Prospective validation of Consensus criteria for the diagnosis of dementia with Lewy bodies

Methods.

Results.

Fifty cases were autopsied out of 63 who died during the first 3 years of the study—an autopsy accrual rate of 79%. Twenty-six cases had a primary clinical diagnosis of DLB (25 probable and 1 possible), 19 had a primary diagnosis of AD (9 probable and 10 possible), and in 5 cases, VaD was the primary diagnosis (2 probable and 3 possible).

Thirty-one cases received secondary clinical diagnoses—21 possible AD, 2 probable DLB, 7 possible DLB, and 1 possible VaD. The relationships between primary and secondary clinical diagnoses are shown in table 1 and confirm previous reports that the majority of DLB cases also meet criteria for other dementia subtypes, particularly possible AD.5

At autopsy, 29 neuropathologic diagnoses of DLB were made, 15 of AD, and 5 of VaD. One case had atypical pathologic features of progressive supranuclear palsy (PSP). The pathologic groups were similar for mean (SD) age at death: DLB, 78.3 (5.5) years; AD, 83.4 (4.7) years; and VaD, 79.7 (7.0) years. Sex distributions were also similar: percentage of women in DLB was 58%; in AD, 68%; and in VaD, 60%. VaD cases presented earlier in their illness after first onset of symptoms, at a mean (SD) of 1.3 (0.6) years, compared with DLB at 3.6 (2.4) years and AD at 3.3 (1.8) years. The prevalence of core clinical diagnostic features of DLB (significant fluctuation, persistent visual hallucinations, spontaneous parkinsonism) in the autopsy-confirmed DLB cases were 85%, 85%, and 87%, respectively.

Neuropathologically, of the 29 autopsy-positive DLB cases, 69% had Consensus criteria LB scores > 6 (neocortical category), 24% had LB scores of 3 to 6 (limbic category), and 7% had scores of 2 or less (brainstem category). Patients in these these groups were not distinguishable by any specific pattern of demographic or clinical features. Using semiquantitative senile plaque estimates, 54% were classified as CERAD definite AD cases, 33% as probable AD, and 12% did not fulfill minimum CERAD criteria for AD. Braak staging was: Stage 0, four cases; Stage 1, four cases; Stage 2, 12 cases; Stage 3, four cases; Stage 4, four cases; Stage 5, one case; and Stage 6, no cases. Only one DLB case, therefore, had isocortical AD using the Braak method, and only two cases (7%) had sufficient neocortical neurofibrillary tangles for a diagnosis of AD using Newcastle criteria.24

Twenty of the 29 DLB cases (69%) had no significant vascular pathology. Six (21%) had “incidental” small focal ischemic lesions predominantly in white matter or basal ganglia, and 3 (10%) had a moderate degree of vascular pathology consisting of small infarcts (<15 mL) combined with arteriosclerosis or congophilic angiopathy. In only one case were these vascular lesions considered sufficient, by themselves, to produce a dementing syndrome.

Discussion.

Using Consensus criteria for the diagnosis of DLB, 24 of 29 autopsy-confirmed cases were correctly identified antemortem—a sensitivity rate of 0.83. Two false positive clinical diagnoses of DLB were made (one probable, one possible), producing an overall diagnostic specificity of 0.91 in this sample. Three additional LB cases were given secondary clinical diagnoses of DLB (two possible and one probable), potentially increasing sensitivity to 0.93, although at the cost of reduced specificity (0.76). These figures compare favorably with the corresponding sensitivity and specificity rates for NINCDS-ADRDA diagnoses of AD (0.87, 0.83) and NINDS-AIREN diagnoses of VaD (0.40, 1.00) made within this same study and similar to those reported elsewhere.28,29

The group of patients studied was heavily biased toward including patients with DLB, who accounted for 52% of the clinical and 58% of the pathologic sample. This weighting was intentional as the primary objective was to test how the Consensus criteria performed in the positive identification of cases. We anticipated, therefore, that we might tend to overdiagnose DLB clinically, and were surprised to find the opposite, with only 2/26 (8%) false positive diagnoses but 5/29 (17%) of autopsy-confirmed cases remaining undetected.

The over-representation of DLB cases (the prevalence of autopsy-confirmed DLB in patients attending our routine dementia assessment service is typically only 15 to 20%) requires an adjustment to be made so that PPV and NPV rates can be generalized to clinical practice. Positive and negative predictive values are dependent on the prevalence of the disease in the sampling frame and can be calculated for any known prevalence—see Statistical methods for details.

Assuming a DLB prevalence of 15%, the estimated PPV for a primary diagnosis of probable DLB is 0.75 and NPV is 0.96. In other words, 75% of cases clinically assessed as probable DLB in our routine dementia assessment service are likely to be correctly diagnosed, as are 96% of those considered not to have DLB. For all DLB primary diagnoses (probable + possible), PPV = 0.61, NPV = 0.97. The corresponding figures for a DLB prevalence of 20% are probable DLB, 0.80, 0.95; and for probable and possible DLB combined, 0.69, 0.95. A similar calculation to adjust AD prevalence rates from the 30% in this research sample to the 60% usually reported in our typical clinic population produces estimates for a diagnosis of probable AD of PPV = 0.95, NPV = 0.73, and for probable and possible AD combined, PPV = 0.88, NPV = 0.81.

Five DLB pathology cases were given incorrect primary clinical diagnoses—three of possible AD and two of possible VaD. The clinical findings were characterized by a lack of spontaneous extrapyramidal features in four of the five and by the absence of visual hallucinations in three or a failure of these to persist over time in one case (Case D). We have previously commented on the importance of clinicians being prepared to diagnose DLB in the absence of parkinsonism,30 as literature review of nine recent studies finds it never to occur in up to 20% of cases.12 Review also finds visual hallucinations to be present at any point throughout the illness in only 46% (range 13 to 80%). Therefore, although both of these core features are useful “signposts” alerting clinicians to the possibility of DLB, they cannot be relied upon to detect approximately one fifth of cases. In Case E, the presence of an etiologic factor (previous intracranial surgery) possibly accounted for the atypical, fluctuating clinical presentation. He had superficial cerebral hemosiderosis due to an anterior communicating artery aneurysm and brainstem LB in this case may represent epiphenomenal LB.31

All three cases incorrectly assigned as having possible AD were additionally given secondary diagnoses of DLB, indicating that the assessors had recognized suggestive clinical features but had difficulty in correctly interpreting their significance in relation to other signs and symptoms. Neither of the two cases misdiagnosed with possible VaD was offered a secondary diagnosis of DLB, even though both met possible criteria. Additional vascular pathology was present in 9 of the 29 cases (31%) classified neuropathologically as having DLB; in 6 cases (21%), the lesions were “incidental” small, focal areas of ischemia, predominantly in white matter or basal ganglia, of the type commonly reported in the brains of cognitively unimpaired older people. They had no effect on accuracy of DLB diagnosis.

In the three remaining cases (10%), the vascular pathology was more extensive and in two was associated with a clinical misdiagnosis of possible VaD. Review of Case C shows that fluctuating, confusional episodes preceded the patient’s later stroke by at least a year and the incorrect clinical diagnosis was partly a consequence of attempting to find a unitary explanation for all symptoms, rather than opting for a “mixed pathology” category. In Case G, a combination of sudden onset, asymmetric tendon reflexes (recorded on only one occasion) and brain scan findings (CT and SPECT) indicative of cerebrovascular disease contributed to the selection of a primary vascular diagnosis. Barber et al.32 have shown that periventricular lesions and white matter hyperintensities on MRI occur in similar proportions of DLB and AD patients, both being significantly more prevalent than in aged controls and less than in VaD. Minor brain imaging abnormalities, which are often interpreted as evidence of cerebrovascular disease, should not automatically exclude a DLB diagnosis—they are consistent with it and may indicate additional vascular pathology. The decision-making process appears analogous with that for interpreting imaging findings when diagnosing AD.

In four of the five autopsy DLB cases not clinically diagnosed as such, significant cognitive fluctuation was recorded in the records, confirming that this is a key symptom in detecting the syndrome.7,9 Mega et al.5 have suggested that well-defined operationalized criteria for distinguishing the fluctuations that characterize DLB are lacking. They suggested that fluctuations of 5 or more points on the MMSE over three administrations in a 6-month period could meet this criterion, although within that particular study this criterion did not achieve satisfactory inter-rater reliability. Because our subjects were assessed only annually, we could not directly test this hypothesis. The report of the Second International Workshop on DLB33 recommends that a variety of complementary techniques be developed to improve the assessment of cognitive fluctuation. These include the use of questionnaires and diary keeping by a reliable informant, which provide a more comprehensive evaluation of different aspects of function, which may fluctuate. The administration of specific psychometric procedures, including computer-based tasks, which are sensitive to very short-term variations in attentional performance may also help to discriminate DLB from AD.34

Three previously published, retrospective chart reviews examining the validity of the Consensus criteria have already been mentioned.5,6,10 Although each used slightly different neuropathologic methods, all three approximated closely to the Consensus guidelines and technical variations are unlikely to have contributed greatly to any discrepant findings. Twenty-five percent of LB pathology cases from a community-based dementia case register in London10 had additional vascular pathology, similar to the 31% that we found. Our experience is that patients with mixed pathology pose greater problems in accurate clinical diagnosis than “pure” pathologic cases. Chart review identified only two of nine DLB autopsy cases without additional vascular pathology in the London series (sensitivity = 0.22) with no false positive clinical diagnoses (specificity = 1.0). Inspection of the data presented in that article shows that two of these nine cases did not receive a clinical diagnosis of DLB because of a >1 year history of motor PD before the onset of neuropsychiatric features. Although the Consensus statement does stipulate this arbitrary 1-year cutoff, its primary function was to distinguish marker points along the spectrum of LB disease rather than to discriminate from other pathologies. If these two motor onset cases are included, sensitivity increases to 0.44. Two further cases were excluded because of “stroke disease as evidenced by brain imaging,” although precise details of the imaging changes are not given. Minor periventricular lucencies or white matter hyperintensities would be consistent with a primary diagnosis of DLB.32 Another group6 has suggested that four clinical features have utility in discriminating DLB and AD cases: unspecified hallucinations, unspecified extrapyramidal features, fluctuating course, and rapid progression. The presence of only one of these symptoms had high sensitivity (0.91) but low specificity (0.62) and was therefore analogous with the Consensus “possible” DLB category. Two, three, or more features were associated with improved specificity, but as this reached unity, sensitivity fell to 0.43.

In the most recent study to examine these issues,35 clinical abstracts from 18 autopsy-confirmed DLB and 76 other non-LB demented cases were reviewed. Inter-rater reliability was acceptable for all key diagnostic items, ranging from 0.57 for fluctuation to 0.90 for extrapyramidal features. Consensus criteria for probable DLB had sensitivity of 0.61 and specificity of 0.84. Diagnostic accuracy improved further (0.78, 0.85) when the analysis was confined to patients with mild or moderate disease or when DLB cases also fulfilling pathologic criteria for AD were excluded (0.78, 0.81).

Several other groups of investigators have presented limited data in abstract form about clinicopathologic correlations in patients clinically diagnosed as having DLB by Consensus criteria. In the Oxford Project To Investigate Memory and Aging (OPTIMA) project,36 26 of 102 (25%) of a consecutive series of demented cases were found to have nigral LB. At study entry the sensitivity value of the Consensus criteria was 0.58; specificity, 0.89; PPV, 0.89; and NPV, 0.86. Patients clinically diagnosed with possible DLB were no more likely to have LB at autopsy than were other diagnostic groups, reflecting our own experience that possible DLB as currently defined is a nonspecific label. A discriminant function analysis including 12 items found seven factors that were able to correctly distinguish 85% of autopsy-proven DLB cases from an AD group.37 These were neuroleptic sensitivity, nonvisual hallucinations, clouding of consciousness, syncope, spontaneous parkinsonian features, fluctuations, and gender. One study38 found the Consensus criteria to be highly sensitive (0.89) but the false positive diagnosis rate was also high (71%).

This prospective evaluation of the Consensus criteria for the clinical diagnosis of DLB made few false positive diagnoses. Even when diagnostic accuracy rates are corrected to take account of the high proportion of DLB cases in this series, the estimated PPV for probable DLB in a typical dementia clinic population is 0.75 to 0.80. This is similar to the PPV of 0.75 recently reported in a study sample that had a 19% prevalence of DLB cases,35 suggesting that the core features of DLB, particularly items relating to fluctuation, can be identified when the appropriate clinical methods detailed in the Consensus article are actively applied. Seventeen percent of autopsy-confirmed cases in our sample remained undetected, even when the index of suspicion for DLB was high within a research study. Lack of extrapyramidal features and visual hallucinations were the usual reasons for misdiagnosis, which was more likely to occur in the presence of additional vascular pathology. This suggests that in routine clinical practice a significant proportion of DLB cases are continuing to go undiagnosed. Fluctuating cognition appears to be a more frequent and specific diagnostic feature than either visual hallucinations or parkinsonism. The Second International Workshop on DLB33 recommended that the Consensus criteria should continue to be used in their current format, the only modification being the addition of REM sleep behavior disorder and depression to the list of additional features “supportive” of a diagnosis of DLB. Standardized methods for identifying fluctuation need to be established and diagnostic accuracy may be additionally improved by the use of appropriate investigations. Medial temporal lobe atrophy on MRI is significantly less in DLB than in AD38 and reduced dopamine transporter activity in the striatum39 may be visualized by FP-CIT SPECT in DLB but not AD.40 Future clinicopathologic validation studies should incorporate such diagnostic investigations applied to a suitably heterogenous case mix, including a greater proportion with possible DLB and patients presenting with motor features of PD. Neuropathologic identification of DLB cases should be augmented by the use of antibodies to α-synuclein, which are emerging as the most sensitive markers of LB disease.41 A final important observation is that there is a high prevalence of mixed pathology, even in a highly selected cohort of demented subjects such as these, serving as a reminder that Occams’s razor does not always serve the clinician well. Future developments in clinical classification of the dementias should take account of this.