Population Surveillance of Dementia Mortality

Geographic and temporal variation in occurrence of dementia within the US has received little attention despite its importance for generation of new etiologic hypotheses and health services research. We examine methodological problems in the use of vital statistics data for assessing variation over time, among states and within states in the US. We analyzed the US multiple cause of death files for 2005–2006 and 1999–2000 US deaths with Alzheimer’s Disease (International Classification of Disease 10th revision code G30) and other dementias (codes F01, F02, R54) coded as underlying or contributing cause of death based on the death certificate. Age-adjusted death rates were computed by year, state or county for persons aged 65 years and over. In 2005–2006 combined, 555,904 total deaths occurred with any dementia type (212,386 for Alzheimer’s disease) coded as underlying or contributing cause. Among the states, age-adjusted rates per 100,000 per year varied by two fold ranging from 458 in New York to 921 in Oregon. Similar geographic patterns were seen for Alzheimer’s disease. However, between 1999–2000 and 2005–2006 the US death rate for all dementia increased only from 559 to 695 (24%) while that for Alzheimer’s disease doubled from 135 to 266. Use of specific (G30, F01) versus non-specific diagnoses (F02, R54) varied among states and over time, explaining most of the temporal increase in rate of Alzheimer’s disease. Further research is needed to assess artifacts of diagnosis, certification or coding, utilization of health services, versus biological variation as possible causes of temporal and geographic variation to enhance utility of mortality data for dementia monitoring and research.


Introduction
Already a major burden in affluent countries, few doubt that population aging trends will continue to increase the burden of dementia worldwide [1,2]. By 2002, Alzheimer's disease had become the sixth leading cause of death for persons 65 years and over in the US [3]. Many publications have detailed the geographic variation in stroke mortality in the US [4][5][6]. However, geographic variation in occurrence of dementia within the US has received little attention despite its potential importance for generation of new etiologic hypotheses and health services research. A few authors have reported international variation in dementia prevalence rates [7][8][9][10]. The hypotheses are that the rate of death with Alzheimer's Disease or other dementia varies among states, within states and over time in the US.

Experimental Section
Data were analyzed from the US multiple cause of death files for 2005-2006 and 1999-2000 [11]. Data from each death certificate were coded to yield a single underlying cause of death and up to twenty multiple causes as well as demographic data. These mortality data were examined for all deaths occurring in the fifty states and the District of Columbia. Deaths of foreign residents and of residents who died abroad were not included. Deaths with Alzheimer's Disease (International Classification of Disease 10th revision [ICD-10] codes G30.0, Alzheimer's disease with early onset; G30.1, Alzheimer's disease with late onset; G30.8, Other Alzheimer's disease; G30.9, Alzheimer's disease, unspecified) coded as underlying or a contributing cause of death were enumerated [12]. Also enumerated were vascular dementia (F01.0, Vascular dementia of acute onset; F01.1 Multi-infarct dementia; F01.2, Subcortical vascular dementia; F01.3, Mixed cortical and subcortical vascular dementia; F01.8, Other vascular dementia; F01.9, Vascular dementia, unspecified); Unspecified dementia (F03); and senility (R54). Death counts shown are totals for each two-year period (1999)(2000)(2005)(2006). Coding was completed by computer software and does not vary among states. Using the 2000 US standard population, age-adjusted death rates per 100,000 per year (average annual rate for each two-year period) were computed by state or county for persons aged 65 years and over; population estimates and standard methods are documented elsewhere [11]. Maps of death rates were created to show approximate quartiles of state or county rates using mapping software created by the Centers for Disease Control and Prevention [11]. Rates were considered unreliable when based on fewer than 20 deaths (hatched areas on maps). White areas labeled -other‖ represent counties with fewer than 6 deaths or population less than 100,000. -Background‖ refers to any areas excluded by the user.

Results and Discussion
Results. Among 555,904 with any dementia diagnosis in 2005-2006, 212,386 deaths occurred with Alzheimer's disease coded as underlying (142,518) or contributing cause (69,868). Among the states, age-adjusted rates per 100,000 for all dementia varied by two fold ranging from 458 in New York to 921 in Oregon (Table 1). Rates varied by over three fold for Alzheimer's disease ranging from 133 in New York to 419 in Washington (Supplementary Table 1). Figure 1 shows that high rates (>816 per 100,000) for dementia were seen in the Pacific Northwest, some mountain states, Northern New England, East-south central states and the Carolinas. Rates >300 per 100,000 for Alzheimer's disease were seen in the Pacific Northwest, Northern New England, East-south central states and the Dakotas (Supplementary Figure 1). Low rates were seen in Southern New England and the Mid-Atlantic, Hawaii, Nevada and Florida for both categories.      To assess the extent to which geographic patterns and trends in Alzheimer's disease rates might be attributed to variation in diagnostic practices, the numbers of deaths with different diagnostic codes were compared in the US and in states with highest and lowest Alzheimer's Disease rates (Table 2). Interstate differences were marked with respect to the proportion of deaths assigned to the specific diagnoses of Alzheimer's disease or vascular dementia versus non-specific diagnoses (dementia unspecified, senility); the former were used in 59% of cases in Washington but only 32% of cases in New York. The age-adjusted rate per 100,000 of death for all dementias was 559 in the US, 724 in Washington and 368 New York in 1999-2000 and 695, 840 and 458, respectively, in 2005-2006. Therefore, the difference between Washington and New York in rates of dementia is not merely due to use of difference diagnoses or ICD-10 codes. However, the increase in deaths with any dementia diagnosis in the US or these states between 1999-2000 and 2005-2006 was much less for dementia than that seen for Alzheimer's Disease, which was mostly due to increasing use of that specific diagnosis over time.
Discussion. This analysis of data from the US multiple cause files is the first to our knowledge to establish inter-and intra-state patterns of geographic variation in death with or Alzheimer's disease and other dementias. Strengths of this study include its provision of previously lacking data, its use of recent data, its complete coverage of the US population eliminating sampling errors. High rates of death with dementia were seen in the Pacific Northwest, Northern New England, and Southern Appalachia. Low rates were seen in Southern New England and the Mid-Atlantic, Hawaii and Florida. Between 1999Between -2000Between and 2005Between -2006 use and coding of the diagnosis of Alzheimer's disease increased so that the US death rate doubled for Alzheimer's disease but increased only 24% for all dementia. Mechanisms. Major geographic variation in mortality with dementia could be explained by variations in disease prevalence, a function of incidence and duration. This could result from more sophisticated or accessible care resulting in longer survival with the disease, higher incidence, or both in areas with higher versus lower rates. Cardiovascular risk factors such as hypertension have been hypothesized to be risk factors for Alzheimer's disease and other dementias [12]. However, geographic patterns of stroke mortality were different from those for dementia, with relatively low rates in the Pacific Northwest and northern New England and high rates in the Atlantic Coastal Plain [4]. However, east, and south-central states were high for both. Further, an analysis of trends in stroke mortality predicted that Oregon and Washington, Pacific states, would rise greatly in relative ranking consistent with dementia patterns in 2005-2006 [13]. Like dementia, coronary heart disease mortality was reported to be high in the Ohio River Valley and Carolinas, but, unlike dementia, low in the Pacific Northwest and high in New York [5]. If healthy elderly persons moved from northern New England and Ohio to south and central Florida this could produce higher death rates in the former than in the latter [14,15].
Limitations of this study mean that statistical artifact due to patterns of diagnostic practice and completion of death certificates cannot be excluded as a contributor to geographic and temporal variation in mortality with all dementia [15]. Diagnostic inaccuracy may result in under-ascertainment of specific conditions at death despite introduction of codes for them with ICD-10 [16,17]. Multiple cause data were used to reduce tendency of underlying cause data to underestimate to role of chronic conditions such as dementia on mortality due to immediate causes such as pneumonia and cardiovascular disease. Yet, diagnostic practice may be driven by the availability of specialist care and diagnostic modalities such as magnetic resonance imaging and computerized tomography as well as patients' access to all these. Lack of personnel and facilities or poor access may tend to increase the use of non-specific diagnoses such as unspecified dementia and senility as well as failure to diagnose dementia before death. Fewer than 10% of deaths from natural causes were autopsied [18]. For deaths occurring outside an institution, the death certificate might be completed by a coroner without physician input. The increasing use of specific diagnoses nationwide seen in this study likely reflects the spread of awareness among physicians and the public and increasing utilization of diagnostic technology among the elderly [2]. Location of dementia research programs might increase awareness in some areas. However, of 30 Alzheimer's Disease Centers funded by the National Institutes of Health at major medical centers, only 4 were located in the 10 states with the highest Alzheimer's disease death rates while 7 were located in the 10 states with the lowest rates [19]. California with 6 centers ranked 17th. The current study could not address failure by certifiers to mention chronic dementia at all. When death rates are computed, inaccuracies of population counts and estimates can also limit the utility of data. However, every attempt is made by the U.S. Bureau of Census to understand and compensate for known problems. For example, to accommodate geographic shifts of the Alabama, Louisiana, Mississippi, and Texas populations resulting from Hurricanes Katrina and Rita in 2005, the U.S. Census Bureau developed adjustments in the methodology for state and county population estimates [11]. Aging of the population may not totally be accounted for by age-adjustment, e.g., in trend analysis, since the top bracket was 85 years and over. If elderly persons move from their customary state of residence after dementia diagnosis, their residence at death may not reflect long-term environmental exposure and hence hinder hypothesis generation [20]. Further research addressing these methodological issues would enhance the feasibility of ecological studies of Alzheimer's disease with respect to known or suspected risk-modifying factors (e.g., hypertension) has been done in the past for cancer [21][22][23].

Conclusions
Marked geographic variation in rates of death with all dementia and Alzheimer's disease occurred among US states and counties and over time. Further research is needed to assess variation in artifacts of certification versus variation in disease incidence or duration as possible causes of geographic and temporal variation. Data from multicenter studies of dementia in population-based samples of the US elderly population with validation of death-certificate diagnoses and estimation of incidence, survival and prevalence are needed to establish the magnitude and causes of geographic variation.