A Comparison of an Australian Observational Longitudinal Alzheimer’s Disease Cohort to Community-Based Australian Data

INTRODUCTION

Our understanding of the clinical progression of Alzheimer’s disease (AD) including amyloid-β (Aβ) aggregation, tau hyperphosphorylation, neuroinflammation and cognitive decline have been improved through longitudinal observational studies of AD cohorts over the last two decades.¹ These have been via routine clinical assessments, cognitive ratings, lifestyle questionnaires, genetic testing and monitoring of biomarkers including blood, cerebrospinal fluid (CSF) and neuroimaging (brain magnetic resonance imaging [MRI], amyloid and tau positron emission tomography [PET] scans). These clinical cohorts have enriched the understanding of the natural history of AD, the risk factors for AD, and helped guide the development of AD clinical trials.^2–4

However, there have been concerns regarding the generalizability of findings derived from longitudinal observational AD cohorts. Participants in these cohort studies tend to be predominately Caucasians that are well educated and of higher socioeconomic status, require less visits to see their doctors, have fewer medical co-morbidities, and consume fewer medications compared to the general population.⁵ A comparison of the Alzheimer’s Disease Neuroimaging Initiative (ADNI) cohort in the United States (US) to a community-based cohort from the Atherosclerosis Risk In Communities (ARIC) study, a prospective observational cohort investigating the etiology of atherosclerosis and cardiovascular disease in the US, by Gianattasio et al. showed significant differences in education attainment (completed high school – ADNI 85% versus ARIC 52%), gender (male – ADNI 55% versus ARIC 41%), ethnicity (black – ADNI 4% versus ARIC 24%), hypertension (ADNI 48% versus ARIC 77%), and apolipoprotein E (APOE) status (APOE ɛ4 allele carrier – ANDI 85% versus ARIC 52%).⁶ Compared to the US census in 2019 for people aged over 60 years old, the ADNI cohort had a lower participation of people who identified as either ethnoculturally underrepresented populations such as Blacks or Latinx (ADNI 11% versus US census 2019 25%) or completed less than 12 years education (ADNI 16% versus US census 2019 44%).⁷ These raise concerns regarding the generalizability of the ADNI cohort to culturally diverse and low education populations and could potentially lead to biased estimates of associations with AD and our understanding on the progression of AD in this population.

To our knowledge, there have not been any studies comparing an Australian observational AD cohort to the community. The Australian Imaging, Biomarkers and Lifestyle (AIBL) Study is an ongoing longitudinal cohort study in Australia that began in late 2006 whose overall goal is to determine the extent that demographics, baseline cognitive profile, biomarkers (blood, CSF, MRI, and PET scans), genetic and lifestyle factors can predict a person’s development of AD.⁸

Our aim was to determine if the characteristics of participants with a diagnosis of AD dementia in the AIBL study are representative of the general population of older Australians (aged 65 years old or over) with a diagnosis of AD dementia. We specifically chose AD dementia for comparison between the two groups of AIBL study and the community Australian population due to there being a population denominator in the community that could be used as a comparator. The comparators were older Australians who were under Therapeutic Goods Administration-approved pharmacological treatment (i.e., donepezil, rivastigmine, galantamine, and memantine) for AD as per the Australian Pharmaceutical Benefits Scheme (PBS) linked with the Australian Census performed in 2021 and older Australians who self-reported dementia in the 2021 Australian Census. The PBS is the Australian Government’s subsidized drug prescription program. Since people undergoing pharmacological treatment for AD under the PBS are required to have a confirmed diagnosis of AD by a specialist/consultant physician,^9–12 we therefore have an appropriate population denominator for comparison. We also compared the AIBL cohort to older Australians who self-reported dementia, which was not AD specific but would include people in the community with AD dementia that were not on PBS-listed pharmacological treatment for AD, in the 2021 Australian Census. Thus, our goal was to identify if the findings from the AIBL study on the natural history of AD are generalizable to community-based cohorts.

METHODS

RESULTS

There have been 3,667 participants enroled in the AIBL study as of April 2023. Of these, 591 (16%) participants aged 65 years and older have had a diagnosis of AD dementia (referred to as AIBL AD cohort). In 2021, there were 54,346 people aged 65 or older in Australia who were dispensed with one of the four medications for the treatment of AD that were linked to the Australian census in 2021 (referred to as PBS/census AD cohort). There were 164,425 people aged 65 or older in Australia who self-reported dementia (referred to as dementia/census cohort).

For the PBS/census AD cohort, the most commonly dispensed medication drug class for AD (Table 2) were acetyl-cholinesterase inhibitors (AChEIs; 49,070 people [90%]), of which donepezil was predominant (36,891 people [68%]), followed by rivastigmine (7,589 people [14%]) and galantamine (5,589 people [10%]). Non-AChEI drug memantine was used by 7,331 people (13%). A small portion of the cohort (2,055 people [4%]) used dual therapy of AChEIs and memantine for AD. Medication dispensing records have not been collected in the AIBL study, and therefore no AIBL data is available for this comparison.

A comparison of the baseline characteristics including medical co-morbidities for all cohorts is presented in Table 3 and stratified by sex in Table 4. The AIBL AD cohort was younger compared to the PBS/census and dementia/census cohorts (mean age±SD, AIBL 79±7 years, PBS/census 81±7 years, p < 0.001, dementia/census 83±8 years, p < 0.001). A subgroup analyses of AIBL 2021 data demonstrated that the mean age of AIBL participants with a diagnosis of AD dementia who were assessed in 2021 was similar to the overall AIBL AD cohort (mean age±SD, AIBL 2021 78±6 years versus AIBL 79±7 years, p = 0.409), and was younger than both the PBS/census cohort (p = 0.011) and the dementia/census cohort (p < 0.001). All cohorts had a slightly higher portion of females than males (AIBL 306 people [52%] versus PBS/census 31,088 people [57%], p = 0.011; versus dementia/census 97,579 [59%], p < 0.001). A quarter of the PBS/census AD and dementia/census cohort did not report their highest education attainment (PBS/census 11,479 people [22%], dementia/census 41,671 [25%]). 60% (332 people) of the AIBL AD cohort had less than 12 years of education, compared to 65% (27,317 people) in the PBS/census AD cohort (p < 0.001) and 71% (87,281 people) in the dementia/census dementia cohort. 40% (217 people) in the AIBL AD cohort had greater than 12 years education, with 21% (117 people) reporting post-graduate education (15 + years education). This was much higher than the PBS/census and dementia/census cohorts (>12 years education PBS/census 14,618 [35%] dementia/census 35,473 [29%], 15 + years education PBS/census 1,214 people [2%] dementia/census 2,278 [2%]). There was a higher proportion of females who reported having less than 12 years education compared to males in all cohorts (female < 12 years education AIBL 188 people [66%], PBS/census 17,735 people [77%], dementia/census 57,283 [81%]; male < 12 years education AIBL 144 people [54%], PBS/census 9,582 people [51%], dementia/census 29,998 [58%]).

19% (111 participants) in the AIBL AD cohort,6% (3,388 participants) in the PBS/census AD cohort and 6% (9,350 participants) in the dementia/census cohort did not report their primary language. The majority of participants in both cohorts reported English as their primary language (AIBL 466 people [97%] versus PBS/census 40,897 people [80%]). Only 3% (14 participants) in the AIBL AD cohort reported their primary language to be other than English, which is a much lower culturally and linguistically diverse (CALD) population compared to 20% (10,061 participants) and 22% (33,580 participants) in the PBS/census and dementia/census cohorts respectively.

In the AIBL AD cohort, 8–11% of participants did not complete the questionnaire on the selected co-morbidities (excluding self-reporting a history of mental health condition). Mental health condition was defined as a combination of anxiety, depression and psychiatric illness. Each variable on its own has 9–10% unknown. However, combined together, only one participant did not report one of the three. In comparison, only 4% (2,356 participants) of the PBS/census AD cohort did not complete the questionnaire on the selected co-morbidities. All participants in the dementia/census cohort completed the questionnaire on co-morbidities. In all cohorts, there was a declining proportion of people who self-reported having more than three of the selected conditions compared to only one of the selected conditions, while the AIBL AD cohort has a higher portion of people with three or more medical co-morbidities compared to that for the PBS/census cohort, but a lower portion of people with three or more medical co-morbidities in the dementia/census cohort (one co-morbidity – AIBL 197 people [33%], PBS/census 15,600 people [29%], dementia/census 47,817 [29%]; three or more co-morbidities – AIBL 77 people [13%], PBS/census 5,515 people [10%], dementia/census 28,025 people [17%]). A lower proportion of people reported having none of the selected medical co-morbidities in the AIBL AD cohort (165 people [28%]) than in both the PBS/census AD cohort (24,050 people [44%]) and dementia/census cohort (52,872 people [32%]).

The most commonly self-reported co-morbidities in all cohorts were arthritis and mental health, with strong evidence that the proportion was higher in the AIBL AD cohort compared to the PBS/census and dementia/census cohorts (arthritis – AIBL 243 people [45%], PBS/census 14,584 people [28%], p < 0.001, dementia/census 56,255 [34%], p < 0.001; mental health AIBL 205 people [38%], PBS/census 11,443 people [22%], p < 0.001, dementia/census 48,086 [29%]). The proportion of people in the AIBL AD cohort that reported a history of cancer was two-fold higher compared to people in PBS/census AD cohort (AIBL 97 people [18%] versus PBS/census 4,569 people [9%], p < 0.001) and higher in proportion to the dementia/census cohort (17,887 people [11%], p < 0.001).

DISCUSSION

To our knowledge, this is the first study that compares an Australian observational AD cohort to a community-based AD population. People from a CALD background were underrepresented in the AIBL study (3%) compared to the PBS/census AD cohort (20%) and dementia/census cohort (22%). English was the predominant language spoken in the AIBL AD cohort, with 97% of the cohort reporting English as their primary language. However, this proportion is affected by 19% of participants in the AIBL AD cohort not reporting their primary language. This information bias would have caused differential misclassification, resulting in potentially under or overestimating of the proportion. This observation in AIBL is comparable to The Sydney Memory and Ageing Study, an Australian observational older adult cohort monitoring cognition, where 94% of participants reported English as their primary language.¹⁴ Compared to a non-Australian cohort study, the CALD proportion in the AIBL study (3%) was lower than that reported by the ADNI study in the USA, where 11% of the ADNI participants were identified as being from ethnocultural populations such as Black and Latinx.⁷

In Australia, First Nations people account for 3.8% of the Australian population in 2021.¹⁵ The Koori Growing Old Well study (KGOWS) was a prospective longitudinal cohort study investigating aging and dementia across five Aboriginal communities in New South Wales, Australia across two metropolitan Sydney sites and three regional mid-North Coast sites from 2010 to 2018. The study identified that older age, male sex, unskilled work history, polypharmacy, past smoking and carrying an APOE ɛ4 allele to be associated with incident mild cognitive impairment/dementia.¹⁶ However, there are currently no AD cohort studies actively recruiting First Nations people in Australia. First Nation status is not collected in the AIBL study nor in The Sydney Memory and Ageing Study, while the Prospective Imaging Study of Ageing: Genes, Brain and Behaviour (PISA), an Australian observational cohort study of prodromal AD reported only 0.2% of participants identified as First Nation.¹⁷ Given that the prevalence rate of people with dementia is thought to be 3–5 times greater in the First Nations population than the general Australian population,^18,19 more efforts are required to improve recruitment and retention of First Nations people to participate in AD cohort studies to improve our understanding of the natural history of AD in this population.

Increasing the CALD population in dementia research in high-income countries has been recognized as an area that requires improvement.^20,21 This study suggests that there needs to be more efforts to improve the CALD population recruitment and retention in the AIBL study. A systematic review into potential barriers in recruitment and retention of underrepresented populations in trials has identified issues with mistrust, stigma, a lack of access to information of results and competing demands to participants in studies.²² One of the potential barriers for AIBL to recruit CALD participants could be that the study consent form, questionnaires, and cognitive tests were only in English, which makes it challenging to establish trust and make potential participants understand how the study runs. In addition, participants would be involved in multiple investigations over multiple days which include invasive tests such as a lumbar puncture (although optional for AIBL) and can be burdensome. Recruitment and retainment of First Nations people in research has been challenging due to issues of mistrust in research within First Nation communities and poor engagement with the First Nation people from researchers.²³ The AIBL research group could engage people from the CALD Australian communities and First Nations people to identify the barriers and develop enablers to promote CALD recruitment. Since 2020, ADNI has addressed concerns of low underrepresented ethnocultural communities in their AD cohort by forming an ADNI3 Diversity Task Force, which established less burdensome investigations making lumbar punctures optional and having an advertising agency to develop culturally tailored digital media campaigns to improve the outreach effort.²⁴ This campaign has increased screening and enrollment of ethnoculturally underrepresented communities from 11% in ADNI to 26.7% in the ADNI 3 cohort. The AIBL study could adopt a similar approach. Questionnaires specifically asking social determinants of health would also be critical to understanding its association with AD. This would enhance our knowledge of the influence of culture and ethnicity on the natural history of AD and determine future treatment options.

40% of participants with a diagnosis of AD dementia in the AIBL study reported greater than 12 years education, which was higher than 35% in the PBS/census cohort and 29% in the dementia/census cohort. While there was strong statistical evidence of a difference (p < 0.001), caution should be noted that the PBS/census AD cohort had 23% of participants and the dementia/census cohort had 25% of participants who did not report their highest education attainment. Differential misclassification would have resulted from the information bias, resulting in potentially under or overestimating the population’s highest education attainment above or below 12 years. In comparison to AIBL participants with a diagnosis of AD dementia, 84% of ADNI participants had reported an education greater than 12 years,⁷ which was higher than 56.3% in the 2019 United States census. Globally in AD observational cohorts, the average education of participants was approximately 14 years.²⁵ Low education has been identified by the Lancet commission on dementia prevention as a key risk factor for the development of dementia.²⁶ Cohort studies like AIBL should ensure to recruit not only those with more years of education but also those with less years of education for better generalizability of research findings.

There was strong statistical evidence of a higher proportion of participants in the AIBL AD cohort self-reporting having one or more of the selected medical co-morbidities compared to the PBS/census AD cohort taking medications for AD (p < 0.001) and the dementia/census cohort (p < 0.001). This is an intriguing paradoxical finding, as other observational AD cohorts have reported less co-morbidities than in the general population,⁵ likely due to selection bias from healthy volunteer bias. A potential contributor to this finding is information bias since 8–11% of AIBL participants did not fill out the questionnaire on selected medical co-morbidities. It is also possible that the AIBL AD cohort self-reported more medical comorbidities than the PBS/census cohort and census dementia cohort due to seeing their GP/specialists more often whilst they were on the study and thus able to report more comorbidities.

Several limitations of the current study must be noted. Firstly, the overall AIBL AD cohort was used instead of using a cross-section of AIBL data in 2021, which may lead to bias when used as a comparator to the PBS/census cohort. In our study we used PBS/census data that only included participants with a diagnosis of AD dementia for comparison, which limited us using all the AIBL participants who were assessed in 2021. To increase the sample size of AIBL participants with a diagnosis of AD dementia, we used cumulative AIBL data. Although the mean age of the AIBL 2021 AD cohort was comparable to the overall AIBL AD cohort (mean age±SD, AIBL 2021 78±6 years versus AIBL 79±7 years, p = 0.409), using the cumulative data rather than cross-sectional data can potentially introduce artifacts. Secondly, the overall AIBL AD cohort reported the age of participants at the time of diagnosis of AD dementia (the incidence), while the PBS/census cohort reported the prevalence of AD dementia in Australia in 2021. This difference has likely contributed to the younger age of the AIBL cohort to the PBS/census cohort. Thirdly, we are limited in the community-based PBS/census cohort where we only have a population that has been dispensed medications for AD. Not all people with AD would have been prescribed medications for AD leading to selection bias. There are potentially people in Australia who do not have access to pharmacological treatment for AD due to health inequities such as being unable to see a specialist for a diagnosis of AD dementia and then procuring the medications. There is an underdiagnosis of dementia.²⁷ People with AD dementia identified via the PBS are also younger and more educated compared to other administrative datasets in Australia (hospital inpatient records, aged care assessments, aged care funding instruments, death certificates) given that medications are prescribed at the early stages of AD dementia.²⁸ Geographic remoteness and socioeconomic disadvantage have also been associated with reduced cholinesterase inhibitor prescription rates in Australia.²⁹ Anticholinesterase inhibitors and memantine can also sometimes be prescribed to patients for other types of dementia or conditions, but for their use under the PBS, it requires a diagnosis of AD by a specialist/consulting physician. We included the 2021 census self-reported dementia cohort as an additional comparator since it would include who were diagnosed with AD dementia but were not on PBS-listed pharmacological treatment. This is however limited that the census dementia cohort is not AD specific, encompassing all types of dementia, the underdiagnosis of dementia,²⁷ and that there are different estimates of the prevalence of dementia in Australia depending on the dataset used (estimated age- and sex-standardized prevalence in Australians aged 60 years or more –31.4 per 1000 people (2021 census), 21.4 per 1000 people (2018–19 NPS MedicineWise survey), and 65.9 per 1000 people (2021 Australian Institute of Health and Welfare).³⁰ Finally, self-reporting for co-morbidities via a questionnaire were used for all cohorts, which could contribute to potential recall bias,³¹ especially for medical history obtained from participants with known cognitive impairment or family members. AIBL participants were asked to self-report 28 conditions including date of diagnosis and include details such as kind, severity, and treatment where possible. The AIBL questionnaire did not ask if the conditions were diagnosed by a doctor or if the conditions were long term (lasting for longer than six months or more). In comparison, in the 2021 census, people were asked if a doctor or nurse had told them if they had ten health conditions and if it had lasted or was expected to last for six months or more. The different self-reported questionnaires could have influenced how participants responded. In the AIBL cohort, arthritis could have been self-diagnosed by the participant rather than diagnosed by a doctor. A mental health condition reported by a participant in the AIBL cohort may not have been long term and current, and thus would not been reported by participants in the census. Self-reported long-term medical conditions was a new question in the census in 2021 compared to the census performed in 2011 and 2016. A report on the quality of the 2021 census data suggested that long-term health conditions results compared similarly to the 2017–2018 National Health survey for all ages, with the exceptions that cancer was reported higher in the census than the National Health survey, and mental health condition was reported higher in the National Health survey than the census (National Health survey – cancer 1.8%, mental health condition 20.1%; 2021 census – cancer 2.9%, mental health condition 8.8%).³² Further investigation with other datasets in older Australians with dementia is warranted to validate the self-reported long-term health conditions in the 2021 census. A more systematic approach to compare AD cohort studies and the general population should ideally be incorporated at the design phase of future cohort studies, and it can be implemented in AIBL study.

Future studies are encouraged to compare the AIBL cognitively normal and mild cognitive impairment cohort to a community-based cohort in Australia and comparing other observational AD cohorts in Australia and globally to a community-based population to determine generalizability of their findings. The Australian Dementia Network registry encompassing patients newly diagnosed with mild cognitive impairment or dementia from memory clinics and specialists in Australia was commenced in 2020,³³ and could be used in future studies as a community-based comparative cohort.

Overall, compared with the older adults who used medications for AD dementia in Australia in 2021 linked with the Australian census, people with a confirmed diagnosis of AD dementia in the AIBL cohort were younger, received more years of education, and had more medical co-morbidities. The AIBL study has facilitated AD research, improved our understanding on the natural history of AD, and helped in the design of interventional clinical AD trials. However, care is required regarding the generalizability of the AIBL cohort and the interpretation of the findings from the AIBL study, in particular in the CALD population. Increased efforts are required to enrich the AIBL cohort with a more culturally diverse population.

AUTHOR CONTRIBUTIONS

Andrew Liem Hieu Huynh (Conceptualization; Data curation; Formal analysis; Methodology; Writing – original draft); Yihan Wang (Data curation); Liwei Ma (Formal analysis); Yi Ling Clare Low (Formal analysis; Writing – review & editing); Weisi Chen (Data curation; Formal analysis; Visualization); Christopher Fowler (AIBL data collection); Edwin CK Tan (Data curation; Writing – review & editing); Colin L Masters (AIBL data collection); Liang Jin (Conceptualization; Methodology; Supervision; Writing – review & editing); Yijun Pan (Conceptualization; Funding acquisition; Methodology; Project administration; Supervision; Writing – review & editing).