Alzheimer Disease in the US Population:  Prevalence Estimates Using the 2000 Census FREE

CHANGING AGE structure of the US population markedly affects the occurrence of Alzheimer disease (AD). Because the disease has a great effect on the quality of life for affected individuals and caregivers and because it places heavy demands on the health care system, estimates of its occurrence and projections of future occurrence are essential for public health planning.

Available estimates^1- 2 vary in how data from source studies are applied to the US population, and, more importantly, in how AD is ascertained by source studies. Most studies have used a 2-stage approach, with a second stage of expert clinical evaluation required to measure disease. Some studies have used a screening approach, restricting second-stage evaluation for disease to persons failing a screening cognitive test. Others have evaluated disease in random fractions of both those failing and passing such a test, usually in several strata having different probabilities of random selection.

As prevalence estimates derived from restricting disease evaluation to those failing a screening test assume perfect screening test sensitivity,³ while random sampling of all fractions of the population does not, screening approaches typically give lower AD prevalence estimates. Differences resulting from using the 2 methods have been only infrequently examined,⁴ but may be large, depending on the severity of disease a study aims to detect. Gradual development of AD from normality results in a second variation in data collection methods. Although criteria⁵ for the disease reflect good conceptual agreement, translation into an operational cutoff point along the continuum between normality and advanced disease varies. A third variation in data collection is in "blinding" examiners performing the second-stage evaluations for disease to each subject's first-stage cognitive test performance. Although we are unaware of systematic investigations of blinding, judging from its importance in trials,⁶ its effect here seems potentially great.

Analytic variation occurs in applying study estimates to the US population. One method directly applies the prevalence of AD from a study to the number of people in the United States. The alternative life-table approach estimates prevalence from incidence of AD from a study. The first approach is more direct, but the second better adjusts for survival differences and takes advantage of incidence being a more accurate measure of disease occurrence, especially in studies of groups with large cultural differences.⁷

We use incidence estimates from a population study using stratified random sampling and blinded disease evaluation in a biracial urban community. We combine them with 2000 US census and National Center for Health Statistics mortality data to estimate prevalence of AD in the US population. We then project future population disease prevalence, using US Census Bureau high, middle, and low series of population growth projections.

Estimates of AD incidence are from a study⁸ of a biracial (black and white), geographically defined community of 3 adjacent neighborhoods in Chicago, Ill —Morgan Park, Washington Heights, and Beverly. Of all residents older than 65 years, 79% participated. Each study data collection cycle consists of an in-home interview of all participants and clinical evaluation for AD of a stratified random sample. Alzheimer disease incidence was measured in a cohort of 3838 persons determined to be free of AD at baseline, with disease developing during an average interval of 4.1 years between the baseline cycle and cycle 2. Sampling for clinical evaluation was based on age, race, sex, and change in cognitive function from the first to the second cycle home interview of the entire cohort, with persons randomly selected for evaluation from all levels of cognitive change. All 835 persons examined received identical structured clinical evaluations by examiners blinded to population-interview cognitive testing and sampling category. Criteria for AD were those of the of the National Institute of Neurological and Communicative Disorders and Stroke–Alzheimer's Disease and Related Disorders Association Working Group for probable AD,⁵ except that persons who met these criteria and had another condition-impairing cognition were retained.

We used the methods of Brookmeyer et al² with the following modifications. First, we calculated separate incidence estimates for 432 groups defined by single year of age, sex, 2 racial groups, and 3 educational groups, based on a logistic model weighted for the complex sample design with terms for age, age squared, race, educational groups, and follow-up interval. We then computed a weighted average incidence across the educational groups for each age, sex, and racial group. The weights were based on the current level of education from US census data⁹; future education was obtained by aging the current population. Second, instead of using the total population death rate for each age, sex, and racial group as the probability for the disease-free segment and multiplying that by 1.44 for the diseased segment, we distributed the total number of expected deaths between the diseased group and disease-free group using an iterative algorithm to obtain the ratio of 2.14 reported previously.¹⁰ Third, for the numbers of people alive at age 65 years and age-, sex-, race-, and birth cohort–specific death rates, we used National Center for Health Statistics life tables through 1999¹¹ and US census projected life tables up to 2050.¹² For the number of people in the US, we used the 2000 US census¹³ and the US Census Bureau estimates of US population growth through 2050.¹² We repeated the estimation for the year 2000, using different model specifications to examine sensitivity to variations in the logistic model.

We estimated the sample-weighted proportion of persons in each disease severity category by age group, using prevalence data from the baseline cycle of the Chicago study. Severity was classified by score on the Mini-Mental State Examination (MMSE)¹⁴ as mild (≥18), moderate (10-17),^10- 17 or severe (≤9).

In 2000, we estimate there were 4.5 million people in the United States with AD. A fairly small number, 0.3 million (7%), were between the ages of 65 and 74 years, 2.4 million (53%) were between the ages of 75 and 84 years, and 1.8 million (40%) were 85 years of age and older. The estimate was insensitive to variations in statistical method; specifically, excluding a term for black race (not significant), including a term for sex (not significant), modeling educational level linearly by year, modeling educational level in different groups, and modeling age as a single linear term produced results within 0.5 million of the 4.5 million estimate.

In the Chicago population study, with most affected persons still living in the community, 48% of prevalent cases of AD were classified as mild, 31% as moderate, and 21% as severe. Prevalence of disease and proportion of severe disease increased with age (Figure 1). Among persons aged 65 to 74 years, 17% of these cases were severe compared with 20% among persons aged between 75 and 84 years and 28% among persons 85 years and older.

Both the number of persons with AD in the US population and their age distribution will change substantially over the next 50 years (Table 1). Prevalence is expected to increase almost 3-fold to 13.2 million persons in 2050 (Table 1 and Figure 2). To indicate the uncertainty in these future estimates, we bracket the estimates derived from the middle-series US Census Bureau estimates of population growth with estimates derived from the high- and low-series estimates of population growth. For 2010, the estimate derived from the middle series is 5.1 million persons with AD and is bracketed by an estimate derived from the low series of 5.1 million persons and an estimate derived from the high series of 5.3 million persons. By 2050, the uncertainty increases substantially. The estimate of 13.2 million persons with AD derived from the middle series is bracketed by a figure of 11.3 million derived from the low series and 16.0 million derived from the high series.

As rapid growth of the oldest age groups of the US population continues, the age distribution of AD will change (Table 1 and Figure 3). The number of persons aged 75 to 84 years with the disease will double to 4.8 million by the year 2050. The number of persons older than 85 years with AD, however, will more than quadruple to 8.0 million. The number of persons aged between 65 and 74 years with the disease will remain fairly constant at 0.3 to 0.5 million.

Projected decline in death rates among persons older than 65 years in the US population to about half their current annual values by 2050¹¹ will, first, increase the number and proportion of persons who survive to the oldest ages where AD is most frequent, and, second, result in increased survival of persons with AD, and thus increased prevalence. We applied an estimate of the current ratio in survival of 1:2.14 to persons without the illness⁹ and assumed it will remain constant.

These estimates suggest that prevalence of AD in the US population will substantially increase, as older age groups increase in size. Further, the age distribution of the disease will change, as much of this increase will be attributable to the number of persons older than 85 years with AD.

Our estimate of AD prevalence in the year 2000 closely agrees with a projection from the earlier East Boston Study,¹ despite a different statistical method. The number of affected persons projected in 2050 is 35% higher than estimated from the East Boston Study because our methods assumes that projected increases in survival over the next 50 years will affect both persons with and without AD, with the ratio of survival among unaffected persons to survival among affected persons remaining constant at its present level.

Another study² provided lower estimates, especially for the absolute number of persons affected by AD in future years. These differences appear due more to differences in the studies providing estimates of current AD occurrence rather than to fairly small differences in how these estimates of AD occurrence were applied to US Census Bureau estimates of population growth.¹⁵ Both this study and the earlier one suggest marked growth in numbers of persons affected by AD. The estimates from the prior study used 4 different source studies^16- 19 to estimate AD occurrence. One was the incidence phase of the East Boston Study,¹⁹ which used stratified random sampling to ascertain disease. The other 3 studies, however, used methods that gave lower estimates, especially in the oldest age groups. One¹⁷ used only medical records to identify cases; another¹⁸ eliminated the low-scoring 10% of persons from the dementia-free cohort and used a restrictive protocol to detect incident disease. Another¹⁶ examined a highly educated volunteer cohort that was likely healthier than the total population. Because the older subgroups of the US population will experience the greatest growth, these lower estimates of disease in the oldest aged groups result in large differences in projected AD prevalence in future years.

Strengths of this study include an estimation of population prevalence from source study incidence and a source study of a single population of black and white Americans that used random sampling for detailed evaluation from all strata of cognitive test performance and examiners blinded to prior cognitive test results. Weaknesses include a limited proportion of persons of Hispanic origin in this source study. While use of a single, rigorously designed study of AD is a strength, no single community is likely to exactly represent the entire country. The average educational level in the study community was fairly similar to that of the US current educational levels of persons older than 25 years suggesting that there will be little increase in the average educational level of people older than 65 years in the next 50 years. The effect of race on AD occurrence is unclear. We assumed the risk of death for persons with AD compared with that of unaffected persons will remain constant. Life expectancy with AD has a substantial effect on projected prevalence of the disease and is subject to such influences as improvements in care.

These estimates of a substantial increase in AD prevalence assume that the age-, race-, and education-specific risk of disease will remain constant over the next 50 years. The large public health challenge is to make these projections obsolete and irrelevant by discovering routes to the prevention of the illness through better understanding of its underlying biology and by discovery of modifiable risk factors.

Corresponding author and reprints: Denis A. Evans, MD, Rush-Presbyterian–St Luke's Medical Center, Suite 675, 1645 W Jackson Blvd, Chicago, IL 60612 (e-mail: Denis_Evans@rsh.net).

Accepted for publication April 22, 2003.

Author contributions: Study concept and design (Drs Hebert, Scherr, Bennett, and Evans); acquisition of data (Drs Hebert, Scherr, Bennett, and Evans); analysis and interpretation of data (Drs Hebert, Scherr, Bienias, Bennett, and Evans); drafting of the manuscript (Drs Hebert, Bienias, and Evans); critical revision of the manuscript for important intellectual content (Drs Hebert, Scherr, Bienias, Bennett, and Evans); statistical expertise (Drs Hebert, Scherr, and Bienias); obtained funding (Dr Evans); administrative, technical, and material support (Drs Hebert, Bennett, and Evans); study supervision (Drs Bennett and Evans).

This study was supported by grants R01-AG 11101 and P30-AG 10161 from the National Institute on Aging, National Institutes of Health, Bethesda, Md; and by a grant from the Alzheimer's Association, Chicago, Ill.