Background
The largest single risk factor for developing dementia is aging. As longevity increases, more people are diagnosed with dementia. In Canada, 24.8% of the population aged ≥65 years lives with dementia or mild cognitive impairment,Reference Graham, Rockwood and Beattie 1 with annual estimated costs of $8.8 billion.Reference Wimo, Winblad and Jonsson 2 Accurate determination of life expectancy after dementia diagnosis is important for patients, families, and communities for planning services, estimating costs of forthcoming care, and community support needs.
Previous studies indicate that survival rates vary considerably. Although the previous commonly cited survival estimate in dementia are 8-12 years,Reference Stubendorff, Hansson, Minthon and Londos 3 more recent prospective studies using published diagnostic criteria estimate median survival after dementia diagnosis from 3.2Reference Jagger, Clarke and Stone 4 to 8.3Reference Brookmeyer, Corrada, Curriero and Kawas 5 years. These differences may be attributed to study design and population, but individual patient factors also influence survival.
Canadian survival studies following dementia diagnosis are limited.Reference Wolfson, Wolfson and Asgharian 6 - Reference Hogan, Thierer, Ebly and Parhad 8 The studies are older, with shorter follow-up, and without systematic investigation of potentially important contributing risk factors. The Canadian Cohort Study of Cognitive Impairment and Related Dementias (ACCORD) was an eight-center study across Canada that included subjects referred to dementia specialty clinics for assessment of cognitive impairment. All subjects received comprehensive neurological, cognitive, and functional ability assessments before being classified as being “cognitively normal,” “cognitively impaired but not demented” (CIND), or “dementia.”Reference Feldman, Levy and Hsiung 9
The present study evaluates the long-term survivorship of the inception cohort referred to the lead and coordinating ACCORD site, at the University of British Columbia Hospital’s Clinic for Alzheimer Disease and Related Disorders (CARD). Clinic for Alzheimer Disease and Related Disorders is a tertiary dementia referral clinic for the province of British Columbia (BC). This study aimed to evaluate the effects of demographics, genetic and vascular risk factors, dementia treatment, and comorbid medical illness on survival, and to compare survival of this cohort with a reference BC population.
Methods
Subjects
Canadian Cohort Study of Cognitive Impairment and Related Dementias enrolled an inception cohort of 1136 subjects from eight centers across Canada between 1997 and 1999. All participants were enrolled and consented consecutively, after referral by family physicians or other specialists. Detailed description of the methods have been previously published.Reference Feldman, Levy and Hsiung 9 There were 229 newly referred patients enrolled through CARD, 169 of whom were initially diagnosed with dementia. For the current study, we undertook a survival analysis of this inception dementia cohort in 2015. One subject was removed because of significant missing baseline data, allowing an evaluable sample of 168 subjects.
Diagnosis
Diagnostic classification was made by CARD site investigators as cognitively normal, CIND, or dementia according to DSM-IIIR criteriaReference Fratiglioni, Grut, Forsell, Viitanen and Winblad 10 after evaluation of all clinical, laboratory, and neuroimaging data. Dementia etiology was ascertained in the following manner: Alzheimer’s disease (AD) by NINCDS-ADRA criteria;Reference McKhann, Drachman, Folstein, Katzman, Price and Stadlan 11 vascular dementia (VaD) was diagnosed on the basis of history of stroke, neurological signs consistent with cerebrovascular accidents, history of vascular risk factors or neuroimaging evidence of ischemic changes and Hachinski Ischaemic Score (HIS) score; frontotemporal dementia (FTD) was diagnosed on the basis of Lund and Manchester group criteria; 12 and dementia with Lewy bodies (DLB) using the criteria by McKeith et al.Reference McKeith, Galasko and Kosaka 13 All dementia diagnoses besides AD were grouped as “Other Dementias” (OD) for this analysis owing to small numbers.
Baseline demographic information included date of birth; sex; education; living status (alone or with others); and cholinesterase inhibitor (AchEI) use. Vascular risk factors, including diabetes mellitus, hypertension, hypercholesterolemia, atherosclerosis, and smoking, were assessed by nurse clinical interview or the presence of those risk factors through evaluation of medical records and current medications. These risk factors were recorded as dichotomous variables (yes/no). For a subset of participants (43%) within a separate consent, apolipoprotein 4 (ApoE-4) carrier status was analyzed.
Baseline scores of standardized assessments were also collected: Mini-Mental Status Examination (MMSE)Reference Folstein, Folstein and McHugh 14 screens global cognition, with lower scores indicating more severe impairment (range 0-30); Cumulative Illness Rating Scale (CIRS)Reference Conwell, Forbes and Cox 15 rates impairment of 14 organ systems (including neurological, where the question was changed to “brain and spinal cord, excluding dementia” in order not to confound results given our cognitively impaired cohort), with higher scores indicating more severe comorbid medical disease (range 0-56); HISReference Hachinski, Iliff and Zilhka 16 identifies underlying vascular components for dementia (range 0-18), with scores >7 indicating significant cerebrovascular contribution; Disability Assessment for Dementia (DAD)Reference Gélinas, Gauthier, McIntyre and Gauthier 17 rates everyday functioning including activities of daily living, with lower percentages indicating greater disability (range 0%-100%); Functional Rating Scale (FRS),Reference Crockett, Tuokko, Koch and Parks 18 a derivative of the Clinical Dementia Rating Scale,Reference Berg 19 provides global staging of dementia severity, where higher scores indicate greater cognitive and functional impairment (range 8-32); and Neuropsychiatric Inventory (NPI)Reference Cummings, Mega, Gray, Rosenberg-Thompson, Carusi and Gornbein 20 assesses frequency and severity of neuropsychiatric disturbances (range 0-144), with higher scores indicating greater burden.
Determination of Vital Status
We obtained vital status for every subject on June 20, 2015 through the BC Medical Service Plan records, which has a record of all citizens in BC. Death dates and moves from BC are registered at the first of the new month after which a person died or moved. The 14th day of the previous month was used as censor date. Survival was calculated from the date of baseline assessment to censor date. Six participants left BC during follow-up time and were counted as censored at the time of their move.
Statistical Methods
Baseline characteristics were compared between diagnostic groups using independent t-test for continuous variables, using Levene’s test to verify equality of variances in the samples. We used Pearson’s χ 2 to compare categorical data. Significance was defined as two-tailed p<0.05.
We performed separate unadjusted Kaplan-Meier curves and log-rank tests to determine survival from diagnosis for AD and OD groups for each variable. Continuous variables were stratified for these analyses. For age, we initially looked at 5-year age strata and found significant difference between groups of those under and over 80, but not for any other age strata. We then used ≥80 as a binary cut-off point to maximize statistical power to examine for group differences. For the MMSE and FRS, which have published cut-offs that are indicative of clinical differences,Reference Tombaugh and McIntyre 21 , Reference Feldman, Schulzer, Wang, Tuokko and Beattie 22 those scores were used as strata. For the DAD, NPI, CIRS, and Hachinski scores, which do not have validated cut-off points marking severity, we initially divided the sample into quartiles. Given our small sample, quartiles without significant difference were grouped together to form a dichotomized variable to evaluate severity. Stratified groups were tested for proportional hazards, but assumptions were not met with diagnosis (AD vs. OD), and thus time-dependent Cox regression was used. Survival by 5-year age group was compared with BC population life expectancies for years 1990-1999, 23 reflecting the period of recruitment.
All analyses were performed using SPSS version 23.0 (IBM Corp., Armonk, NY, USA).
Results
Cohort Characteristics
Of the 168 participants diagnosed with dementia, 135 had a primary diagnosis of AD; 33 had a primary diagnosis that was non-AD, including 8 VaD, 9 DLB, and 12 FTD; and 4 were diagnosed with another dementia. As the numbers in each subtype of dementia were small, they were grouped as “OD” for the purpose of comparison. Table 1 lists the characteristics of the cohort. Mean age at diagnosis was 78.24 (SD 8.76). The groups differed significantly from each other in that the AD group was significantly older at the time of diagnosis, more likely to be female, and had lower MMSE scores than the OD group; the OD group in turn had significantly higher total CIRS, NPI, and Hachinski scores and lower DAD scores than the AD group. When comparing all dementia types by age strata <80 and ≥80, we found no significant difference between groups, except that those ≥80 were more likely to live alone (40% vs. 17%, p=0.004) and there was a trend toward lower DAD scores (59 vs. 68, p=0.052); data not shown.
Baseline characteristics of cohort
AchEI=cholinesterase inhibitor; AD=Alzheimer’s disease; ApoE=apolipoprotein E; CIRS=Cumulative Illness Rating Scale; DAD=Disability Assessment for Dementia; FRS=Functional Rating Scale; MMSE=Mini-Mental Status Exam; NPI=Neuropsychiatric Inventory; OD=other dementias.
Percentages for dichotomous data are valid percent accounting for missing data.
Bolded text indicates significant values.
Cognitive Diagnosis, Mortality, and Survival
In total, 158/168 (94.0%) subjects died during the mean follow-up time of 16.61 years (SD 0.82): 128/135 (94.8%) of AD, and 30/33 (90.9%) of OD. Unadjusted median survival was 7.08 years (SE 0.41) for all-cause dementia. There was no significant difference in the unadjusted median survivorship between AD (7.33 years, SE 0.25) and OD (4.33 years, SE 1.44, log-rank 1.72, p=0.190).
We found that older age at diagnosis, higher CIRS, and lower DAD scores were significantly associated with shorter survival (Table 2; Figures 1A-1D). There was no interaction between DAD and CIRS scores. There was a trend toward significance between smoking (p=0.070) and shorter survival. There was no significant association between other variables including AchEI use and survival. Given the relatively small sample size, only whole-sample significant factors were included in the subsequent regression model.
(A-D) Kaplan-Meier survival curves. AD=Alzheimer’s disease; CIRS=Cumulative Illness Rating Scale; DAD=Disability Assessment for Dementia; OD=other dementias.
Survival table by characteristics for Alzheimer’s disease (AD) and other dementia groups
AchEI=cholinesterase inhibitor; ApoE=apolipoprotein E; CIRS=Cumulative Illness Rating Scale; DAD=Disability Assessment for Dementia; FRS=Functional Rating Scale; MMSE=Mini-Mental Status Exam; NPI=Neuropsychiatric Inventory; OD=other dementias.
* Remain significant after time-dependent Cox regression adjustment for diagnosis, age, CIRS, and DAD.
Bolded text indicates significant values.
A model was built using time-dependent Cox regression adjusting for diagnosis (AD, OD), age at diagnosis (<80, ≥80), CIRS (<7, ≥7), and DAD scores (<55%, ≥55%). Shorter survival was associated with age ≥80 years (hazard ratio [HR] 1.56, 95% confidence interval [CI] 1.05-2.32); DAD scores <55% (HR 1.47, 95% CI 1.04-2.08); and CIRS scores ≥7 (HR 1.51, 95% CI 1.08-2.12). There was no difference in survival by group (AD vs. OD) in this model.
Median survival times by 5-year strata were compared with population life expectancies in BC 23 (Table 3). Survival decreased with each stratum, but years of potential life lost after a dementia diagnosis was particularly compromised for younger subjects: 15.36 years lost for those diagnosed <65 years, tapering to 1.82 years for those ≥80 years.
Comparison of median survival in Alzheimer’s disease and other dementias with life expectancy in British Columbia
* Data from Human Mortality Database.
** Data from age group 60-64.
*** Life expectancy averaged from age groups 80-84 (8.58 years), 85-90 (6.20 years), and 90-94 (4.41 years).
Discussion
In the current study, we report on the survival analysis of a well-characterized cohort referred to a tertiary-care dementia clinic in the province of BC, Canada. The median survival from diagnosis for all-cause dementia was 7.08 years. Important predictors of shorter survival within the dementia groups included older age, lower DAD, and higher CIRS scores. Comparing the median dementia survival with BC general population life expectancies, there was a reduction in survivorship across all age groups, with strongest effect on younger subjects.
Although survival time was reduced in subjects ≥80 years with dementia, the effect of reduced survival was strongest for those <80 years, consistent with other studies.Reference Brookmeyer, Corrada, Curriero and Kawas 5 , Reference Ostbye, Hill and Steenhuis 7 , Reference Brodaty, Seeher and Gibson 24 - Reference Rizzuto, Bellocco, Kivipelto, Clerici, Wimo and Fratiglioni 27 Age at diagnosis (≥80) was a significant risk factor for survival (HR=1.56), a finding that is consistent with mostReference Jagger, Clarke and Stone 4 , Reference Ganguli, Dodge, Shen, Pandav and DeKosky 28 - Reference Xie, Brayne and Matthews 33 but not allReference Mölsä, Marttila and Rinne 34 - Reference Stern, Tang and Albert 36 other studies.
Our cohort had relatively long survivorships after diagnosis—roughly 7 years compared with published median survival times of 3-8 years—but direct comparison with other studies is difficult. Differences can be at attributed to community sample;Reference Jagger, Clarke and Stone 4 , Reference Wolfson, Wolfson and Asgharian 6 , Reference Ganguli, Dodge, Shen, Pandav and DeKosky 28 , Reference Mölsä, Marttila and Rinne 34 , Reference Aevarsson, Svanborg and Skoog 37 - Reference Scarmeas, Luchsinger, Mayeux and Stern 39 older population age;Reference Jagger, Clarke and Stone 4 , Reference Wolfson, Wolfson and Asgharian 6 , Reference Ganguli, Dodge, Shen, Pandav and DeKosky 28 , Reference Aevarsson, Svanborg and Skoog 37 , Reference Scarmeas, Luchsinger, Mayeux and Stern 39 - Reference Rascovsky, Salmon and Lipton 41 shorter follow-up times;Reference Jagger, Clarke and Stone 4 , Reference Wolfson, Wolfson and Asgharian 6 , Reference Aevarsson, Svanborg and Skoog 37 - Reference Rait, Walters, Bottomley, Petersen, Iliffe and Nazareth 40 , Reference Mehta, Yaffe and Pérez-Stable 42 different methods of ascertainment; and definitions of diagnoses.Reference Jagger, Clarke and Stone 4 , Reference Mölsä, Marttila and Rinne 34 , Reference Mölsä, Marttila and Rinne 38 , Reference Rait, Walters, Bottomley, Petersen, Iliffe and Nazareth 40 , Reference Rascovsky, Salmon and Lipton 41 Furthermore, our findings may be influenced by a secular trend where overall life expectancy 43 is increasing in recent years. Of particular note, BC has the longest life expectancy in Canada and one of the highest in the world.Reference Zhang and Rasali 44
A cumulative medical comorbidity index (CIRS) was a strong predictor of mortality (HR=1.51). The CIRS, a composite measure of medical comorbidity severity, is sensitive to the effects of age, education, and cognitive impairment,Reference Borson, Scanlan, Lessig and DeMers 45 factors that affect performance on scales such as these. Not surprisingly, the degree of functional disability associated with dementia, measured by the DAD, was also found to be a significant and independent predictor of survival (HR=1.47). This finding recognizes that functional disability better predicts survival than cognitive impairment on screening cognitive measures such as the MMSE.Reference Todd, Barr, Roberts and Passmore 46 , Reference Rountree, Chan, Pavlik, Darby and Doody 47 Our study suggests that a global health assessment such as the CIRS and a functional disability rating such as the DAD are useful adjuncts to predicting survival with dementia.
The OD group had substantially shorter median survival compared with the AD group (4.33 years vs. 7.33 years). This difference did not reach significance, possibly due to the small sample numbers in the OD group. Consistent with other studies,Reference Larson, Shadlen and Wang 25 , Reference Rountree, Chan, Pavlik, Darby and Doody 47 individual vascular risk factors did not predict survival in dementia. However, studies are needed that measure the presence and degree of control of vascular risk factors during midlife as differentiated from the presence or absence at the time of dementia assessment. This inception cohort was unique for its enrollment when AchEIs were first becoming available in Canada for use in AD, but not covered by government health insurance in BC. Only about half of the participants were taking AchEIs; no survival difference was found (p=0.490). There have been conflicting reports about the effects of AchEIs on survival, with some reporting no difference,Reference Rountree, Chan, Pavlik, Darby and Doody 47 , Reference Lopez, Becker and Wahed 48 and others reporting a survival benefit,Reference Ott and Lapane 49 , Reference Gasper, Ott and Lapane 50 but studies showing prolonged survival were retrospective within nursing-home populations. Although consistent with a number of other studies,Reference Scarmeas, Luchsinger, Mayeux and Stern 39 , Reference Wattmo, Londos and Minthon 51 , Reference Helzner, Scarmeas, Cosentino, Tang, Schupf and Stern 52 we found no significant survival difference overall with ApoE-4 (p=0.104), similar to a previous study.Reference Dal Forno, Carson, Brookmeyer, Troncoso, Kawas and Brandt 53 Apolipoprotein 4’s lack of significant survival overall effect may indicate that ApoE-4 has a progression effect earlier in the disease course. It is noteworthy that prevalence of ApoE-4 carriers was higher in the OD group compared with the AD group. We believe that there may be two elements at play to explain this: selection bias of OD subjects willing to undergo genetic testing, which may be driven by family history, with only 10 of the 33 in the OD group consenting to genetic testing. The second element is that, in this study, we used primary clinical diagnoses, reflecting usual clinical paradigms. A previous study using this British Columbia ACCORD cohortReference Woodward, Mackenzie, Hsiung, Jacova and Feldman 54 has shown that that out of the 45 cases that went to autopsy, 47% had mixed pathologies. Of these, 86% of them were mixed Alzheimer’s pathology, which may also account for the higher prevalence of ApoE-4 carriers among the OD group. Further studies with larger samples using integrated clinical, genetic, and pathological data would be helpful in clarifying this issue. Sex is often cited as a significant predictor of survival in dementia,Reference Wolfson, Wolfson and Asgharian 6 , Reference Ganguli, Dodge, Shen, Pandav and DeKosky 28 , Reference Agüero-Torres, Fratiglioni, Guo, Viitanen and Winblad 30 , Reference Xie, Brayne and Matthews 33 which we did not find to be the case, along with a number of others studies.Reference Brookmeyer, Corrada, Curriero and Kawas 5 , Reference Agüero-Torres, Fratiglioni, Guo, Viitanen and Winblad 29 , Reference Guhne, Matschinger, Angermeyer and Riedel-Heller 35 , Reference Helzner, Scarmeas, Cosentino, Tang, Schupf and Stern 52
The strengths of this study include the complete survivorship data on a cohort of consecutively referred individuals 16 years after initial recruitment. As a provincial tertiary care center, the CARD receives a broad range of referrals, and multiple etiologies of cognitive change are encountered. All diagnoses were made according to a uniform set of research criteria and reviewed for inconsistencies within the diagnostic algorithm. We have previously reported CARD’s clinical-pathological diagnostic accuracy,Reference Woodward, Mackenzie, Hsiung, Jacova and Feldman 54 with very high accuracy for primary pathologies, and lower accuracy for multiple mixed and secondary etiologies, an effect that was not evaluated within this current study. Further strengths of the current study include our ability to include a large number of potential risk factors and long follow-up time in our survival analyses.
There are several potentially important limitations. As a referred sample, there may be selection bias, which may result in either underestimation or overestimation of survival times. Survival may be underestimated, as referred individuals may be farther along in disease course. Length-timeReference Wolfson, Wolfson and Asgharian 6 and referral bias may contribute to overestimating survival. This study design did not allow for the recruitment of a cognitively normal control group, and within the study we had only a small group of participants who were classified as being not cognitively impaired. Another constraint is our relatively small sample size, which limits our ability to address differential survival across CIND, OD, and a more fine-grained assessment of dementia of mixed etiologies. To date, the other study centers of the ACCORD have had high loss to follow-up, and therefore are not able to undertake a similar survival analysis. Comparisons with BC population data is of interest, but must be interpreted with caution: by nature, these data do not represent a recruited sample and include people with dementia, although these factors would serve to strengthen the survival differences found.
Conclusion
A diagnosis of dementia confers a shorter survival than population life expectancy in BC in any age stratum, with greatest impact on younger patients. Age at diagnosis, higher number of medical comorbidities, and greater disability predict shortened survival better than cognitive impairment. Instruments such as CIRS and DAD may be more valuable indicators for survival in dementia than the MMSE, and would be useful in mapping the patient journey and shaping family and societal expectations.
Acknowledgments
The authors thank all of the subjects who participated in the study, and the study staff at the CARD, as well as all of the investigators and staff members of the ACCORD.
Funding
The ACCORD study was funded by the Medical Research Council of Canada (now CIHR) and the Pharmaceutical Manufacturers Association of Canada, MRC PMAC program (grant-in-aid no. PA14197), with partner support from Sandoz, SmithKline Beecham, Pfizer Canada Inc., Hoffman LaRoche Ltd., Janssen Pharmaceutical Inc., and Astra. A further extension for ACCORD-II added sponsors Eisai, Hoechst Marion Roussel Canada Research Inc, Bristol Myers Squibb (US), and Eli Lilly Canada.
Disclosures
NF, BLB, and BM have nothing to disclose. MLL reports grant support from the Association for Frontotemporal Degeneration.
G-YRH reports receiving funding support for conducting clinical trials from Eli Lilly, AstraZeneca, Tau-Rx, and Roche, outside the submitted work. DF reports personal fees and non-financial support from Pfizer Canada, personal fees and non-financial support from Novartis Canada, personal fees and non-financial support from Lundbeck Canada, personal fees and non-financial support from Janssen Pharmaceuticals, and personal fees from Eli Lilly, during the study.
HHF reports grants from Canadian Institutes of Health Research (CIHR) and Pharmaceutical Manufacturers Association of Canada (PMAC) Partnership Program (ACCORD-I), and grants from Pharmaceutical Company Grant in Aid Funding (ACCORD-II), during the study, as well as clinical trial funding and service agreement with Lilly Pharmaceuticals, service agreement with Eisai Pharmaceuticals, and service agreement with Banner Institute and Genentech Pharmaceuticals, outside the submitted work. Before 2009, during the ACCORD study period (unrelated to the ACCORD), he received personal fees from Pfizer Inc, Sandoz (now Novartis), SmithKline Beecham (now GlaxoSmithKline), Hoffman LaRoche, Janssen Pharmaceuticals, Astra (now Astra Zeneca), Hoechst Marion Rousel (now Sanofi Aventis), Eli Lilly, Eisai, and Bristol Meyers Squibb. Before 2009, UBC site payments were received for sponsored clinical trials with Pfizer Inc, Novartis, GlaxoSmithKline, Hoffman LaRoche, Janssen Pharmaceuticals, and Sanofi Aventis. On leave from UBC between 2009 and 2011, HHF was a full-time employee of Bristol Myers Squibb.
Statement of Authorship
MLL: study concept and design, acquisition of data, analysis and interpretation of data, drafting, revising, and final approval of the manuscript. NF: study concept and design, analysis and interpretation of data, and revising and final approval of the manuscript. BM: acquisition of data, and revising and final approval of the manuscript. G-YRH: acquisition of data, and revising and final approval of the manuscript. DF: acquisition of data, and revising and final approval of the manuscript. BLB: acquisition of data, and revising and final approval of the manuscript. HHF: study concept and design, acquisition of data, analysis and interpretation of data, and revising and final approval of the manuscript.
