The incidence of dementia in Canada

The sample.

In 1991, representative samples of people aged 65 or over were drawn from 36 urban and surrounding rural areas in the 10 Canadian provinces.28 The study involved 9,008 people from the community and 1,255 from institutions. The community sampling frame used the computerized records of the provincial universal health insurance plans, except in the province of Ontario where an aggregated list based on election and other municipal records was used. Age-stratified random samples were drawn in each sampling area, following an optimum allocation procedure.28 The institutional sampling frame included nursing homes and chronic care facilities. Institutions were stratified by size and we drew random samples of people aged 65 or over from them.

For CSHA-2, survivors who had been diagnosed with dementia at baseline were reassessed to study the progression of dementia; they are not further discussed here. The incidence sample was based on the remainder who did not have dementia at CSHA-1; the derivation of the sample is illustrated in figure 1. Five hundred eight subjects who screened positive for cognitive impairment at CSHA-1 but who did not attend the clinical examination had to be omitted because we cannot exclude the possibility that they had dementia at CSHA-1. For the incidence analysis we classified people into the community or institutional samples based on CSHA-1 status, even though community participants may subsequently have moved into an institution. We used this approach because we could not be certain of the chronologic sequence of institutional admission and the development of dementia.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1. The Canadian Study of Health and Aging (CSHA) incidence study sample.

Data collection.

With minor modifications, the same approach was used in both waves of the study. For those living in the community, a home interview included the Modified Mini-Mental State Examination (3MS) screening test for cognitive impairment.38,39 All who screened positive (scores <78) were asked to attend a clinical examination. In addition, random samples of those who screened negative were examined clinically at both waves of the study (494 at CSHA-1 and 489 at CSHA-2). This permitted estimation of the false-negative rate of the screening test. Those who were in an institution at both waves of the study proceeded directly to the clinical examination. The home interview also covered a range of other topics, including scales for the presence of chronic health conditions (range, 0 to 12) and activities of daily living (ADL) (range, 0 to 14).40

In the clinical examination a registered nurse administered the 3MS and assessed vital signs, hearing, vision, height, and weight; the nurse also collected information on medications and obtained the subject’s medical and family history from a relative using the Cambridge Examination for Mental Disorders of the Elderly (CAMDEX) interview.41 Next, a physician carried out a physical and neurologic examination. Finally, a psychometrist administered 11 neuropsychological tests to all consenting subjects deemed testable (defined as a 3MS score ≥50).42 Test scores were interpreted by a neuropsychologist. The physician and neuropsychologist independently formed preliminary diagnoses using criteria from the Diagnostic and Statistical Manual of Mental Disorders, 3rd ed., revised (DSM-III-R).43,44 These diagnoses were compared at a consensus conference involving the whole clinical team, and those diagnosed with dementia were subclassified using the National Institute of Neurologic and Communicative Disorders and Stroke and Alzheimer’s Disease and Related Disorders Association (NINCDS-ADRDA) criteria for AD;45 International Classification of Diseases, 10th revision (ICD-10) criteria46 were used to diagnose vascular dementia. The consensus diagnoses formed the basis for the incidence figures reported here; the incidence figures for AD were based on “probable” cases only.

For participants who died before the follow-up study, the date and cause of death were obtained from the Provincial Registrar of Vital Statistics, and a relative or other informant was interviewed to assess the subject’s physical and cognitive status 3 months prior to death. The interview included ADL questions and a selection of indicators of cognitive status, taken from the CAMDEX cognitive history interview,41 that had been found predictive of dementia in analyses of the CSHA-1 data set.

Data quality.

All questionnaires were checked for accuracy and completeness; double data entry was used until checks revealed fewer than one error per 1,000 variables entered. Following this, extensive computerized logical and consistency checks were applied within and across the various components of the data set (screening questionnaire, clinical examination, decedent interview), and against the baseline information. The reliability of the CSHA-2 clinical diagnosis was evaluated by assessing agreement between the consensus diagnosis and a second physician’s independent diagnosis based on study records alone, for 201 randomly selected patients. The results showed a κ coefficient of 0.72 (95% CI, 0.61 to 0.84) for distinction between dementia and nondementia, indicating substantial agreement.47

Estimating incidence.

Incidence rates were calculated for Canada (except for Newfoundland) and were age-standardized to the nondemented Canadian population. We were unable to include incidence data from Newfoundland because a 1996 legal interpretation of the province’s advance directives legislation prohibited the use of proxy consent for persons unable to give full, informed consent for research. Subjects who did not participate in CSHA-2 and those who screened positive at CSHA-2 but were not clinically examined were also excluded from the calculations.

As most dementias are irreversible, we used the estimated Canadian population without dementia in 1991 as the denominator, rather than the total population. However, projections onto the total population also permit comparison with other studies that have used this approach. Standard actuarial methods were used to estimate person-years at risk. The date of onset of dementia was assumed to be midway between the date of the baseline assessment and the date of death or of the follow-up assessment.

Although the 5-year delay between baseline and follow-up allowed time for incident cases to accumulate, it also meant that mortality had to be formally considered. Some incidence studies have assumed that those who died had the same chance of having dementia as survivors,6,11 but as mortality rates are higher among those with dementia,48 this approach is likely to underestimate the true incidence. We used three sources of information to estimate the dementia status of decedents:

1.

Death certificates were examined to identify those citing dementia as the underlying or a contributing cause. Sensitivity was expected to be low because death certificates under-report dementia,49-53 and this was confirmed in our study through analysis of 71 people who died within 5 months of undergoing the clinical examination. Sensitivity was 33%, although specificity was 96%.

2.

The interview with a decedent’s informant included the question: “Was [the deceased] ever diagnosed with memory problems, for example Alzheimer’s disease or senile dementia? If Yes, specify _____.” A positive answer was considered indicative of dementia. For the 71 subjects, sensitivity was 44% and specificity was 96%.

3.

Again using the same 71 subjects, a logistic regression algorithm was developed to estimate the probability that decedents had had dementia prior to death. The algorithm used baseline data and responses from the CSHA-2 interview with the decedent’s informant. The variables that significantly predicted dementia status are shown in Appendix 2. The model was then validated by applying it to 2,431 live subjects who received the CSHA-2 clinical examination. Using a conservatively high threshold probability value of 0.95 as indicating dementia, the estimated number of cases fell 7.7% below the true number in the validation sample. Using the probability threshold of 0.95, the algorithm was applied to the decedents.

Decedents who met one or more of these criteria were classified as having had dementia for the incidence analyses; a threshold value of 0.95 was used for the algorithm. This approach did not allow us to classify dementia by type, so we could not include decedents in the analyses that present incidence figures for AD (table 1).

Participants.

The numbers participating are summarized in figure 1. Of 9,008 people in the original community sample, the 402 who had dementia at baseline and the 508 who screened positive but were not clinically assessed were excluded. Similarly, 730 of the 1,255 people in the institutional sample were diagnosed with dementia at CSHA-1 and were excluded. As explained above, the 395 participants from Newfoundland had to be excluded from the follow-up study. Of the eligible subjects, 1,846 died before CSHA-2; further details are given below.

Of the 6,152 live community subjects, 5,432 (88.3%) participated in the CSHA-2 screening interview. Of these, 1,134 screened positive or could not undergo the screening interview; 987 (87%) consented to the clinical examination. Of the 230 live institution subjects, 210 (91.3%) consented to the clinical examination.

There were 1,846 deaths among eligible subjects in the incidence cohort. No source of information was available for 217 (11.8%), who were omitted from further analysis; 104 death certificates mentioned dementia and 118 informant interviews reported dementia. We had data to apply the predictive algorithm for 1,629 (88.2%); this classified 410 as having dementia. A total of 468 decedents (28.7%) were classified as having had dementia from one or more of these sources of information. These have been included in the incidence results that consider all forms of dementia.

To guide inferences about the possible biases that the losses to follow-up may have produced in our incidence estimates, table 2 compares demographic and health status variables for participants and nonparticipants. Those who died were the oldest, included more men, and had reported more chronic health complaints at baseline; those who refused to participate at all in CSHA-2 were comparable to participants in terms of age, education, chronic health complaints, and disability scores at baseline. Those who screened positive but refused the clinical examination were older and had lower baseline 3MS and health scores.

Incidence rates.

Table 3 indicates the numbers of incident cases of dementia in the sample by gender, age-group, and residence (community versus institution), and shows incidence rates per 1,000 in the population without dementia. It should be recognized that these are not weighted population estimates, and that the community sample includes some people who were admitted to an institution between the two waves of the study. Further analyses considered the influence of cognitive status at baseline. In the community sample, the dementia incidence rates for both genders were five times higher for those who screened positive at CSHA-1 and were diagnosed as cognitively impaired (but not demented) compared with those who screened normal. In the institutional sample, the equivalent incidence rates were three times higher for women and two times higher for men.

Table 4 presents weighted estimates of incidence of dementia for the Canadian population by age-group and gender; community and institutions are combined. Rates are shown for two denominators; the rates per 1,000 nondemented give the most appropriate age-specific incidence estimates, but the rates per 1,000 total population are also shown, as these have been used in some other studies and can be more easily used for planning where information is available only on the total population. The figures for both genders combined show roughly a doubling of incidence for each 5-year increase in age. Men in their 70s showed comparatively high incidence rates.

Table 1 shows the number of cases and the age-specific incidence rates for AD. The figures are based on survivors only, as we were unable to form a differential diagnosis of AD for those who had died. Due to the small numbers of cases, community and institutional samples have been combined. The nondemented population is used as the reference.

The incidence rates suggest that there will be approximately 60,150 new cases of dementia in Canada each year. Of these, 36,320 will be women and 23,830 men; 46,670 will arise in the community and 13,480 in institutions (more details are available on the Neurology Web site at www.neurology.org). We previously estimated the prevalence of dementia in Canada to be 252,60028; from this, the estimate of 60,150 new cases per year suggests that the average duration of dementia is 5.8 years (6.7 years for women and 4.6 years for men).

Comparison with other studies.

In many ways our findings confirm prior expectations. The logarithm of our incidence rates for all dementias is reasonably linear across age until the group aged 90 years or more, at which it shows a slowing of the increase with age. Unfortunately, the Canadian census does not provide a finer age break-down for those aged over 90 in institutions, so we cannot weight our results appropriately according to community or institutional samples. As in most studies, we found the overall incidence rate to double with approximately every 5 years of age.27 The age gradient is greater for community than for institutional subjects in both women and men. As with the Bordeaux,15 Lundby,2 and Rotterdam13 studies (but in contrast to the Baltimore data17), our incidence figures for AD on a logarithmic plot were convex upward, such that the increase in rates declines from about 80 years of age onward.

The CSHA incidence figures are comparable to those of other, smaller studies for AD, but lie toward the upper end of the spectrum for all dementias combined (figures 2 and 3⇓). This is arguably because our estimate of AD incidence based on survivors is low, whereas our figures for all dementias include decedents. The comparatively wide spread of estimates between studies at higher ages has been discussed by Jorm and Jolley26 and Gao et al.27 and may reflect instability in the estimates due to small sample sizes among people aged 80 or over. This is, however, more apparent than real; on a logarithmic incidence scale there is a constant spread across age between the studies.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2. Age-specific incidence of dementia per 1,000 persons in selected studies. CSHA = Canadian Study of Health and Aging.

• Download figure
• Open in new tab
• Download powerpoint

Figure 3. Age-specific incidence of AD per 1,000 persons in selected studies. CSHA = Canadian Study of Health and Aging.

The meta-analysis by Gao et al.27 showed no difference in incidence by gender for all dementias combined, although women had an OR of 1.56 for AD. Similarly, the differences in incidence by gender in our study were not large, and using Poisson regression analysis the ratio of the two rates (men: women) was 1.19 (95% CI, 1.00 to 1.41) for community and 0.91 (0.59 to 1.40) for institutional subjects. Our comparatively high incidence rate for men in their 70s may be due to a higher incidence of vascular dementia in this group, and a relatively higher prevalence of cognitive impairment at baseline. As would be expected, we found greatly increased incidence rates for people with previously diagnosed cognitive impairment; our figure of a fivefold greater incidence in the community sample is probably conservative because the comparison group comprised those who screened negative for dementia; some of these individuals might well have been diagnosed as cognitively impaired had they received the clinical examination.

The threshold for defining a case may have an impact on incidence estimates, although it would have a greater impact on prevalence. It may, however, account for the studies with the lowest incidence rates in figure 2: those from Rochester and Finland. These reports were based on hospital populations or registries, which typically include more advanced cases and show lower incidence estimates;20,58,59 this was illustrated graphically by Bickel and Cooper.56

This Canadian study adds to the growing international effort to document the incidence of dementia. Several of the estimates agree closely (at least for people up to the age of 80), such that a consensus seems to exist. Ways to overcome methodologic issues that may have led to the variation in estimates for the very old, such as the question of which denominator to use, have been proposed here. Some uncertainty remains, however, in substantive areas due to variations in results between studies: the relative incidence in the two genders, and the balance between subtypes of dementia are examples of issues that remain to be resolved.