A Clinicopathological Comparison of Community-Based and Clinic-Based Cohorts of Patients With Dementia FREE

DEMENTIA IS a syndrome of cognitive impairment associated with functional and socioprofessional repercussions.¹ In the United States and Europe, the most common cause of dementia is considered to be Alzheimer disease (AD), followed by vascular dementia (VD).^{2 - 6} However, in Japan^{7 - 8} and some other countries⁹ where stroke and its risk factors are prevalent, VD is the main cause of cognitive impairment. The diagnosis of probable AD is based on clinical criteria^{1 ,10} but is proven only on neuropathological examination.^{11 - 12} Similarly, clinical criteria have been developed for the diagnosis of probable VD, but definite diagnosis requires pathological confirmation.¹³

The accuracy of clinical diagnosis vs pathological confirmation varies widely in the literature.^{14 - 17} Most of the clinical and clinicopathological studies on dementia were carried out in hospital-based or clinic-based populations. To our knowledge, no studies have correlated clinical and pathological diagnosis in a cohort of patients from the community. Hence, no information exists regarding the accuracy of clinical diagnosis in that setting. These patients may differ demographically and clinically from patients with dementia who were drawn from the clinic or hospital settings. These differences may in turn affect the causes of dementia and the clinical accuracy of diagnosis compared with pathological findings.

The purpose of this study was to compare the clinical and neuropathological diagnoses of 2 groups of patients with dementia. The first group was selected from a memory disorder clinic and the second from a community-based cohort of patients. Our main objectives were 3-fold: (1) to compare the sensitivity and specificity of the clinical diagnoses of the 2 groups, (2) to determine whether the distribution of causes differs according to patient population, and (3) to evaluate the effects of different demographic and clinical characteristics on the final diagnosis.

The study population consisted of 2 groups of patients. The first group comprised 63 patients from the Memory Disorders Clinic that is associated with the federally supported Alzheimer's Disease Research Center, a tertiary referral facility for patients with cognitive complaints. The second group included 26 patients selected from a large community-based study of elderly individuals over the age of 65 years who were participating in an epidemiologic study of aging and dementia in northern Manhattan.

The diagnosis of dementia was based on the Diagnostic and Statistical Manual, Revised Third Edition (DSM-III-R), criteria.¹⁸ Alzheimer disease was diagnosed according to the National Institute of Neurological Communicative Disorders and Stroke–Alzheimer's Disease and Related Disorders Association (NINCDS-ADRDA) criteria.¹⁰ The National Institute of Neurological Disorders and Stroke and the Association Internationale pour la Recherche et l'Enseignement en Neurosciences (NINDS-AIREN) criteria were used for the diagnosis of VD.¹³ The clinical evaluation included a structured neurological and psychiatric examination, neuropsychological testing, and a series of laboratory tests. Each patient's complete blood cell count, serum electrolyte levels, serum vitamin B₁₂ level, and folate level were measured. In addition, liver and renal function tests, Veneral Disease Research Laboratory tests, thyroid function tests, and brain imaging (either computed tomography or magnetic resonance imaging) were done. Findings on examination and laboratory studies were used to reach the final clinical diagnosis at a consensus conference of neurologists, psychiatrists, and neuropsychologists. While all patients in the clinic group received the entire workup, neuroimaging or laboratory data were not available for many in the community group. Both populations were evaluated by the same group of physicians.

The patients' characteristics included were sex, race/ethnicity (white, black, or Hispanic), education (years), stage of disease at entry and at the last visit as assessed by the Clinical Dementia Rating scale,¹⁹ age at the time of entry into the study, age at the time of death, and vascular risk factors.

All patients and families in the Memory Disorders Clinic were given information about autopsy and gave nonbinding verbal consent if they wished to participate in the autopsy program. In the community, no recruitment effort was systematically attempted because of the concern that participation in the aging study might be adversely affected. Older patients with severe dementia who resided in nursing homes and had family members who were accessible or involved in their care were targeted for discussions about autopsy. Legal consent was obtained upon death for all patients.

Neuropathological examinations were performed at the Columbia Presbyterian Medical Center, New York, NY, and diagnosis was based on either the Consortium to Establish a Registry for Alzheimer's Disease criteria¹¹ or the Khachaturian criteria¹² for AD. Diagnoses of other dementias were made using standard published clinical and pathological criteria.^{11 ,20 - 22}

For purposes of statistical analysis, clinical and pathological diagnoses were classified as either AD or non-AD. The clinical and pathological diagnoses were used in 2×2 contingency tables to determine sensitivity (the number of patients with the clinical diagnosis of AD who met the pathological criteria for AD, or true positives, divided by the total number of patients with the neuropathologically definite diagnosis of AD) and specificity (the number of patients without the clinical diagnosis of AD who did not meet the pathological criteria for AD, or true negatives, divided by the total number of patients without a neuropathologically definite diagnosis of AD). Bivariate analyses were done to compare different subgroups of patients; t test and χ² analysis were used for continuous and discrete variables, respectively. Chi squared analysis was also used to compare the sensitivity and specificity of clinical diagnoses in the 2 groups. Logistic regression analyses, including variables of interest, were used to identify the determinants of brain autopsy and to further compare the 2 groups. Statistical analyses were performed using SPSS for Windows (version 8.0; SPSS Inc, Chicago, Ill).

Compared with clinic patients who died but did not have autopsies, autopsied clinic patients were younger at the time of diagnosis (P<.001) and death (P=.05), more educated (P=.001), more likely to be white (P=.009), more likely to have been in the study for a longer period (P=.008), and more likely to have a diagnosis of probable AD (P<.001) (Table 1). Race lost its significance when we adjusted for age and education in a logistic regression model.

The same analyses were carried out in the community cohort of patients (Table 2). Community patients who underwent autopsies were more likely to have been in the study for a longer period (P<.001) and more likely to have a diagnosis of probable AD (P<.001).

Patients from the clinic and from the community who underwent autopsies were compared (Table 3). Bivariate analyses showed that clinic patients were younger at the time of entry into the study (P<.001), more educated (P<.001), and more likely to be white (P<.001). Community patients had more severe dementia at the time of intake (P<.001) and death (P<.001) (data not shown), had a shorter period of participation in the study (P=.03), and were older at the time of death (P<.001).

Of the 63 clinic patients who underwent autopsies, 3 (5%) did not have clinically diagnosed dementia, 40 (63%) had a diagnosis of probable AD, 6 (10%) had a diagnosis of possible AD, and 14 (22%) had another diagnosis (ie, chromosome 17–associated familial dementia [n=2], frontal lobe dementia [n=1], dementia associated with Creutzfeldt-Jacob disease [n=4], Huntington disease [n=2], amyotrophic lateral sclerosis with dementia [n=1], adult polyglucosan disease [n=1], Parkinson disease–associated dementia [n=2], and dementia of unknown cause [n=1]). In the community group of patients, the clinical diagnoses were distributed as follows: 3 (12%) had no dementia, 11 (42%) had probable AD, 3 (12%) had possible AD, and 9 (35%) had another diagnosis (Parkinson disease–associated dementia [n=5], VD [n = 1], multisystem atrophy [n=1], and dementia of unknown cause [n=2]). The clinical diagnoses of the 2 groups of patients who had autopsies were compared and did not differ significantly.

Of the clinic patients (n=63), 29 (46%) had a pathological diagnosis of definite AD (6 with Lewy bodies), 15 (24%) had a diagnosis of mixed AD (AD associated with cerebrovascular disease), and 19 (30%) had a diagnosis of some other type of dementia (Creutzfeldt-Jacob disease [n=4], progressive degenerative disease [n=3], diffuse Lewy body disease [n=2], Huntington disease [n=2], multisystem atrophy [n=1], frontal lobe dementia [n=1], progressive supranuclear palsy [n=1], amyotrophic lateral sclerosis [n=1], Parkinson disease [n=1], multiple sclerosis [n=1], adult polyglucosan disease [n=1], and subcortical gliosis [n=1]).

The pathological diagnoses in the community group were distributed as follows: 6 (23%) had definite AD; 6 (23%) had mixed AD (AD associated with cerebrovascular disease); 6 (23%) had cerebrovascular disease, including infarcts and hypertensive vasculopathy (5 were nonwhite); and 8 (31%) had a diagnosis of another type of dementia (ie, Parkinson disease [n=3], progressive supranuclear palsy [n=2], Wernicke disease [n=1], diffuse Lewy body disease [n=1], and frontal lobe dementia [n=1]). The difference in distribution of pathological diagnoses between the clinic group and the community group was significant only for cerebrovascular disease (P<.001). However, this significance disappeared when the distribution of pathological diagnoses was adjusted for differences in race, age, and education between the 2 groups. Vascular risk factors (ie, smoking, high blood pressure, atrial fibrillation, coronary artery disease, congestive heart failure, hyperlipidemia, and diabetes mellitus) were significantly more common in the community group (54% [n=14]) than in the clinic group (27% [n=17]) (P=.02). However, adding this variable to the logistic regression model did not alter the outcome of the analysis.

For patients who were seen at the clinic, sensitivity was 98% and specificity was 84% (Table 4). Four patients were incorrectly diagnosed in the clinic. Three with clinically diagnosed AD actually had diffuse Lewy body disease, subcortical gliosis, and multiple sclerosis, as determined at the time of the postmortem examination (false positives). One patient with clinically diagnosed dementia of unknown cause had AD on pathological examination (false negative). There was no change in the clinical diagnosis over time.

For patients in the community group, the sensitivity of the last clinical diagnosis of AD was 92% and the specificity was 79% (Table 5). Clinical and pathological diagnoses were different in 4 patients. Three patients with clinically diagnosed AD (false positives) had pathological diagnoses of cerebrovascular disease (n=2) and frontal lobe dementia (n=1), and 1 patient with a clinical diagnosis of dementia of unknown cause had typical changes of AD (false negative). The clinical diagnoses changed in 5 community patients (19%) over time (mean duration of follow-up, 2 years; SD, 1.7 years). When the clinical diagnosis at the time of the first visit was considered, sensitivity was 75% and specificity was 86%. The improvement in sensitivity was a result of the clinical reassignment of 2 patients who were initially not thought to have AD to a diagnosis of probable AD. Both of these patients had AD on pathological examination. The slight deterioration in specificity is explained by the changes in the clinical categories of 3 patients—2 patients who were initially considered not to have clinically diagnosed dementia were rediagnosed with AD (whereas pathological diagnosis indicated cerebrovascular disease), and 1 patient with a pathological diagnosis of non-AD who was initially diagnosed with AD was reassigned to the clinical non-AD category on follow-up.

The differences in sensitivity and specificity between the clinic and community patients were not statistically significant.

The aims of this clinicopathological study were to compare the sensitivity and specificity of the clinical diagnosis of AD, the distribution of pathological causes, and the demographic and clinical characteristics of 2 different groups of patients with dementia, 1 selected from a tertiary memory disorder clinic and the other selected from a community-based cohort.

We found that clinic patients who had autopsies were younger at the time of diagnosis and at the time of death, more educated, more likely to be white, and more likely to have been in the study for a longer period than patients who did not have autopsies. Community patients who had autopsies were more likely to have been in the study for a longer period. In accordance with our findings, Harrell et al²³ reported on 69 deceased patients in a memory clinic and found that those who underwent autopsies were more likely to be white and younger at the time of presentation and at the time of death. However, there were no major differences in clinical diagnoses for autopsied and nonautopsied patients. They concluded that ". . . the frequency of occurrence of various dementias obtained through autopsied series are representative of the demented population." In our experience, those who underwent autopsies were more likely to have received the diagnosis of probable AD as opposed to possible AD or other types of dementias. We believe this to be owing to the fact that patients with the diagnosis of probable AD represent our purest category of patients with AD clinically and are most frequently recruited into research studies. Consequently, they are examined more closely and more frequently, which possibly enhances their relationship with the clinic staff and improves the odds of their families consenting to have an autopsy performed. Fillenbaum et al²⁴ found that out of 308 consecutively deceased patients with AD at 24 Consortium to Establish a Registry for Alzheimer's Disease sites, the 167 who had autopsies were more likely to be white, to be better educated, to have been in the study longer, and to have had a longer total duration of AD. The racial difference persisted even after adjustment for education and other demographic characteristics in a logistic regression model. In contrast, we found that when we adjusted for education, age, and clinical diagnosis in the clinic sample, race lost its significance, suggesting that race was confounded by these other variables, and that by itself it was a negligible factor in determining which patients had autopsies in this group. Because of the general demographic characteristics of the patients seen at our Memory Disorder Clinic, 18% of all autopsied patients were nonwhite. This proportion is higher than the 6%²⁴ and the 15%²³ reported in 2 previous studies.

WHEN CLINICAL and pathological diagnoses were compared, the sensitivity and specificity of the clinical diagnosis of AD were not significantly better in the group of patients from the clinic. This could be explained by the fact that the same group of clinicians evaluated the 2 samples. The sensitivity and specificity at our Memory Disorder Clinic are among the highest reported in the literature^{14 - 15 ,25} and are enhanced by the reliance on rigorously established clinical criteria and the consensus process.²⁶ The lower values in the community-based sample were a result of 3 false positives and 1 false negative, reducing the specificity to 79%. There was a tendency to overdiagnose AD in the community group. One reason that may account for this finding is that clinical diagnosis in the community group was often made without the use of laboratory tests and neuroimaging procedures. Two of the 3 erroneously diagnosed patients with AD actually had vascular disease on pathological examination. This diagnosis might have changed if brain imaging had been available. However, in the absence of a clinical history suggesting an association between the cognitive symptoms and neuroimaging findings, these patients would have been classified at best as having mixed AD and VD (or possible AD, according to the NINCDS-ADRDA criteria), which would not have changed their clinical diagnoses. Furthermore, the patients in the community group were part of an epidemiological study on aging and were selected randomly from the general population, whereas the patients in the clinic group were all referred by primary care physicians because they had memory complaints. Nevertheless, the 92% sensitivity and 79% specificity achieved in the clinical diagnosis of AD in the community group seem quite good compared with the values for the clinic group. In fact, these values are better than those reported by several clinicopathological studies carried out in memory disorder clinics and in hospital settings that based their diagnoses on extensive clinical and laboratory workups.^{14 ,16 ,27 - 29} This suggests that by doing a structured clinical evaluation that includes an extensive history, a detailed physical examination, and neuropsychological testing in a group of elderly patients in the community, very good clinical sensitivity and specificity can be achieved in the diagnosis of AD. A study evaluating the role of a neuropsychological paradigm similar to the one we used in the diagnosis of AD^{30 - 31} showed it to be very useful in minimizing interobserver diagnostic variability over time and in contributing to the improved accuracy of diagnosis. Furthermore, the improvement in sensitivity with follow-up underscores the importance of repeated evaluations in reaching a correct diagnosis.

Comparison of clinic and community patients revealed significant demographic differences. Patients in the community cohort were older, less educated, and more likely to belong to a minority group, which can be explained by the characteristics of the population of our catchment area. Clinically, these patients were older at the time of death, had a shorter period of participation in the study, and had more severe dementia at the time of intake and death. These differences are owing in part to the fact that we focused our recruitment efforts in the community on older age, a more advanced stage of dementia, and admission to a nursing home.

A vascular cause for dementia on pathological examination was found more often in the community group than in the clinic group. In fact, none of our clinic patients who underwent autopsy was diagnosed with pure VD on clinical grounds or on pathological examination. The scarcity of VD in the pathological reports of patients with dementia from memory disorder clinics has been recently confirmed by several studies.^{14 ,17 ,32 - 37} One reason for this finding is that clinicians are reluctant to diagnose VD in patients with obvious symptoms and signs of stroke.³⁸ Hence, this subgroup of patients is less likely to be referred to memory disorder clinics³⁹ and more likely to be referred to stroke clinics. These referral patterns may also vary between countries, since geriatric services in Europe seem to receive a higher proportion of patients with VD.^{33 ,40} Some authors have criticized the NINDS-AIREN criteria for including all types of cerebrovascular disease and have suggested using the term primary dementia to describe VD and the term stroke with secondary dementia in patients with overt stroke and cognitive symptoms.⁴¹ Pathological diagnosis of cerebrovascular disease was significantly more prevalent in the community-based cohort of patients. This is not surprising in light of the higher frequency of vascular risk factors in the community group. Five of the 6 patients with this diagnosis were nonwhite, and when pathological diagnosis was adjusted for differences in race, age, and education between the 2 groups, this significance disappeared. This suggests that demographic characteristics may play an important role in the etiology of certain types of dementias. Recent studies have shown that, in addition to the typical risk factors for cerebrovascular diseases (eg, older age, being male, hypertension, hyperlipidemia, diabetes mellitus, atrial fibrillation, congestive heart failure, coronary heart disease, carotid bruit, cigarette smoking, and alcohol abuse² ), being nonwhite⁴² and having fewer years of education,⁴³ among other factors, predispose patients to VD.

In conclusion, our study shows that the sensitivity and specificity of clinical diagnoses differed only slightly between the clinic group and community group when structured clinical diagnoses were made and careful follow-up was maintained. The prevalence of different subtypes of dementias also differed between the 2 groups. These differences can be explained, at least in part, by the different demographic and clinical characteristics of these 2 populations and by the types of clinical evaluations they received.

Accepted for publication December 14, 1998.

This study was supported by grants AG07232, AG10963, and RR00645 from the National Institutes of Health, Bethesda, Md; by the Charles S. Robertson Memorial Gift for Alzheimer's Disease Research from the Banbury Fund, Cold Spring Harbor, NY; and by a grant from the Blanchette Hooker Rockefeller Fund, New York, NY. Dr Massoud is supported by a grant from the R. Samuel McLaughlin Foundation, Ottawa, Ontario.

Presented as a poster at the American Neuropsychiatric Association Meeting, New Orleans, La, January 31–February 2, 1999.

Corresponding author: Richard Mayeux, MD, MSc, G. H. Sergievsky Center, 630 W 168th St, New York, NY 10032 (e-mail: rpm2@columbia.edu).