Results

Most patients, 58% of whom were women, were referred by general practitioners (35%) or psychiatrists (40%). The mean (SD) age, MMSE, and CAMCOG18 at study entry were 75.6 (8.0) years, 15.1 (7.6), and 51.8 (26.6) respectively. The mean duration after study entry until death was 2.6 (1.4) years. Controls were well matched to the sample of patients with dementia on sex (59% women) and age (mean 75.3 (5.9)). Lewy bodies were detected in the substantia nigra of 22 (24%) patients with dementia, 17 (77%) of whom also had Lewy bodies in the cingulate, 16 (73%) of whom had probable or definite Alzheimer's disease, and all except one (5%) of whom had at least some neurofibrillary tangles on Braak staging. Of the remaining 70, 43 had probable or definite Alzheimer's disease only, two had possible Alzheimer's disease only, eight had Alzheimer's disease plus other conditions (five vascular), and 17 had no Alzheimer's disease (seven vascular, three frontal lobe dementia, one normal pressure hydrocephalus, one previously unsuspected Huntington's disease, one progressive supranuclear palsy, four cause unknown).

Patients with Lewy bodies were more likely to be anosmic than controls (41% v 6%; p<0.0001) or those with Alzheimer's disease (16%; p=0.029, table 1). Those with Alzheimer's disease were not more likely to be anosmic than controls although there was a trend in this direction (p=0.13). The possibilities that the association of anosmia with Lewy bodies in dementia was due to sex, impaired cognitive function, or age were examined and excluded. Although men were more likely to be anosmic than women (10/27v 6/32; χ² =3.8; p=0.050), they were not more likely to have Lewy body pathology (11/27v 11/38;χ² =0.98; p=0.32). The CAMCOG score was no worse in those who were anosmic (n=49) than in those who were not (n=16) (50.6(25.5) v42.5(27.2); t=−1.1, df=63, p=0.29). There was no difference in the age of those who were or were not anosmic (76.8 (9.8) v 75.4 (7.3)). Those with Lewy body pathology did not have better cognitive function (CAMCOG=47.3 (23.8)) than those with Alzheimer's disease only (49.2 (27.2); df=63,t=0.27, p=0.8). Cohen's κ for the test-retest reliability of anosmia after 1 year was 0.52. Controls were less likely to fail to detect lavender (n=6) than would be expected to be anosmic on the UPSIT (n=15).

The possibility that coexisting Alzheimer's disease pathology contributed to the association of Lewy body pathology with impaired olfaction was examined further and excluded. In the sample of 65 with pure Alzheimer's disease or Lewy bodies the severity of neurofibrillary pathology was not different in those with and without anosmia (Braak stage: 4.3 (1.2) v 4.5(1.4); Mann Whitney U=288,z=−1.18; p=0.24). Separate analyses on those with Lewy bodies (n=20, two patients with missing Braak data), and those with Alzheimer's disease only (n=43) also failed to show any relation between severity of neuropathological Alzheimer's disease and anosmia (patients with Lewy bodies: Braak stage=3.5 (1.0)v 3.7(1.9); Mann WhitneyU=41, z=−0.51, p=0.62. Alzheimer's disease only: Braak stage=5.2 (0.42)v 4.8(1.1); Mann WhitneyU=118, z=−0.27, p=0.81). Finally, the possibility that Alzheimer's disease and Lewy body pathology acted synergistically to produce anosmia was examined and excluded. Those with both Lewy body and Alzheimer's disease were no more likely to be anosmic (seven of 16 patients) than those with Lewy body but without Alzheimer's disease (two of six patients) (Fisher's exact p=1.0).

The relative contributions of cortical Lewy body pathology and SN cell loss to anosmia were examined in the sample of patients with Alzheimer's disease or dementia with Lewy bodies (n=65). There was a strong association between anosmia and a higher cortical Lewy body score (3.3 (2.7) v 1.3 (1.8); Mann WhitneyU=188, z=−3.1, p=0.002). The SN cell counts were also lower in those with anosmia (318 (178) v 471(218); df 58, t=−2.4, p=0.02). Confining the sample to the 22 patients with Lewy bodies in the SN, the mean cortical Lewy body score was greater in the nine patients with anosmia (4.8 (2.4)v 2.5 (2.5);t=2.1, df=20, p=0.05), but there was no significant difference in the SN cell count (242 (119)v 308 (126);t=−1.2, df=19, p=0.24).

Logistic regression analyses were performed, in which the dependent variable was anosmia. The following independent variables were entered in a forward conditional analysis: mean SN cell count, cortical Lewy body score, Braak Alzheimer's disease stage, sex, CAMCOG, and age at study entry. In the final model, cortical Lewy body score (B=−0.48, SE=0. 18, Wald=7.09, df=1, p=0.008) and sex (B=1.93, SE=0.83, Wald=5.4, df = 1, p=0.02) contributed independently to the presence of anosmia, but the other variables did not. The results were unaltered in simultaneous entry and backward stepwise analyses and by omission of CAMCOG score from the variable list.

An exploratory analysis of cortical Lewy body density in individual brain areas showed that anosmic patients had significantly greater density of cortical Lewy bodies in the cingulate gyrus than those without (mean density Lewy body/mm²= 0.64 (0.59)v 0.27 (0.54); Mann WhitneyU=25, z=−2.25, p=0.025). This was not apparent in the other five cortical areas examined.

The consensus criteria for probable dementia with Lewy bodies17 performed very well in patients without CERAD definite Alzheimer's disease, identifying all patients with Lewy body in the SN and cingulate (sensitivity=100%, PPV=73%). The criteria were specific but lacked sensitivity in patients with definite Alzheimer's disease (table 2). Overall, the positive predictive value of anosmia was only 43%. In patients with CERAD definite Alzheimer's disease, anosmia was slightly more sensitive (55%) than the consensus criteria for dementia with Lewy bodies (33%). A new criterion for probable dementia with Lewy bodies that included anosmia and demanded two of four symptoms (parkinsonism, visual hallucinations, fluctuation, and anosmia) identified two more true positive patients than the existing criteria, but also yielded two more false positives.

Discussion

This is the first clinicopathological series to correlate olfactory function in life with necropsy findings. Our results confirm the presumed association of impaired olfactory function with Lewy body pathology, and show that this association extends to patients with dementia as well as those with clinical Parkinson's disease. One explanation for the surprising discrepancy between our data and those clinical studies which have found impaired odour detection in Alzheimer's disease may lie in the failure of clinical criteria for Alzheimer's disease to exclude patients with Lewy body pathology.

Why is dementia with Lewy bodies, but not Alzheimer's disease, associated with anosmia? There are no olfactory brain areas that are heavily affected by Lewy bodies in dementia with Lewy bodies, but remain unaffected in patients with pure Alzheimer's disease. The possibility of an additive effect on olfactory function of Alzheimer's disease and Lewy body pathology is attractive as, in this series, all patients with Lewy bodies except one also had at least some Alzheimer's disease pathology. Different neurons in the anterior olfactory nucleus or olfactory cortices might be affected by Lewy bodies and tangles,21 or, in the amygdala, both abnormalities may be present in the same cells.22 However, this hypothesis is not supported by either simple χ² or regression analyses of our data on cortical pathology. Although Lewy body involvement and cell loss in the anterior olfactory nucleus is extensive in Parkinson's disease, with almost complete group separation from controls,23 cell loss and tangle formation are also more severe in the olfactory bulb of those with Alzheimer's disease than controls.24 21 Unfortunately the loss of olfactory bulbs during processing and storage made it impossible to examine the effect of additive pathology in peripheral structures.

Our failure to find an association of anosmia with Braak Alzheimer's disease stage could be explained if a ceiling in the effect on olfactory function of Alzheimer's disease (anosmia) is reached at very early Braak stages. However, this possibility would contradict data from clinical studies which show that olfactory dysfunction is correlated with degree of dementia in those with established Alzheimer's disease.10 25 Furthermore, without doing olfactory bulb biopsies, or large clinicopathological studies of controls, this explanation is untestable.

Odour detection may be dependent on dopaminergic neurons, which are selectively affected by Lewy bodies.26 The anterior cingulate, a predilection site of Lewy bodies, is rich in dopaminergic innervation, which may explain our finding of increased cingulate Lewy body density in patients with anosmia. The cingulate and frontal regions are activated by odours in normal volunteers, as well as regions previously recognised as olfactory cortex (entorhinal and orbitofrontal).27 In patients with hyposmia, activation is reduced in the cingulate, orbitofrontal, medial temporal, and posterior temporal regions, but is restored when olfactory function is improved.28

A strength of the study was the standardised assessment of fluctuation, which, although it is a core feature of dementia with Lewy bodies in the consensus clinical criteria,17 is difficult to define. We found that short term fluctuations lasting “minutes or hours” segregated with other elements of the clinical syndrome of dementia with Lewy bodies and were associated with Lewy body pathology whereas longer fluctuations of “hours or days” were not. The within patient blinding of pathological assessment was also a strength. It reduced the possibility of confounds due to confusion of ubiquitinated globose tangles and cortical Lewy bodies. Although α-synuclein immunohistochemistry may have been more sensitive, it was not available at the time of this work, and there are emerging doubts about its specificity for Lewy bodies.29 30 The only report comparing ubiquitin and α-synuclein in the cortex found similar numbers of Lewy bodies staining with each.31

The main limitation of the present study lies in the method of olfactory testing. As there were no dummy bottles, suggestibility and response bias could not be ruled out, but the cognitive function of the Alzheimer's disease and dementia with Lewy bodies groups was not different and there is no obvious reason why those with Alzheimer's disease should be more likely to say that they could detect an odour than those with dementia with Lewy bodies. Conventional tests of detection threshold use sequential presentation of differing concentrations of odorants but these longer tests, although more reliable,32 would not be feasible for patients with dementia. The test-retest reliability over a year (κ=0.52) was modest but is likely to be confounded by patients who genuinely developed anosmia over that period, and was better than the 2 week reliability of single ascending butanol detection threshold tests in normal subjects.32Fewer controls were anosmic on this test than expected, suggesting that the olfactory stimulus might have been of a higher concentration than occurs in UPSIT and raising the possibility that some patients reported detection of the odour because of a trigeminal effect. Such an effect would have worked against detection of an association of anosmia with Lewy bodies.

The main problem with the existing clinical diagnostic criteria for dementia with Lewy bodies is poor sensitivity due to the masking of symptoms by concomitant Alzheimer's disease (table 2). The crude test we used did not improve the predictive value of the criteria. Odours such as pizza and wintergreen may be better at discriminating patients with Parkinson's disease from controls,2 although this awaits confirmation. It is not known whether these would be better discriminators between Alzheimer's disease and dementia with Lewy bodies than the test that we used as reports of reduced odour sensitivity in Alzheimer's disease depend on the odour being tested.10 The finding that patients with dementia with Lewy bodies were less likely than those with Alzheimer's disease to be able to detect lavendar on a classic bedside test holds out the possibility that a simple test may yet be developed which has the right sensitivity profile to be a useful addition to existing diagnostic criteria.

Smells are probably less well semantically encoded in humans than other stimuli, making olfactory recognition and identification relatively more vulnerable to early Alzheimer's disease than detection.8 33 This also explains why tests of odour identification may be markers for subsequent cognitive decline.34 35

Nevertheless, given that detection is a prerequisite for identification, our results suggest that neuropathological validation is necessary before any impairments in olfactory function can be confidently attributed to Alzheimer's disease rather than dementia with Lewy bodies.