Trends in Dementia Prevalence, Incidence and Mortality in the United States (2000–2016)

Abstract

Objectives:

The prevalence of dementia has declined in the United States; how this parallels changes in incidence and mortality, and how improvements in educational attainment may have influenced these trends, is not known.

Methods:

Using the Health and Retirement Study (2000–2016), we estimated logistic regression models to examine trends in dementia prevalence and incidence, along with mortality for those with and without dementia.

Results:

The relative decline for dementia prevalence was about 3% per year and 2.0% for dementia incidence. Mortality declined similarly for those with and without dementia. Improved educational attainment accounted for decline in incidence, some of the decline in prevalence, and had a negligible role in mortality.

Discussion:

The declines in dementia incidence provide evidence that dementia prevalence should continue to decline in the near future. These recent declines in incidence and prevalence are largely driven by increased levels of education among older adults.

BACKGROUND

Several studies have reported a decline in age-standardized dementia prevalence in the United States (Hudomiet et al., 2018; Langa, 2014; Rocca et al., 2011). For example, using the Health and Retirement Study (HRS), Langa et al. (2017) found a decline in prevalence for adults 65 years and older from 11.6% in 2000 to 8.8% in 2012. Less clear, however, are the ways in which dementia incidence as well as mortality among people with dementia are contributing to the downward trend in prevalence. This is true for both prevalence trends up to this point and future trends which will be driven by current and earlier incidence and mortality patterns. Our study sheds new light on the connections between dementia prevalence, incidence, and mortality by documenting recent trends for the 2000–2016 period among Americans 65 years of age and older.

The current prevalence of any condition reflects the stamp on the surviving population of past trends in incidence as well as the length of time that people live with the condition. For conditions such as dementia that are not cured by treatment, the length of time with the condition is defined by onset and mortality. Thus, current trends in incidence and mortality provide insight into how prevalence may change in the near future. Although localized studies document recent declines in incidence (Derby et al., 2017; Gao et al., 2019; Rocca et al., 2011; Satizabal et al., 2016), information on national trends in dementia incidence for the United States is sparse. Using the National Health and Aging Trends Study, Freedman et al. (2018) reported a decrease in dementia incidence over a 4-year period, 2011–2015, for persons without vascular conditions. They also found that mortality trends for people with and without dementia remained stable for the 4-year period. Longer term trends are less clear.

We are able to document these trends over a more extended period of time, 2000–2016, drawing on nine waves of the longitudinal Health and Retirement Study (HRS), one of the most commonly used national surveys for studying trends in cognitive health (Prince et al., 2016). In examining these trends, it is important to note that the 16-year prevalence trend has been influenced by incidence and mortality trends operating prior to the period that have left their stamp on the surviving population. The observed incidence trend, on the other hand, while some influence on prevalence, has clear implications for both current and future prevalence. Mortality trends can also influence dementia prevalence by impacting the number of people living with dementia relative to the total older adult population. For example, if mortality rates for people with dementia decline, then the total number of people living with dementia will increase due to increased survival, which will drive up dementia prevalence, so long as the mortality rate for people without dementia is unaltered. Any combination of the two trends will impact future prevalence. We acknowledge that our analysis is restricted to the documentation of trends in the underlying demographic forces that drive prevalence. A formal demographic assessment of how prevalence, incidence and mortality are interrelated is not possible given the limited time frame of the data (to evaluate the impact of incidence and mortality trends on dementia prevalence in this period we would have to obtain data that predate the survey observation period).

Explanations of observed declines in prevalence and incidence have often been linked with increases in the educational level of the population (Freedman et al., 2018; Hayward et al., 2021; Skoog et al., 2017). Higher levels of education are associated with a lower risk of dementia which has been theorized to operate primarily by building cognitive reserve or improving cognitive development (Clouston et al., 2019; Crimmins et al., 2018; Levine et al., 2018; Zhang et al., 2016). Cognitive development (or reserve) in early life is then carried forward, which lowers the risk of having dementia (for a review, see Lövdén et al., 2020). Recent studies provide evidence that is consistent with this idea. For example, the increase in the educational attainment of older Americans explained about half of the decline in life expectancy with dementia between 2000 and 2010, and half of the increase in life lived cognitively intact (Crimmins et al., 2018).

In contrast to dementia incidence and prevalence, the associations between higher educational attainment and mortality with and without dementia are mixed. Among people with dementia, the risk of mortality is higher with greater levels of education (Alley et al., 2007). Therefore, at the population level, the increase in educational attainment among older adults with dementia may lead to increasing mortality across this time period. In contrast, for people without dementia, mortality risk decreases with greater levels of education (Lleras-Muney, 2005; Montez et al., 2012). Therefore, increasing educational attainment should lead to a declining mortality trend. Taking these two patterns into account, the improvements in education among older adults from 2000 to 2016 may influence the mortality trends in diametrically opposite ways.

The current study builds on this prior work by examining how education is associated with the trends in the specific components, which drive changes in prevalence over time.

METHODS

DATA

To examine the trends in dementia prevalence and incidence, as well as mortality for people with and without dementia, we use the HRS from 2000 to 2016. The HRS is a large, biannual nationally-representative study of adults over age 50 in the United States. The sample is also replenished every 6 years with younger cohorts to continue being nationally representative of Americans 50 years and older. The survey collects comprehensive socioeconomic and health information and tests cognitive performance for those able to perform the tests. Proxy respondents provide information on cognitive ability for those unable to do the tests as well as those unwilling to answer for themselves. Our analytic sample is limited to adults 65 and over at each observation wave; cognitive performance was not assessed for persons younger than 65 years of age.

The prevalence analysis is based on age-eligible person observations at each of nine waves. In total, the nine survey waves yielded 97,101 person-observations. Dementia incidence and cognitive status-specific mortality are based on changes in cognitive status and mortality between adjacent observation waves approximately two years apart. To analyze these changes, we constructed a person-interval file for the 8 intervals: 2000–2002, 2002–2004, 2004–2006, 2006–2008, 2008–2010, 2010–2012, 2012–2014, and 2014–2016. The “risk set” for dementia incidence refers to people classified as not having dementia at the beginning of each interval and alive at the end of the interval (because respondents needed to provide cognitive status information) (N=68,562). The definition of “risk set” for mortality among persons not classified as having dementia is slightly larger because it also includes people who have died (N=77,672). The “risk set” for mortality among persons with dementia consists of people who have dementia at the beginning of the interval and were alive or dead at the end of the interval (N=9,739). Detailed information on the number of person-observations (prevalence) and person-intervals (incidence and mortality) is provided in Table 1.

Table 1.

Number of Person Observations for Dementia Prevalence and Number of Person Intervals for Dementia Incidence, Mortality without Dementia, and Mortality with Dementia by Wave, HRS (2000–2016)

	Dementia Prevalence		Dementia Incidence	Mortality for People with Dementia	Mortality for People without Dementia
Year	# of Person Observations	Year Interval	# of Person Intervals
2000	10725
2002	10930	2000–2002	8213	1334	9370
2004	11104	2002–2004	8659	1198	9704
2006	11395	2004–2006	8853	1186	9905
2008	11342	2006–2008	9101	1237	10124
2010	10941	2008–2010	8879	1171	10168
2012	10736	2010–2012	8720	1232	9702
2014	10366	2012–2014	8349	1225	9502
2016	9562	2014–2016	7788	1156	9197
Total	97101	Total	68562	9739	77672

VARIABLES

DEMENTIA

Building on other HRS-based studies of dementia, we use a classification scheme based on cognitive test performance for self-respondents or proxy scores for those who do not answer for themselves. The scheme was developed based on validating HRS scores using a clinically-evaluated neuropsychological examination on a subset of the HRS called the Aging Demographics and Memory Study (Crimmins et al., 2011). The self-respondent score is based on a battery of tests used to assess cognitive functioning: immediate recall of 10 words, delayed recall of 10 words, 5 trials of serial 7s and backward counting. Proxy reports are based on a proxy assessment of memory, an interviewer assessment of difficulty completing the interview because of cognitive limitation, and a report of difficulty in performing 5 instrumental activities of daily living such as meal preparation, grocery shopping, taking medication, making a phone call and managing money. While the scheme allows for classifying respondents as cognitively normal, cognitively impaired without dementia (CIND), and having dementia, because our objective is to document trends in the most severe form of cognitive impairment, we adopt a dichotomous approach focusing on dementia compared to not having dementia (this category includes adults with CIND or that are cognitively normal).

MORTALITY

The HRS tracks the mortality of respondents using the National Death Index (for mortality from 2000 to 2011) or exit interviews in which proxy respondents provided information on death. The accuracy of the mortality tracking in the HRS is of very high quality and complete based on period/cohort comparisons with life tables (Weir, 2016).

TREND

To capture the trends over the 2000–2016 period, a trend indicator (YEAR) is constructed identifying “years since 2000.” The range for the prevalence analysis is 0–16 (2000 – 2016), while the range for the incidence analysis is 0–15 (respondents were first at risk in 2000–2002 and last at risk in 2014–2016). This indicator should be interpreted as the number of years since the baseline wave (prevalence) or baseline interval (incidence and mortality). This indicator allows us to examine the annual decrease across this time period.

OTHER VARIABLES

Gender is collected on each respondent (0-Male, 1-Female). Age is measured in years. Education is measured as years of completed schooling, ranging from 0 to 17.

ANALYSIS

To examine the dementia prevalence, incidence, and mortality trends over this historical time period (2000–2016), we use logistic-regression models. For the prevalence trend estimation, we are assessing how the age-standardized proportion of people who drop below the threshold score for dementia changes across the survey years. The dominator is all eligible persons at each survey wave. These models are based on a person-observation file. For the incidence and mortality trends, we estimate parallel sets of Markov-based logistic regression models. These models are based on changes in dementia status or mortality between adjacent survey waves. For example, the incidence in the 2000–2002 period is estimated from the number of people with a dementia score in 2002 divided by the number of people without dementia in 2000. Incidence is measured in the same way for the remaining 2-year time periods. And mortality is also measured in this way. These models are based on person-interval files. Additionally, these models use a Markov assumption in that prior history is ignored and the incidence rates refer to the risk of an event occurring from time t to time t +1. This approach is analogous to the calculation of mortality incidence in period-based life tables.

We evaluate models of mortality for people with and without dementia because both mortality trends will impact future prevalence. For example, if mortality trends with dementia decrease faster than mortality trends without dementia, then dementia prevalence will most likely increase because the proportion of older adults with dementia will grow relative to the proportion of older adults without dementia. In contrast, if mortality trends for people with and without dementia decline at similar rates, then the declines in mortality would have no impact on dementia prevalence because a constant change in mortality across different states would create a stable population (Schoen, 1988). We also tested the difference between the two trends in a supplementary model (see Table S1).

Models are specified for each outcome as follows: Model 1 contains the trend indicator identifying years since 2000, with age and gender as covariates to adjust for changes in population composition. Model 2 introduces educational attainment, and changes in the parameter estimates for the trend indicator for Models 1 and 2 are used to evaluate whether educational attainment may account for the trend.

Additionally, we conducted several sensitivity analyses that included different functional forms of age and education (see Tables S2, S3, and S4). We found no notable differences.

All models are adjusted using the sampling weights provided by the HRS. This adjustment ensures that our estimates are nationally representative.

RESULTS

Overall, dementia prevalence, dementia incidence, and mortality for people with and without dementia (separately) declined from 2000 to 2016 (see Tables 2, 3, and 4). The declines, however, were not uniform. The relative annual decline in prevalence was about 3.0% [OR .970 (.966-.975), Table 2]. The relative annual decline in incidence was about 2.0% [.980 (.972-.987), Table 3]. Mortality decline for respondents with and without dementia was 0.8% [OR .992(.981–1.003), Table 4] and 1.3% [OR .987 (.979-.995), Table 4]—a fairly similar decline for both groups (we also found no statistically significant differences between the two trends in additional analysis, Table S1). The total declines for dementia prevalence, incidence, and mortality with and without dementia are illustrated in Figure 1. Prevalence declined from 11.5% in 2000 to 7.7% in 2016, totaling a 33.0% reduction. Incidence declined from almost 5.0% in 2000–2002 to 3.8% in 2014–2016, totaling a 24% reduction. And mortality declined from 32.7% in 2000–2002 to 30.3% in the 2014–2016 for people with dementia, a 7.3% reduction, and for people without dementia from 6.3% in 2000–2002 to 5.4% in 2014–2016, a 14.3% reduction.

Table 2:

Time Change in Dementia Prevalence 2000–2016: HRS

	Model 1			Model 2
	OR	95% CI	p	OR	95% CI	p
Year	0.970	0.966–0.975	<.0001	0.987	0.982–0.992	<.0001
Age	1.123	1.120–1.127	<.0001	1.119	1.115–1.123	<.0001
Female	1.118	1.063–1.176	<.0001	1.118	1.060–1.178	<.0001
Education				0.812	0.807–0.818	<.0001

Note: Year: 0 to 16 (2000–0; 2002–2; 2004–4; 2006–6; 2008–8; 2010–10; 2012–12; 2014–14; 2016–16)

N=97,101 person observations

Table 3:

Time Change in Dementia Incidence 2000–2016: HRS

	Model 1			Model 2
	OR	95% CI	p	OR	95% CI	p
Year	0.980	0.972–0.987	<.0001	0.995	0.988–1.002	0.168
Age	1.113	1.107–1.119	<.0001	1.111	1.105–1.117	<.0001
Female	1.089	1.019–1.165	0.013	1.062	0.987–1.143	0.103
Education				0.827	0.816–0.839	<.0001

Note: Year: 0 to 14 (2000–2002 – coded as 0 as this interval is the baseline for the incidence analysis, 2002–2004 – coded as 2; 2004–2006 – 4; 2006–2008 – 6; 2008–2010 – 8; 2010–2012 – 10; 2012–2014 – 12; 2014–2016 – 14.

N=68,562 person intervals

Table 4:

Change in Mortality among Those with and without Dementia 2000–2016: HRS

With Dementia
	Model 1			Model 2
	OR	95% CI	p	OR	95% CI	p
Year	0.992	0.981–1.003	0.141	0.987	0.976–0.998	0.018
Age	1.066	1.059–1.074	<.0001	1.066	1.059–1.073	<.0001
Female	0.707	0.638–0.783	<.0001	0.686	0.617 – 0.762	<.0001
Education				1.067	1.051 – 1.083	<.0001
Without Dementia
	Model 1			Model 2
	OR	95% CI	p	OR	95% CI	p
Year	0.987	0.979–0.995	0.003	0.991	0.982–0.999	0.027
Age	1.100	1.096–1.104	<.0001	1.099	1.095–1.103	<.0001
Female	0.669	0.636–0.704	<.0001	0.660	0.628–0.694	<.0001
Education				0.959	0.946–0.973	<.0001

N = 9,739 with dementia; without dementia 77,672

Figure 1.

Trends in Predicted Dementia Prevalence and Predicted Transition Probability for Dementia Incidence, Mortality with and without Dementia, adjusted for age and gender, HRS (2000–2016)

Additionally, changes in educational composition had mixed results on dementia prevalence, dementia incidence, and mortality for those with and without dementia. Controlling for education either reduced, eliminated, or had no effect on the time trend, respectively. For dementia prevalence, the inclusion of education, or assuming that education did not change across the time period, the OR changed from .970 to .987 (Table 2). The smaller odds ratio, however, remains statistically significant. We interpret the significance as indicating that a 1.3% annual relative decline in dementia prevalence was not explained by improvements in educational composition of the population. In contrast, changes in educational composition largely explained the reduction in incidence. When educational attainment was included in the model, the effect of the trend indicator was reduced to statistical insignificance (OR .995 (.988–1.002), Table 3). The result indicates that reduction in dementia incidence for older Americans from 2000 to 2016 can be attributed largely to improvements in educational composition of the older adult population. For mortality among people with and without dementia, educational composition had a negligible effect, signifying that changing educational composition did not influence the time trend in mortality among older adults with dementia.

In addition to these models, we also preformed sensitivity analysis to assess whether our estimations of state probabilities (i.e., prevalence) for dementia and transition probabilities for incidence and mortality would be unduly influenced by the functional form modeling of age and education effects. We tested models with age, age-squared, and the natural log of age. We also tested whether education’s association was non-linear (e.g., education and education-squared). As shown in Table S2, the time trend coefficients for incidence were nearly identical: for the models without education, the coefficient was exactly the same .980. For models with education, the coefficients were .995 and .996. As shown in Table S3, the time trend coefficients for mortality without dementia barely changed. In models without education, the time coefficient ranged from .986 to .988, while with education coefficients ranged from .990 to .991. Lastly, for models of mortality with dementia, we observed similar patterns across all models as well (Table S4). The time trend coefficient was .992 for model without education and ranged from .986 to .987 for models with education. In total, we found no significant differences in the patterns observed across all supplementary tables, providing evidence that our presented findings are robust.

DISCUSSION

The decline in dementia prevalence over the 2000–2016 period reflects the history of dementia incidence (onset) and survival prior to 2000 as well as trends over the period. Altogether, we observed a decline in dementia prevalence, dementia incidence, and mortality for adults with and without dementia from 2000 to 2016. The declining trend in dementia prevalence is similar to findings from other studies. While the absolute numbers of dementia cases increased in the United States due to population aging (“2020 Alzheimer’s Disease Facts and Figures,” 2020), the proportion of older adults with dementia decreased, suggesting declining risk in the recent past and currently. We also documented a concurrent decline in dementia incidence. This finding suggests that prior declines are continuing to be mirrored in recent years. Additionally, because dementia prevalence is largely a process of previous trends in mortality and incidence, the decline in incidence suggests that future prevalence rates should also decline, barring any significant changes in mortality.

In our analysis of mortality, we did not find any conclusive evidence that differential mortality trends will impact near future dementia prevalence. While mortality rates for people with dementia declined in this period, this coincided with a decrease in mortality among those without dementia. We tested to see whether or not these trends differed from one another, and we did not find evidence to this effect (Table S1). The similar mortality trends would suggest that dementia prevalence will largely be shaped by incidence trends: if left to mortality alone, similar relative declines would create a stable population with no disproportionate increase or decrease in prevalence of either cognitive state (Schoen 1988). Therefore, this finding indicates that future prevalence is likely to be shaped by preceding declines in incidence.

In addition to documenting declining dementia prevalence and incidence, our study also sought to assess the role of education—a key component in dementia prevention widely discussed in the social science and medical literature (Cerasuolo et al., 2017; Langa, 2014). Education is thought to be a strong predictor of cognitive reserve and/or cognitive functioning (Lövdén et al., 2020; Stern, 2006, 2009). In the United States, education levels significantly increased (and will continue to increase for about another ten years) among recent cohorts of older Americans (Heckman & Lafontaine, 2007; Leggett et al., 2019). These improvements in education among older adults most likely led to better cognitive functioning in old age, either through higher peaks of cognitive function that are carried forward or slower declines (Lövdén et al., 2020). As such, it should not be surprising that improvements in education among older Americans have also led to age-standardized declines in dementia incidence and prevalence across this time period because improvements in cognitive functioning would either delay or prevent older adults from having dementia. Additionally, changes in education also may be mirrored in adult exposures and behaviors such as the job-related demands and improved treatment of other upstream medical conditions often discussed in the literature (James & Bennett, 2019; Langa, 2014) that reinforce the association between education and cognitive reserve or cognitive function. While the specific etiology of dementia is beyond the scope of our study, education has a broad impact on dementia development across multiple pathways, leading to a decline in prevalence of multiple subtypes. More research on this issue is crucial to understand the way in which social exposures affect dementia trends.

Additionally, we did not find strong evidence that improving educational attainment among older adults influenced the observed mortality trends. When education was added to the models, the trends changed in the expected directions (a steeper decline for people with dementia and a more muted decline for people without dementia) but it never reached statistical significance. The “lack of effect” for education on the trend suggests other factors or combination of factors explain declines in mortality for both groups. Future studies should examine other potential factors that may have influenced trends in mortality with and without dementia that is not related to education, such as improvements in dementia care, broad changes in CVD or cancer mortality, or changes in other health behaviors and social factors that are related to mortality, net of education.

The importance of improvements in education levels for older Americans for continued decline in dementia in prevalence, incidence, and mortality is unclear. At the low end of the education distribution, the expansion of educational attainment has slowed. Cohorts born between 1890 and 1935—the 65+ population in 2000—had high school completion rates that ranged from 20% to 50%. Cohorts born between 1932 and 1949—the 65+ population in 2016—had high school completion rates around 75% (Heckman & Lafontaine, 2010). High school completion rates have since remained relatively stable. At the same time, college completion rates have grown among all major demographic groups in the United States. How the expansion of the college-educated group will contribute to future dementia trends is unclear. Much will depend on college education’s association with cognitively demanding jobs, strong social relationships, and health behaviors and disease conditions that confer low dementia risk (or if it establishes even greater levels of cognitive functioning than high school alone). Ultimately, research introducing a cohort perspective will be crucial for understanding these processes.

We note three important limitations in our research. Our analysis is limited to older Americans from 2000 to 2016 who are predominantly white. Minority groups will compose a greater share of the older population in future generations. Research has shown that racial minorities are more likely to have dementia and live with it longer (Farina et al., 2019). Therefore, to understand changing national trends, future work should attend to how trends for racially/ethnically underrepresented groups are changing, and the role of education in reducing risk for these groups. Second, our measurement of education captures the number of years and not the quality of education, which may have an important role in cognitive reserve development. The HRS collects information on the number of years a person has attended school or the credentials obtained, which given the importance of education for cognitive reserve would imply greater cognitive capacity among older adults. Nonetheless, the role of quality has been also an open question for researchers (Manly et al., 1999, 2005; Mehta et al., 2009; Schaie, 2008). Our study cannot assess how quality of education along with more years of education impacts cognitive reserve in the population. While the number of years of education still provides a basic understanding of how cognitive reserve in the population has improved, future work that explores the impact of education on dementia risk would benefit from assessing quality and quantity simultaneously. And lastly, there is an open debate about practice effects and their potential impact on the analysis of dementia trends (Hale et al., 2020; Lee et al., forthcoming). Studies using the HRS data have found modest practice effects among well-educated individuals with improvements in some sections of the TICS exam (Alley et al., 2007). To explore this issue, we modeled prevalence trends by education group. We found flat trends within education groups, which provides evidence that practice effects are not driving the dementia trend: if practice effects had been a factor, we would expect to see a decline in dementia prevalence among adults with greater levels of education. While practice effects may pose challenges to assessing cognitive functioning or status for individuals, it is less evident that practice effects are impacting the observed downward trends.

CONCLUSION

Our study documented that the declining trends in dementia prevalence were accompanied by declining dementia incidence and mortality for people with and without dementia. It is reassuring that the age-standardized prevalence decline observed in the US older adult population from 2000 to 2016 was accompanied by a strong decline in incidence since this suggests that dementia prevalence will continue to decrease in the near future. The continued decline in dementia prevalence points to one of the few areas of improving population health in the United States today. Nonetheless, the observed decline documented here was primarily driven by improved educational composition of older adults – a trend that has slowed. Researchers and policy makers should consider the slowing of educational advancements among the older population to better predict future financial and medical costs of the population as well as caregiving needs, especially if a treatment for dementia remains elusive.