are described in detail by Herzog and Wallace 1997. For self-respondents, Herzog and Wallace developed a series of 35 items, including immediate free recall; delayed free recall; Serial 7s to test working memory; and tests of knowledge, language, and orientation. The core items are derived from the Telephone Interview for Cognitive Status (TICS; Brandt, Folstein, and Folstein 1988) that is modeled after the Folstein Mini-Mental State Exam (1975). The TICS was chosen as the basis of the AHEAD measures because those aged 70–79 were routinely interviewed by telephone, as were those aged 80 and over who preferred a telephone assessment to an in-home interview. Individuals received scores ranging from 0 to 35, 10 possible points from immediate free recall, 10 from delayed free recall, 5 from the Serial 7s, 2 from counting backward, and 1 each for the correct day of the week, day, month, and year, as well as being able to name cactus, scissors, the president and vice president. The scoring combines two indices: the memory index that ranges from 0 to 20, and the mental status index from 0 to 15. We followed the approach used by Herzog and Wallace to identify three levels of cognitive function based on the total score for these items. We coded the 917 individuals (14% of the self-respondents) with scores of 0–12 as having low cognitive function, those with scores ranging from 13 to 24 as having moderate cognitive function (4,510 or 68%), and we classified those with scores of 25–35 as having high cognitive function (1,224 or 18%).

For those who refused or were unable to participate in the interview (n = 792), proxies were asked to rate the sample members on a series of attributes and behaviors. From these, we selected memory, judgment, organization, getting lost in familiar places, and wandering off as measures of cognitive function. We created a set of dichotomous variables indicating a problem in each of these areas and then summed these from 0 (no problems) to 5 (problems in all of these areas). Memory, judgment, and organization were each coded as problematic if the proxy rated the sample member as fair or poor in that area. Getting lost in familiar places, or wandering off, were each coded as a problem if they ever occurred. Higher scores were associated with more problems. Cronbach's alpha for this scale was 0.78, indicating that the items are highly correlated and made a credible scale (DeVellis 1991). Using this scale, 36% had no problems with cognitive function, 29% had 1–2 problems, and 35% had three or more problems with cognitive function.

One of our key challenges was to develop a unified typology of cognitive function incorporating both the proxy and self-respondents, even though different measures were used for the two groups. We considered this task key because excluding proxies, as done previously, undoubtedly would exclude highly impaired persons. However, it would be inappropriate to assume that all proxy responses are due to cognitive function, because there are other reasons for proxy responses. We conducted detailed analyses of the reasons for proxies, and the relationship between the cognitive measures and reason for proxy. As reported in Table 1 , we found substantial variation in the reason for proxy: 73% citing ill health, 21% refusal by the sample member, and 6% citing language barriers. There is also variation in cognitive function across these groups. For example, 58% of those too ill to complete the survey themselves were judged to have a memory impairment, compared with 22% of those who refused, and 32% of those who did not speak English or Spanish. As seen in Table 2 , more than one third of the proxy respondent group had no evidence of cognitive impairment by these measures, including 25% of those who needed a proxy because of ill health. Finally, we integrated the proxy cognitive function measures into the three levels of cognitive function derived for the self-respondents. The combined typology, integrating proxy and self-respondents based on cognitive function, is summarized in Table 3 .

Proxy status.
We created a dummy variable indicating proxy status (proxy response = 1). Because proxy response may be a marker for poor health, for the availability of others to note health problems and seek health care services, and for better recall, we hypothesized that proxy respondents would report higher rates of health care use than the self-respondents, all else being equal.

Demographic characteristics.
We included age (measured in years), gender (female = 1), urban versus rural (urban = 1), living alone, and dummy variables for Black, White, Hispanic, and other race (with White and other race combined as the omitted group in the multivariate analyses), and education level measured as having at least a high school education. We did not include marital status, because it was highly correlated with living alone, which we considered the more important of the two measures to include.

Because income is commonly discussed in the policy literature in terms of the federal poverty limit, we constructed income measures taking into account the total number of persons in the household and total income, compared with the federal poverty level (FPL) in the study year (1993). We used this to create four levels: income at or below poverty, 10–200% of poverty; 201–300% of poverty, and income greater than 300% of the FPL (the omitted group in the multivariates). In 1993, the FPL for a single individual was $6,930, and $8,740 for a two-person household. Our hypothesis was that lower income levels, although a risk factor for poor health, would serve as barriers to health care.

Insurance coverage.
Virtually all of the sample (97%) reported having Medicare coverage. However, there is variation in the proportion with additional coverage, including Medicaid coverage and other additional health insurance coverages, such as Veterans Administration or Medigap policies. Although Medicaid is associated with low income, the Medicaid eligibility guidelines in many states are lower than the FPL, individuals who may be income eligible may have assets greater than those allowed, and it is well known that many eligible persons are not covered by Medicaid. Indeed, there was only a 32% correlation between Medicaid and income under the FPL in this sample. Our assumption is that additional insurance would facilitate access to care and would be associated with increased outpatient utilization.

Health status and health behaviors.
Health status was specified through self-rated health (excellent, very good, good, fair, and poor, with excellent used as the reference group in the multivariate analyses). We also constructed a count of health conditions from 0 to 10, based on self-reported high blood pressure, diabetes, cancer, lung disease, heart disease, angina, stroke, serious injury, arthritis, and emotional problems. We included two health behaviors: current cigarette smoking (any amount) and alcohol intake. We had two competing hypotheses in mind regarding cigarette smoking. The deleterious effects of smoking could increase utilization. Alternatively, smoking might be associated with decreased health-seeking behavior. If the latter was true, we would expect less outpatient utilization among those who smoke, but perhaps the same or more inpatient utilization. Drinking behavior is specified through two dummy variables based on the self-reported amount of alcoholic drinks typically consumed per day, with moderate drinking coded for consumption of 1-2 drinks per day, and heavy drinking indicated for three or more drinks per day. Heavy alcohol consumption is also a risk factor for poor health, and hence increased utilization, although light-to-moderate alcohol consumption may have health benefits. Like smoking, heavy alcohol consumption might also be associated with less help-seeking behavior.

Functional Status.
We included three measures of functional status: ADLs, IADLs, and a count of lower body limitations. Our measures of ADL and IADL impairments are dichotomous, based on the presence of one or more ADL or IADL impairments. We examined ADL and IADL counts (and report the means in our descriptive findings), but found the relationship to utilization more strongly related to having any limitations at all. Conceptually, functional status served as a general marker of health status in this study, compared with its use as a need indicator in studies related to long-term care or home health utilization. In our previous work with the Medicare Current Beneficiary Survey, we have found lower body limitations to be associated with increased Medicare costs (e.g., Pope et al. 1998), and researchers have shown them to be precursors of ADL and IADL limitations (Guralnick et al., 2000). As a result, we included a count of lower body limitations (0–4), based on reported difficulty walking several blocks, difficulty climbing the stairs, difficulty pulling or pushing large objects, and difficulty lifting or carrying weights more than 10 pounds.

Analytic strategy.
After creating a typology based on cognitive function, we compared the characteristics and health service utilization of the three groups as defined by low, moderate, and high cognitive functions. We used chi-square tests to evaluate whether the observed differences were statistically significant and F statistics to test coefficient differences for continuous variables. Our dependent variables included both dichotomous and continuous variables. We employed a two-part model to separate the probability of acute care use from the intensity of use. We used logistic regression to assess the impact of cognitive function on the probability of any use of hospitalization, physician services, medications, and dentists in the last 12 months. We used ordinary least squares (OLS) regression to assess the impact of cognitive function on the number of nights hospitalized, the number of physician contacts, and the number of medications taken each month among those who used these services.

Because of the complex multistage sample design of the AHEAD survey, we used STATA software for all descriptive statistics and multivariate analysis. STATA SVY commands provide correct standard errors and p values not available in statistical packages that assume a simple random sample. Without adjusting for the complex sampling design effects, the standard errors would be underestimated, which could lead to falsely rejecting the null hypothesis.

	Results

TOP Abstract Conceptual Model Hypotheses Methods Results Discussion References

 
Descriptives
In Table 4 , we present the weighted variable means, standard errors, and coding algorithms for the independent variables. Sixteen percent of the sample fell into the lowest cognitive function category, 64% fell into the moderate cognitive function group, and 20% fell into the highest functioning group. The mean age of the sample was 77.5 years, with 62% being female, 72% living in urban areas, and 37% living alone. The sample is 85% White, 10% African American, 4% Hispanic, and 1% other races. Only 57% of the sample had a high school education or more. Almost one half of the sample had incomes at or below 200% of the FPL. Only 12% had family incomes above 300% of the FPL. Nine percent reported Medicaid coverage, and 76% reported at least one other source of health insurance coverage in addition to Medicare (e.g., Veteran's Administration or Medigap policies).

Table 5 shows the means, standard errors, and coding algorithms for the dependent variables: health service utilization. Almost one fourth of the sample had at least one hospital stay in the last 12 months. Among these users, the mean number of admissions was 1.6 (SE = 0.071), and the mean number of nights hospitalized was 10.9 (SE = 0.548). Eighty-nine percent reported having at least one physician visit in the last 12 months. Among those with any physician visits, the mean number of physician contacts was 5.5. Eighty-five percent of the sample took at least one prescription medication in the last 12 months, with the mean among users just under 3. Slightly less than half of the sample reported a dental visit in the last 12 months.

Multivariate Analyses
We ran logistic regressions for any hospitalization, any physician visit, any medication use, and any dental visits in the last year (reported in Table 7 ), and OLS regressions on the number of hospitalizations, nights in the hospital, physician contacts, and number of medications (reported in Table 8 ). For each type of utilization, we first constructed a base model, including cognitive function, proxy status, demographic variables, and insurance coverage, but no health status variables to further examine the relation of cognitive function, health status, functional status, and utilization.

In contrast, neither level of cognitive function nor proxy status predicted physician use in the base model, whereas those in the lowest cognitive function group had a significantly decreased likelihood (0.69, p < .05) of physician use in the expanded model. This could imply that decreased cognitive function was serving as a barrier to physician use, even for those with comorbid conditions and lower levels of self-rated health. Alternatively, the finding could reflect that those with impaired cognitive function had poor recall of doctor visits. However, one would assume that proxy responders would be able to recall doctor visits. Because the coefficient for the proxy dummy variable was close to one and not statistically significant, those with lower cognitive function may actually have had fewer doctor visits. Lower income and smoking also significantly reduced the likelihood of physician use, athough having Medicaid coverage or additional insurance coverage had positive effects.

In the base model predicting any medication use, level of cognitive function was not significant, while proxy status, female gender, Medicaid coverage, and other insurance each had a significant and positive effect. In the expanded model, those in the lowest cognitive function group had only 57% the likelihood of medication use (p < .01), whereas those with moderate cognitive function were only 79% as likely to report any prescription drug use in the past year (p < .05), compared with the highest cognitive function group. Proxy status was not significant. As with physician utilization, decreased cognitive function may have been a barrier to prescription drugs regardless of the number of comorbid conditions, the level of self-rated health, or the number of lower body limitations. Alternatively, this lower utilization of prescription drugs may have been clinically appropriate.

Including the health status, variables reduced but did not eliminate the effect on medication use of Medicaid and other insurance, or of being female. However, the effects of health status were quite clear: the likelihood of any medication use increased substantially with the number of health conditions and at each progressively worse level of self-reported health. Smoking and heavy drinking had the opposite effect, decreasing the likelihood of medication use; moderate drinking had no effect. ADL and IADL impairments were not significant predictors of medication use, but with each additional lower body limitation, the likelihood of medication use increased 12% (p > .01).

Dental utilization was the only one of these models in which level of cognitive function and proxy status are significant in both the base and full models, suggesting a direct effect of cognitive function. Compared with those in the highest cognitive function group, those with low or moderate function were less likely to report a dental visit in the last 12 months. Having a proxy respondent also reduced the likelihood of seeing a dentist in the last year, as did increasing age, being Black, having lower income levels, and having Medicaid (in both models). Urban residence (significant in the base model only), having at least a high school education, and having additional insurance coverage all increased the likelihood of reporting a dental visit. In the expanded model, decreasing self-rated health, smoking, and IADL impairment all decreased the likelihood of utilization, whereas the number of health conditions and moderate drinking were associated with an increased likelihood of reporting at least one dental visit.

Regression Analyses
As shown in Table 8 , the level of cognitive function was not significant in either the base or expanded models for the number of hospitalizations nor in either model for the number of nights hospitalized. However, having a proxy respondent was significantly associated with increased reports of utilization in the base model for number of hospitalizations and in both base and full models for the number of nights in the hospital. As expected, the health status measures in the expanded models were significant predictors of both hospital stays and nights hospitalized. ADL impairment increased the number of nights hospitalized, whereas IADL impairment increased the number of hospital stays in the last 12 months. Being a smoker had a small, negative impact on the number of hospitalizations, although it had a larger negative impact on the nights in the hospital, perhaps a function of the smaller number of hospitalizations. Reported heavy alcohol consumption also had a negative impact on the number of hospitalizations, and moderate drinking had no impact on either of these measures of hospitalization.

The number of physician contacts in the past year is defined in the AHEAD data to include both office visits and telephone contacts (the two cannot be disaggregated). In the base model, decreasing cognitive function was associated with an increased number of contacts, compared with those in the high cognitive function group. However, neither level of cognitive function was significant in the expanded model, possibly indicating an indirect effect of cognitive impairment. Having a proxy was associated with a statistically significant increase in the number of physician contacts reported and just missed statistical significance (p = 0.05) in the full model. In both the base and expanded models, urban residence, Hispanic origin, and Medicaid coverage were associated with an increased number of physician contacts in the last year. As expected, the number of physician contacts increased with the number of health conditions and with decreasing self-rated health, with ADL impairment, and increasing number of lower body impairments. Smoking and heavy drinking were significantly associated with decreased physician contacts, low cognitive function was associated with an increase, and moderate cognitive function was associated with decreased number of prescription medications. In the full model, lower cognitive function had a small negative coefficient that was not quite statistically significant (p > .06) on the number of prescription medications. Reported prescription medication increased with the number of chronic conditions, with decreasing levels of self-rated health, and with increasing ADL and lower body limitations.

In summary, in the fully specified models, lower levels of cognitive function were associated with decreased reports of some measures of outpatient care, including reporting any physician contacts, any medications, and dental care. In such models, level of cognitive function was not significant in predicting intensity of outpatient care among users, such as the number of medications taken, and the number of physician contacts in these models. Neither was level of cognitive function predictive of reported inpatient care, including the likelihood of any hospitalization and number of hospitalizations, and number of nights in the hospital among users of hospital care. However, in the base models that excluded additional measures of health and functional status, level of cognitive function was predictive of the likelihood of any hospitalization, intensity of physician contacts, and the number of medications taken. Thus, it may be that cognitive function has an indirect effect on utilization through impacts on health status, or reflect the correlation between cognitive impairment and other health problems.

Sensitivity Analyses
Given the controversial nature of mixing the proxy respondents in with the nonproxies for these analyses, we considered various approaches to modeling the multivariate analyses before deciding to use a dummy variable for proxy response in the multivariate models. We conducted sensitivity analyses modeling utilization for the proxy and self-respondents separately (not shown). Because of the small proportion of proxy respondents, the results for the self-respondents only were essentially the same as those for the full model (run without a proxy dummy) and slightly different from those in which the proxy dummy was included. Moderate cognitive function became a significant predictor in the dental use logistic regression when we incorporated the proxy dummy, whereas living alone was no longer a significant predictor of increased likelihood of receiving any dental care. In the OLS regression models, living alone became significant in predicting the number of nights hospitalized, when the proxy dummy was added to the full model. Because of the much smaller sample size, the proxy-only models had much lower F scores (for the logistics) and R² values (for the regressions), and fewer variables attained statistical significance. However, the direction of the coefficients and the odds ratios were the same for the proxy and self-respondent models in virtually every model.

	Discussion

TOP Abstract Conceptual Model Hypotheses Methods Results Discussion References

 
Main Accomplishments
Previous research has associated cognitive impairment with health care utilization primarily through analyses of individuals with specific diagnoses such as AD, or more recently, by use of medical claims with a diagnosis of dementia. These approaches were limited in their ability to identify the full range of individuals with impaired cognitive function. In this study, we created a typology of cognitive function levels and linked those measures to acute care utilization in a representative sample of the community population age 70 and over. The use of direct measures of cognitive function allowed us to look at utilization independent of diagnosis, thus including people who may as yet be undiagnosed with dementia, but who had some cognitive impairment, as well as persons at the other end of the spectrum. We developed an approach to scoring cognitive function that integrated scores from direct measures of memory and mental status for self-respondents with assessments provided by proxies for sample members who were unable or unwilling to participate directly. This typology offered the first look at cognitive function in the community-dwelling aged population as a whole in relation to other characteristics and to service use. Because of the richness of the AHEAD data set, we were able to incorporate a wide range of demographic, insurance, and health and functional status measures as descriptors and as control variables in multivariate analyses. Once linked to administrative data, the typology could be applied to more complex analyses of the relationship between cognitive function and various types of health care utilization and costs.

This exploratory study provides basic information about the relationship of cognitive function to health care utilization. Although the validity of self- and proxy-reported utilization data is questionable (see Limitations), the study yields interesting findings regarding the possible relationships between level of cognitive function, the covariates, and health care utilization. First, lower cognitive function was clearly associated with demographic and health status factors that increased risk of health care utilization. Second, even controlling for these factors, lower cognitive function had a direct and negative effect on outpatient utilization, including physician services, prescription medications, and dental services. Although premature to attach a lot of significance to the multivariate findings, given valid concerns about recall accuracy (see Limitations), several patterns emerge that warrant further investigation. Overall, those with lower levels of cognitive function appear to be using fewer services than those with higher levels of function. This raises the question of whether individuals with cognitive impairment are underserved, because of their own inability to identify or communicate their symptoms or seek care, their caregivers' inability to identify their needs or transport them to health care providers, or other barriers such as the availability of providers willing to serve them.

The case for interpreting the lower level of service use as potential underservice seems most clear regarding access to dental care, and is consistent with our case study findings in previous research. Clinical guidelines indicate that even edentulous individuals should have an annual dental exam to evaluate the gums and mouth, to look for signs of malignancies, and to look for teeth in need of repair for those who have any. An annual dental visit is recommended to evaluate the condition and fit of dentures (which can change overtime because of weight changes, for example), important for chewing, communication, and comfort. For those with cognitive impairments, regularly scheduled dental care may be even more important, given a limited ability to identify and communicate any problems.

The negative association with physician care and prescription medications also suggested potential underservice and could be contributing to the reported increased use of inpatient care. To the extent outpatient care was not adequately addressing comorbid conditions, overall health, or ADL and IADL impairments, the negative association between decreased cognitive function and access to outpatient services may also have contributed to higher inpatient utilization. This finding suggests that primary care physicians may need to actively evaluate cognitive function and provide aggressive outpatient management even for patients not yet diagnosed with AD or other dementias.

Limitations
The use of survey data, based on recall of past utilization, is a major limitation of this study, and hence interpreting the findings should be approached with caution. Although use of survey data is a common method of studying health care utilization, recall ability varies by respondent and has been shown to underreport intensity of services in particular (Wallihan, Stump, & Callihan, 1999). Recall is particularly suspect among individuals with lower levels of cognitive function. To some extent, we address this issue by incorporating proxy-reported utilization for 11% of the sample. However, it is impossible to determine whether the greater utilization associated with proxy respondents in these analyses is real or caused by their more accurate recall of past events. Because 14% of the sample members who responded for themselves were determined to be at a low level cognitive function, the accuracy of their reports, and of those not quite as impaired, is certainly questionable. We will address this limitation in future research by linking actual utilization and lost data to cognitive function, holding other factors constant.

There are several additional limitations to the study. Although it appears that there are both direct (e.g., on dental use) and indirect (e.g., on hospitalization) cognitive impairments on utilization, we cannot fully tease apart the direct and indirect effects nor the causal relation between cognitive function, health status, and utilization. Because of the lack of diagnostic information, neither can we determine whether the causes of the observed cognitive impairments are the result of a primary diagnosis (such as AD) or secondary to other factors (such as depression). Finally, because this sample was restricted to individuals age 70 and over, it does not provide information about the relationship between cognitive function and health care utilization in the under 65 Medicare or SSI populations.

	Acknowledgments

 
This research was funded by a grant from NIA (Contract 1R03AG15708-01).

We thank Marilyn Albert for her participation in the design of the cognitive function typology. We also thank the editor, Fredric Wolinsky, and four anonymous reviewers for their detailed and thoughtful comments on the initial draft of this article.

Edith G. Walsh and Janet B. Mitchell are now with RTI International, Waltham, MA. Bei Wu is now with the Center on Aging and Department of Community Medicine at West Virginia University, Morgantown.

Received for publication August 28, 2001. Accepted for publication May 6, 2002.

	References

TOP Abstract Conceptual Model Hypotheses Methods Results Discussion References

 

• Albert M. S., Drachman D. A., 2000. Alzheimer's disease: What is it, how many people have it, and why do we need to know?. Neurology 55:166-168. [Free Full Text]
• Andersen C. K., Andersen K., Krach-Sorensen P., 2000. Cost function estimation: The choice of a model to apply to dementia. Health Economics 9:397-409. [Medline]
• Andersen K., Lolk A., Nielsen H., Andersen J., Olsen C., Kragh-Sorensen P., 1997. Prevalence of very mild to severe dementia in Denmark. Acta Neurologica Scandinavica 96:82-87. [Medline]
• Binder E. F., Robins L. N., 1990. Cognitive impairment and length of hospital stay in older persons. Journal of the American Geriatrics Society 38:759-766. [Medline]
• Brandt J., Folstein S. E., Folstein M. F., 1988. Differential cognitive impairment in Alzheimer's disease and Huntington's disease. Annals of Neurology 23:551-561.
• Burt V. L., Harris T., 1994. The third National Health and Nutrition Examination Survey: Contributing data on aging and health. The Gerontologist 34:486-490. [Abstract]
• Callahan C. M., Hendrie H. C., Tierney W. M., 1995. Documentation and evaluation of cognitive impairment in elderly primary care patients. Annals of Internal Medicine 122:422-429. [Abstract/Free Full Text]
• Colsher P. L., Wallace R. B., 1991. Longitudinal application of cognitive function measures in a defined population of community-dwelling elders. Annals of Epidemiology 1:215-230. [Medline]
• DeVellis R. F., 1991. Scale development: Theory and applications Sage, Newbury Park, CA.
• DiCarlo A., Baldereschi M., Amaducci L., Maggi S., Grigoletto F., Scarlato G., et al. 2000. Cognitive impairment without dementia in older people: Prevalence, vascular risk factors, impact on disability. The Italian Longitudinal Study on Aging. Journal of the American Geriatrics Society 48:775-782. [Medline]
• Ernst R. L., Hay J. W., 1994. The US economic and social costs of Alzheimer's disease revisited. American Journal of Public Health 84:1261-1264. [Abstract/Free Full Text]
• Ernst R. L., Hay J. W., Fenn C., Tinklenberg J., Yesavage J. A., 1997. Cognitive function and the costs of Alzheimer disease. An exploratory study. Archives of Neurology 54:687-693. [Abstract/Free Full Text]
• Evans D. A., 1990. Estimated prevalence of Alzheimer's disease in the United States. The Milbank Quarterly 68:267-289. [Medline]
• Gill T. M., Williams C. S., Richardson E. D., Berkman L. F., Tinetti M. E., 1997. A predictive model for ADL dependence in community-living older adults based on a reduced set of cognitive status items. Journal of the American Geriatrics Society 45:441-445. [Medline]
• Guralnik J. M., Ferrucci L., Pieper C. F., Leveille S. G., Markides K. S., Ostir G. V., et al. 2000. Lower extremity function and subsequent disability: Consistency across studies, predictive models, and value of gait speed alone compared with the short physical performance battery. Journal of Gerontology: Medical Sciences 55A:M221-M231.
• Gutterman E. M., Markowitz J. S., Lewis B., Fillit H., 1999. Cost of Alzheimer's disease and related dementia in managed-Medicare. Journal of the American Geriatrics Society 47:1065-1071. [Medline]
• Hanlon J. T., Landerman L. R., Wall W. E., Jr. Horner R. D., Fillenbaum G. G., Dawson D. V., et al. 1996. Is medication use by community-dwelling elderly people influenced by cognitive function?. Age and Aging 25:190-196. [Abstract/Free Full Text]
• Herzog A. R., Wallace B., 1997. Measures of cognitive functioning in the AHEAD survey. Journal of Gerontology: Psychological Sciences and Social Sciences 52B: (Special Issue) 37-48.
• Huang L. F., Cartwright W. S., Hu T. W., 1988. The economic cost of senile dementia in the United States. Public Health Report 103: (1) 3-7. [Medline]
• Hy L. X., Keller D. M., 2000. Prevalence of AD among whites: A summary by levels of severity. Neurology 55:198-204. [Abstract/Free Full Text]
• Moritz D. J., Kasl S. V., Berkman L. F., 1995. Living in an elderly community sample. American Journal of Epidemiology 141:41-49. [Abstract/Free Full Text]
• Newcomer R., Clay T., Luxenberg J. S., Miller R. H., 1999. Misclassification and selection bias when identifying Alzheimer's disease solely from Medicare claims records. Journal of the American Geriatrics Society 47:215-219. [Medline]
• Pope G. C., Adamache K. W., Walsh E. G., Khandker R. Z., 1998. Evaluating alternative risk adjusters for Medicare. Health Care Financing Review 20: (2) 109-129.
• Rice D. P., Fox P. J., Max W., Webber P. A., Lindeman D. A., Hauck W. W., et al. 1993. The economic burden of Alzheimer's disease care. Health Affairs (Project Hope) 12: (2) 164-176. [Abstract]
• Ross G. W., Abbott R. D., Petrovitch H., Masaki K. H., Murdaugh C., Trockman C., et al. 1997. Frequency and characteristics of silent dementia among elderly Japanese-American men. Journal of the American Medical Association 277:800-805. [Abstract/Free Full Text]
• Scherr P. A., Albert M. S., Funkenstein H. H., Cook N. R., Hennekens C. H., Branch L. G., et al. 1988. Correlates of cognitive function in an elderly community. American Journal of Epidemiology 128:1084-1101. [Abstract/Free Full Text]
• Soldo B. J., Hurd M. D., Rodgers W. L., Wallace R. B., 1997. Asset and Health Dynamics Among the Oldest Old: An overview of the AHEAD study. Journal of Gerontology: Psychological Sciences and Social Sciences 52B: (Special Issue) 1-20.
• Stommel M., Collins C. E., Given B. A., 1994. The costs of family contributions to the care of persons with dementia. The Gerontologist 34:199-205. [Abstract]
• Wallihan D. B., Stump T. E., Callahan C. M., 1999. Accuracy of self-reported health services use and patterns of care among urban older adults. Medical Care 37: (7) 662-670. [Medline]
• Walsh, E. G. (1998). Beyond ADL counts and the risk of institutionalization: New measures of need and outcomes in long-term care. Unpublished doctoral dissertation: Brandeis University, Waltham, MA.
• Weinberger M., Gold D. T., Divine G. W., Cowper P. A., Hodgson L. G., Schreiner P. J., et al. 1993. Expenditures in caring for patients with dementia who live at home. American Journal of Public Health 83:338-341. [Abstract/Free Full Text]

This article has been cited by other articles:

				B. Wu, B. L. Plassman, J. Liang, and L. Wei Cognitive Function and Dental Care Utilization Among Community-Dwelling Older Adults Am J Public Health, December 1, 2007; 97(12): 2216 - 2221. [Abstract] [Full Text] [PDF]

				B. S. Bloom, S. Chhatre, and R. Jayadevappa Cost effects of a specialized care center for people with Alzheimer's disease American Journal of Alzheimer's Disease and Other Dementias, July 1, 2004; 19(4): 226 - 232. [Abstract] [PDF]

This Article Abstract Full Text (PDF) Services Download to citation manager Citing Articles Citing Articles via HighWire PubMed PubMed Citation