Clinical-pathologic studies of aging and dementia from specialty clinics rather than population or community cohorts may be vulnerable to selection bias. We investigated the effect of sample selection on age-related brain pathology in older persons with and without dementia from two longitudinal community-based cohorts compared to one specialty clinic-based cohort. We found important differences between the cohorts. Community subjects with probable AD were more likely than clinic subjects to have intermediate compared to high likelihood pathologic AD, and more likely to have cerebral infarcts and mixed pathologies. Community MCI subjects showed a similar proportion with pathologic AD but a higher frequency of infarcts and mixed pathologies compared to clinic subjects. Community subjects with other dementias were much less likely to have evidence of frontotemporal lobar degeneration or other atypical pathologies.
Selection or referral bias has been previously reported in subjects with AD [1
]. Studies have indicated significant sociodemographic [1
], clinical [3
], and genetic [2
] differences between subjects referred to specialty clinics compared to other registry and population cohorts. One study [1
] reported differences in neuropathology, most notably an increase in dementia due to cerebrovascular disease in the community cohort. We extend these findings by merging two community cohorts and comparing them to a clinic cohort that employed similar clinical and neuropathologic methods. In addition, we compared neuropathologic data in five diagnostic groups: no cognitive impairment, MCI, probable AD, possible AD, and other dementias.
AD pathology was less severe in the community sample with dementia compared to the clinic cohort. One obvious explanation is that the older persons with dementia in the community had less severe cognitive impairment compared to demented persons in a specialty clinic. Indeed, we had previously reported less severe cognitive impairment in older persons with dementia in the community [17
]. There are several explanations for the differences in severity of cognitive impairment and AD pathology in the clinic versus the community. First, community subjects are recruited after the age of 65 without dementia and develop dementia during the study, whereas clinic patients are typically recruited at baseline with a history of dementia and impairment worsens during the course of the study. Second, doctors may be more apt to refer younger patients with a more rapid progression as evidenced by the younger age for the clinic patients. Third, behavioral issues, mobility disabilities, and incontinence may be related to an increased likelihood of referral as well as more severe cognitive impairment and more widely distributed AD pathology. In addition, clinic subjects may have fewer co-morbidities and greater financial, social, and personal resources, resulting in longer survival after diagnosis, less mixed pathologies, and more severe pathology at death. Finally age itself may be a factor. Older age has been associated with less AD pathology and a lesser association with dementia [35
Mixed pathologies and infarcts were more common in the community sample with probable AD compared to the clinic. Mixed rather than single pathologies were also seen more commonly in the community of MCI subjects compared to the clinic, though the numbers were very small. Interestingly, stroke was not an exclusionary criterion for enrollment in the clinical core. There are other possible explanations for discordant mixed pathology and cerebrovascular disease between clinic and community subjects. Individuals with vascular cognitive impairment may be more likely to visit a stroke and or general neurology clinic rather than be referred to a Memory disorders or AD clinic. In addition, persons who seek medical attention such as those in the clinic may have healthier lifestyles which may make them less prone toward vascular events. Finally, subjects with cognitive impairment in the community were older, allowing additional pathologies such as infarcts to accumulate. Indeed, the younger age of clinic-based compared to community-based subjects is a potential threat to the generalizability of a broad range of studies on dementia [36
Similar biases are likely responsible for the increase in Lewy body disease and atypical pathologies (e.g., frontotemporal lobar degeneration) in the clinic cohort, including age, associated disabilities, and behavioral issues. Lewy bodies have been previously reported to be associated with advancing age [37
]. In addition, the well known association of neocortical Lewy body disease with such problems as hallucinations, REM sleep behavior disorder, and falls are likely to prompt the solicitation of medical advice or referral. The significant behavioral issues associated with atypical pathologies, such as frontotemporal lobar degenerations, are likely to have similar consequences.
The neuropathologic disparities between the community and clinic cohorts with and without dementia may not be surprising; yet, the real-world consequences have broad implications for many types of research. For instance, because infarcts and mixed pathologies are more common in persons with probable AD in the community, trials in specialty clinics are likely to underestimate the potential benefit at preventing or treating vascular disease in older persons with preclinical or clinical AD. Moreover, selection bias will have the potential to exaggerate expected effect sizes for treatments aimed specifically at AD pathology in the community. Because the burden of dementia will dramatically increase in the older age groups over the next decades, it is important to target prevention and treatment studies toward persons that best reflect the population of older persons.
Whereas, the clinic under-represents older persons with infarcts and mixed pathologies, it likely over-represents persons with atypical pathologies. This also has real-world consequences. For instance, when dementia subtypes are extrapolated from autopsy studies derived from specialty clinics to the community, the prevalence of cortical Lewy body disease and frontotemporal lobar degeneration may be overestimated. Though the clinical-pathologic spectrum of both Lewy body disease and frontotemporal lobar degeneration continues to be explored and the prevalence, particularly of frontotemporal lobar degeneration, is currently uncertain, one should be cautious in extrapolating prevalence data from specialty clinics. These data may influence the allocation of resources and alter the direction of research studies.
There are limitations to this study. We had very few persons without dementia in the clinic sample to compare to the community sample. The community subjects were drawn from two different cohorts that are likely to have important differences. There are also advantages. The community subjects from the two cohorts were found to be very similar in both clinical and pathologic features allowing us to pool this data. All three studies had high autopsy rates, lessening the chance for bias when comparing the community to the clinic. Finally, all three groups had their neuropathology evaluated at the same center, lessening the potential for inter-rater differences in the interpretation of pathology.
In summary, sample selection may have a profound impact on the frequency, distribution, and types of neuropathology found in older persons with and without dementia. Persons from the community, compared to the clinic, tend to show less severe AD pathology, more cerebral infarcts, less neocortical Lewy body disease, and less atypical pathologies (e.g., frontotemporal lobar degeneration). These neuropathologic differences vary by diagnostic group. One should be aware of possible selection bias when using or reporting neuropathologic data derived from specialty clinics of older persons with and without dementia.