Am J Geriatr Psychiatry 11:75-82, February 2003
© 2003 American Association for Geriatric Psychiatry
Is the APOE 
4 Genotype Associated With Higher Hospital Costs Among Elderly Patients?
Donald H. Taylor Jr., Ph.D., M.P.A.,
Gerda Fillenbaum, Ph.D.,
Bruce Burchett, Ph.D., and
Dan G. Blazer, M.D., Ph.D.
Received September 14, 2001; revised November 30, 2001, January 18, 2002; accepted January 22, 2002. From the Department of Public Policy Studies, Center for Health Policy, Law and Management, Duke University, Durham, North Carolina (DHT), the Department of Psychiatry and Behavioral Sciences, Center for the Study of Aging and Human Development, Duke University Medical Center (GF,BB,DGB). Address correspondence to Dr. Taylor, 122 Old Chemistry Bldg., Duke University, Durham, NC 27708. e-mail: dtaylor{at}hpolicy.duke.edu
 |
ABSTRACT
|
|---|
OBJECTIVE: The apolipoprotein 
4 (APOE 4) genotype is associated with a number of adverse health outcomes. The authors assessed whether the 4 genotype was associated with higher hospital costs on the basis of data from 1,999 white or black respondents to the Duke Established Population for Epidemiological Studies of the Elderly who consented to be genotyped in 1992–1993. METHODS: They measured hospital costs, using the amount paid by Medicare for hospitalizations from 1992 to 1997. RESULTS: Persons with the 4 genotype did not have higher costs than those who were 4-negative. The highest costs were observed for those who had missing 4 genotype. CONCLUSION: The 4 genotype is not a significant predictor of hospital costs, and so would not be a good risk adjustor for purposes such as setting reimbursement rates for Medicare risk plans.
Key Words: Economics • Genetics • Alzheimer's Disease
|
INTRODUCTION
|
|---|
The literature on apolipoprotein 4 (APOE-4) genotyping can be divided into two distinct, although related, themes. The first category centers on the clinical usefulness of the 4 allele as a predictor of late-onset Alzheimer disease (AD) within the general population and its potential use as a tool for diagnosing or confirming AD in the early stages of dementia in particular patients in a clinical setting.1–11
The second topic focuses on the legal and ethical implications of APOE 4 genotyping,5,12,13 predominantly its relationship to the insurance industry and the possibility of bias with regard to underwriting and insurance premiums, as well as discrimination in the workplace.4,14,15 The resulting legal issues focus on the need for implementation and/or re-examination of regulatory measures to prevent such discrimination.12–16 In both lines of research, the cost and usefulness of APOE testing2,5,6 as a means of predictive assessment in healthy patients with no family history of AD is questionable, given that the costs may outweigh the benefits.3,7,17
This study investigates an emerging area of APOE 4 research, the predictive value of the genotype for understanding healthcare costs, in particular, hospitalization expenditures. Such a linkage is plausible, given the fact that APOE 4 is related to the occurrence of some adverse health outcomes. This line of investigation is important because development of reliable predictors of healthcare utilization is of great importance for purposes such as setting rates for Medicare HMOs. This article addresses several research questions relevant to understanding the possible linkage between APOE 4 and healthcare costs. First, we document the relationship between APOE 4 and hospitalization in a representative older, community-based sample. Next, we quantify the crude and adjusted relationships between APOE 4 and the amount paid by Medicare for hospital care. Third, we determine whether APOE 4 is related to length of stay after hospitalization. Length-of-stay here is a measure of recovery time after an acute hospitalization that may be a more sensitive measure of resource use than Medicare costs because of the prospective nature with which Medicare pays hospitals.
|
METHODS
|
|---|
Subject Sample
Data from this study derive from the Duke Established Population for Epidemiologic Studies of the Elderly (EPESE), a 10-year prospective cohort study that has been described previously.18 This population survey was part of a multicenter, collaborative epidemiological investigation of physical, psychological, and social functioning, and use of health services of persons 65 years of age and older living in East Boston, Massachusetts; Iowa and Washington counties, Iowa; New Haven, Connecticut; and the North-Central Piedmont area of North Carolina.18–20 The North Carolina sample consisted of a stratified household sample of community residents with an oversample of black residents, selected from five contiguous Piedmont counties, one of which was predominantly urban and the other four predominantly rural.
Analysis sample.
Apolipoprotein-E genotype was first assessed at the third in-person interview (P3), conducted in 1992–1993. There were 2,567 respondents to this interview. We excluded persons who were neither white nor black (n=17) and an additional 551 whose APOE 4 genotype was not assessed. The analysis sample comprised the 1,999 respondents who consented to having blood drawn at this interview and whose APOE 4 status was determined.
Data gathered at the same interview included information on age, gender, race, marital status, education, urban versus rural residence, medical conditions (hypertension, heart disease, stroke, diabetes, cancer), cognitive status as assessed by the Short Portable Mental Status Questionnaire (SPMSQ),21 and depressive symptomatology (assessed by the Center for Epidemiological Studies–Depression Scale [CES–D]).22 Medical status was summarized as an index based on the summed weighted measure of the five chronic medical conditions mentioned above.23 The details of determining the APOE 4 genotype of sample members has been described elsewhere.24
Medicare Claims Data
We used Medicare claims data to measure the cost of hospital services received by all members of the analysis sample. We defined cost as the amount that Medicare actually paid for a hospital claim, as reported in the claims data. We examined hospital claims from January 1, 1992, through December 31, 1997, and developed two measures of hospital cost. The first was equal to the total amount Medicare paid for hospitalization on behalf of a sample member during the entire study period. The second measure was the mean cost per hospitalization during this period, which was calculated by dividing the first cost measure by the number of times a subject was hospitalized during the study period; 1,255 respondents were hospitalized at least once between 1992 and 1997. The 744 persons who were not hospitalized had a hospitalization cost value of 0 and were included in both descriptive and multivariate analyses of total cost. Only the 1,255 persons who were hospitalized during the study period were included in analyses of the mean cost per hospitalization.
Other items we used from Medicare claims records included primary discharge diagnosis information, which gives the primary reason for a hospitalization, as determined at discharge (recorded as an ICD-9-CM diagnosis code) and the length of stay of the hospitalization. Length-of-stay was calculated as the difference between the first and last dates of each claim for which Medicare paid for services provided to a sample beneficiary.
Statistical Procedures
We first conducted bivariate analyses to document the relationship between the presence or absence of any 4 allele and hospitalization and cost within the 6-year time interval examined. We used the chi-square test to examine the relationship between hospitalization and each of the explanatory variables noted below, individually. Multivariate logistic regression was used to assess the effect of the explanatory variables on the likelihood of being admitted to a hospital at least once during the study period. We first conducted bivariate analyses of the two cost-dependent variables (study period total costs and mean cost per hospitalization), using the t-test of means to compare costs for both levels of the explanatory variables used in multivariate models. We then estimated median regression models of both cost variables. Median regression was used because the cost-dependent variables were highly skewed. In comparison with ordinary least-squares regression, median regression analyzes variation in the median of the dependent variable over the independent variables (rather than variation in the mean), and median regression fits a regression surface that minimizes the sum of the absolute residuals (rather than minimizing the sums of squares of the residuals).25–27 Median regression also has the benefit of producing coefficients that have a straightforward interpretation in dollars.
The common set of explanatory variables included in multivariate models were the following: First, we controlled for APOE 4 status (positive for APOE 4 versus not). We also controlled for medical status (dichotomized as Poor versus Other), with Poor being a respondent who self-reported two or more of the five chronic conditions noted above. Cognitive impairment was operationalized as a binary measure equal to 1 when a person was cognitively impaired, as determined by the presence of four or more errors on the Short Portable Mental Status Questionnaire (SPMSQ). The SPMSQ, is "intended to offer a rapid screen for cognitive deficit in the community-dwelling elderly (p 120)."28 This measure has been found to be highly predictive of healthcare costs in previous work, even though it may misclassify some persons with clinically relevant dementia.29 We tested the sensitivity of our results to the specification of the cognitive impairment variable (continuous measure, other cut-points) and found that they did not alter our substantive findings. Also, we controlled for demographic characteristics (race [white/black], education [less than high school/high school or more], gender, residence in an urban area versus rural, and marital status [married versus not married]).
|
RESULTS
|
|---|
Approximately one-third of our community-based sample was found to be APOE 4-positive (639 persons; 32.0%), and the remainder (1,360; 68.0%) were not. A majority of the sample persons were black (54.2%); nearly 7 in 10 were female (67.2%), and around half had completed less than 8 years of education (see Table 1).
Nearly two-thirds (1,255) of the 1,999 subjects were hospitalized at least once during the 1992–1997 study period (Table 2). Among persons who were hospitalized, the percentage who were APOE 4-positive was lower (30.9%) than among the proportion who were not hospitalized (33.7%), although the chi-square test was not statistically significant (
2[1]=1.71; p=0.19). Several variables were bivariately related to hospitalization. Persons who were less than 85 years old were less likely to be hospitalized (2[1]=9.53; p=0.002); blacks were less likely to be hospitalized (2[1]=4.53; p=0.03); persons with poor medical status were more likely to be hospitalized (2[1]=37.84; p<0.001); individuals who were cognitively impaired were more likely to be hospitalized (2[1]=9.17; p=0.002); and people who were married were less likely to be hospitalized (2[1]=5.41; p=0.02). Entering all these variables simultaneously into a logistic-regression model, we estimated the probability of admission to the hospital at least once during the study period, and the variables that were strong bivariate predictors of hospitalization generally remained strong predictors after controlling for other factors (Table 3).
The effect of the APOE 4 genotype on unadjusted hospital costs was not statistically significant, regardless of which measure of hospital costs was used (Table 4). Mean cost per hospitalization was $6,988 for persons with the 4 allele, compared with $5,934 for those for whom the 4 allele was absent (t[1, 253]= –1.86; p=0.06). When total hospital cost during the study period was used as the cost measure, the cost differential was even smaller: $12,183 for those with APOE 4, versus $11,758 for those without APOE 4 (t[1, 997]= –0.48; p=0.63). The strongest bivariate predictor of mean cost per hospitalization was being married ($7,152) versus not ($5,807; t[1, 228]= –2.39; p=0.02). Persons who were cognitively impaired had lower costs per hospitalization ($5,161) than did those who were not impaired ($6,491; t[1, 253]=1.92; p=0.05). The only significant bivariate predictor of total hospital costs over the entire study period was having poor medical status; those subjects with poorer medical status had larger costs ($15,139) than those whose medical status was better ($9,659; t[1, 997]= –6.59; p<0.001).
When we controlled for other factors, persons with the APOE 4 genotype did not have significantly increased costs for either cost measure (Table 5). Persons who were cognitively impaired, as well as those who had poor medical status, were found to have higher total costs, after controlling for other variables. These variables were not significant predictors of cost per hospitalization over the same period of time. Persons who were married, were younger than age 85, and who were black had lower total hospital costs; again, these variables were not significant predictors of cost per hospitalization.
We further investigated the effect of APOE 4 genotype on length of hospital stay, contingent upon hospitalization, and found that persons who had the APOE 4 genotype had a significantly longer length of stay than those who did not, after controlling for other factors (Table 6). Persons who were APOE 4-positive spent 2.7 days longer (p<0.001) in the hospital, on average, net of other variables (mean length of stay was 10.7 days). This confirmed a bivariate relationship between length of stay and APOE 4, in which those who were APOE 4-positive had longer length of stay. Persons who were cognitively impaired also experienced a longer average length of stay (2.2 days longer; p=0.01). In addition to the variables we controlled for in the cost models, we also included variables that represented the reasons that persons were admitted to the hospital. Persons who were admitted for stroke had longer average length of stay (3.3 days; p<0.001), as did those who were hospitalized for hip fracture (increased stay by 5.7 days; p<0.001). Those who had heart attacks had shorter average lengths of stay (2.8 fewer days; p<0.001); and persons who were admitted for any type of cancer did not have an average length of stay that differed from all other reasons for hospitalization. All four disease variables identify the effect on length of stay of stroke, hip fracture, heart attack, and cancer versus all other reasons for hospitalization.
|
DISCUSSION
|
|---|
We did not find evidence that persons with the APOE 4 genotype have higher hospital costs than those who do not. Because of the availability of a relatively simple test that allows a person to be genotyped, APOE 4 status could be used to adjust premiums for setting Medicare risk plan rates, for example. However, our results suggest that the APOE 4 genotype is a poor risk adjustor because it does not predict hospital costs. Neither was APOE 4 predictive of being admitted to the hospital, even for some of the diseases cited in the literature, notably heart disease and stroke, whose occurrence has been linked to being APOE 4-positive.30,31
We did find that persons with the APOE 4 genotype had a longer length of stay in the hospital after admission, even after controlling for reason for admission and other measures of health status and demographics. Length-of-stay may be a more sensitive measure of differences in resource use than Medicare hospital costs because Medicare uses prospective payment to reimburse hospitals, which has the effect of reducing variation in payments across patients, but not fixing length of stay. The finding that persons who are APOE 4-positive take longer to recover after a hospitalization exists in spite of the financial incentive for hospitals to reduce length of stay and therefore maximize profitability. This finding was robust to different analytical specifications and analyses, and more work is warranted to better understand this phenomenon, in particular, work that uses diagnosis information obtained from chart review or clinical information to adjust for differential severity of illness.
In spite of the plausibility of a link between APOE 4 and hospital costs, why did we not find such a relationship? First, cost of medical care is difficult to measure, both conceptually and practically. We took the perspective of the cost to the Medicare program, with cost defined as the amount Medicare actually paid for services. It is possible that this measure has subverted some clear cost difference that another measure might detect, but this would require that the bias between measures be systematically related to APOE 4 genotype status, which seems highly unlikely. Second, APOE 4 is a plausible predictor of hospital costs because it has been linked to occurrence of diseases (notably heart disease and stroke) in clinical samples that are important determinants of hospital cost. However, we did not detect a relationship between APOE 4 and hospitalization for heart disease and stroke in the Duke EPESE sample. Although not directly investigated in the literature, it would follow that an increased risk of heart attack and stroke would translate into an increased probability of hospitalization for these causes as well as increased costs; we were seeking to confirm these plausible relationships in a community-based sample, but could not. A possible reason for this discrepancy is that community-based samples are likely to be more diverse than samples obtained through one hospital or clinic. Therefore, clinical samples may introduce selection biases that are not clear a priori, making it important to document relationships between risk factors such as APOE 4 and incident disease in community-based samples to ensure that findings in the literature are not driven primarily by selection effects.
Finally, we have only studied hospital costs, which exclude outpatient costs, such as visiting the doctor or attending specialty clinics. It is possible that a fuller examination of costs would yield a difference by APOE 4 status. This, however, seems unlikely, given that, in previous work, hospital costs have been found to be approximately two-thirds of elderly Medicare beneficiaries' total Medicare-financed healthcare costs.32
The biggest limitation of our study is the fact that 551 persons who answered the P3 survey of the Piedmont Health Survey of the Elderly did not consent to being genotyped. In analyses not shown in this paper, we found that persons with a missing APOE 4 genotype had higher hospital costs than either those with or without the APOE 4 genotype. Those persons who did not consent to being genotyped tended to be older and in poorer physical health and had worse cognitive status. They were also more likely to die within the 5-year period after the P3 observation. Since 4 is related to lower cognitive status, a disproportionately high proportion of those not genotyped may have been 4-positive; if true, this could mean that 4 does predict higher hospital costs. Poor cognitive status is a strong predictor of hospital costs, both in the sample estimated here as well as generally. Furthermore, since hospital costs have been found to be highest in the final year of life, the fact that those who were not genotyped were more likely to die during the study period may alone explain the higher hospital costs of those not genotyped.
Our conclusion that APOE 4 does not have a strong predictive validity for hospital costs argues against the use of APOE 4 as a risk adjustor for projecting future hospital costs and further illustrates that relationships identified in clinical samples may not hold true in population-based samples. This should give rise to hesitancy to make policy for large populations (such as using APOE 4 as a risk adjustor) before documenting key relationships in population-based samples.
|
ACKNOWLEDGMENTS
|
|---|
This study was performed pursuant to NIA grants 1 R01 AG17559 and 1 PO1-AG17937, and contract #N01-AG-2101, but NIA did not conduct this research and is not responsible for this article.
|
REFERENCES
|
|---|
- Holtzman NA: Benefits and risks of emerging genetic technologies: the need for regulation. Clin Chem 1994; 40:1652-1657[Abstract/Free Full Text]
- Mayeux R, Schupf N: Apolipoprotein-E and Alzheimer's disease: the implications of progress in molecular medicine. Am J Public Health 1995; 85:1280-1284[Abstract/Free Full Text]
- Fleck LM: Just caring: the moral and economic costs of APOE genotyping for Alzheimer's disease. Ann N Y Acad Sci 1996; 802:128-138[Medline]
- Kapp M: Medico-legal, employment, and insurance issues in APOE genotyping and Alzheimer's disease. Ann N Y Acad Sci 1996; 802:139-147[Medline]
- Post SG: On not jumping the gun: ethical aspects of APOE gene testing for Alzheimer's disease. Ann N Y Acad Sci 1996; 802:111-119[Medline]
- Frisoni G, Bianchetti A, Trabucchi M: Apolipoprotein-E and the cost of Alzheimer's disease. Int J Geriatr Psychiatry 1997; 12:588-589[CrossRef][Medline]
- Karlawish JHT, Sachs GA: APOE 4 testing for Alzheimer's disease. Journal of the American Psychological Society 1997; 45:1153-1159
- Roses AD: Genetic testing for Alzheimer disease: practical and ethical issues. Arch Neurol 1997; 54:1226-1229[Abstract]
- de Knijff P, van Duijn CM: Role of APOE in dementia: a critical reappraisal. Haemostasis 1998; 28:195-201[CrossRef][Medline]
- Lovestone S: Early diagnosis and the clinical genetics of Alzheimer's disease. J Neurol 1999; 246:69-72[CrossRef][Medline]
- Tsuang D, Larson EB, Bowen J, et al: The utility of apolipoprotein-E genotyping in the diagnosis of Alzheimer's disease in a community based case series. Arch Neurol 1999; 56:1489-1495[Abstract/Free Full Text]
- Natowicz MR, Alper JK, Alper JS: Genetic discrimination and the law. Am J Hum Genet 1992; 50:465-475[Medline]
- Dickens B: Legal issues in predictive genetic testing programs: Alzheimer Dis Assoc Disord 1994; 8:94-101
- Billings PR, Kohn MA, de Cuevas M, et al: Discrimination as a consequence of genetic testing. Am J Hum Genet 1992; 50:476-482[Medline]
- Ostrer H, Allen W, Crandall LA, et al: Insurance and genetic testing: where are we now? Am J Hum Genet 1993; 52:565-577[Medline]
- Gostin L: Genetic discrimination: the use of genetically based diagnostic and prognostic tests by employers and insurers. Am J Law Med 1991; 17:109-144[Medline]
- Van Crevel H, van Gool WA, Walstra GJM: Early diagnosis of dementia: which tests are indicated? what are their costs? J Neurol 1999; 246:73-78[CrossRef][Medline]
- Cornoni-Huntley J, Blazer D, Lafferty M, et al: Established Populations for Epidemiologic Studies of the Elderly: Resource Data Book, Vol. II. Bethesda, MD, National Institute on Aging, 1990
- Blazer D, Burchett B, Service C, et al: The association of age and depression among the elderly: an epidemiological exploration. J Gerontol Med Sci 1991; 46:M210-M215
- Cornoni-Huntley J, Brock D, Ostfeld A, et al: Established Populations for Epidemiologic Studies of the Elderly: Resource Data Book. Bethesda, MD, National Institute on Aging, 1986
- Pfeiffer E: A short portable mental status questionnaire for the assessment of organic brain deficit in elderly patients. J Am Geriatr Soc 1975; 23:433-441[Medline]
- Radloff L: The CES-D Scale: a self-report depression scale for research in the general population. Applied Psychological Measurement 1977; 1:385-401
- Fillenbaum G, Liess J, Pieper C, et al: Developing a summary scale of medical status. Aging: Exp Soc Res 1998; 10:395-400
- Fillenbaum G, Blazer D, Burchett B, et al: Apolipoprotein 4 and risk of mortality in African American and white older community residents. Gerontologist 2002; 42:381-386[Abstract/Free Full Text]
- Chernick MR: Bootstrap Methods: A Practitioner's Guide. New York, Wiley, 1999
- Efron B, Tibshirani R: An Introduction to the Bootstrap. New York, Chapman & Hall, 1993
- Gould WW: Quantile regression with bootstrapped standard errors. Stata Technical Bulletin 1992; 9:19-21; College Station, TX, Stata Corporation, 1992
- Fillenbaum GC: Multidimensional functional assessment of older adults: the Duke Older Americans Resources and Services Procedures. Hillsdale, NJ, Lawrence Erlbaum Associates, 1988
- Taylor Jr. DH, Schenkman M, Zhou J, et al: The relative effect of Alzheimer's disease and related dementias, disability, and comorbidities on cost of care for elderly persons. J Gerontol Psychol Sci Soc Sci 2001; 56B:S285-S293
- McCarron MO, DeLong D, Alberts MJ: APOE genotype as a risk factor for ischemic cerebrovascular disease: a meta-analysis. Neurology 1999; 53:1308-1311[Abstract/Free Full Text]
- Lambert JC, Borusseau T, Defosse V, et al: Independent association of an APOE gene promoter polymorphism with increased risk of myocardial infarction and decreased APOE plasma concentrations: the ECTIM study. Hum Genet 2000; 9:57-61[Abstract/Free Full Text]
- Taylor Jr. DH, Whellan D, Sloan FA: Effects of admission to a teaching hospital on cost and quality of care: evidence from Medicare. New Engl J Med 1999; 318:293-299
TOP
ABSTRACT
INTRODUCTION
