Validity of clinical criteria for the diagnosis of dementia with Lewy bodies

Methods.

Results.

Eighteen cases had DLB, confirmed on postmortem examination. Nine of these also had neuropathologic evidence of AD. Postmortem diagnoses in the remaining 76 cases without cortical Lewy bodies included: AD (33), AD with concomitant strokes (17), vascular dementia (19), dementia lacking distinct histology (2), ALS–dementia (1), Pick’s disease (1), argyrophilic grain dementia (1), and corticobasal degeneration (2).

Demographic characteristics are summarized in table 1. Individuals with DLB were predominantly men; individuals with non–Lewy body dementias were predominantly women (p < 0.03). Patients with DLB were 7 years younger on average at the estimated time of onset (p < 0.02) and 4 years younger at age of death (not significant). The DLB group was seen slightly more often (mean 4.7 visits) than the non-DLB group (mean 3.9 visits; not significant). The frequency of apolipoprotein e4 allele did not differ between the DLB cases and cases with other dementias (data not shown).

Table 2 summarizes the sensitivity, specificity, PPV, and NPV of various clinical criteria for distinguishing any DLB (both pure cases and cases of DLB-AD) from other causes of dementia. Although several criteria gave either excellent sensitivity or specificity, no criterion or group of criteria gave simultaneously excellent sensitivity and specificity. For example, the CDLB criteria for possible DLB (any one major criterion) had excellent sensitivity (89%) but very poor specificity (28%). Criteria for probable DLB (any two major criteria) improved specificity (84%) dramatically at the price of sensitivity (61%). The minor criteria generally had excellent specificity (87–93%) but poor sensitivity (11–39%). Requiring two major or minor criteria provided improved sensitivity (72%) but lower specificity (79%) than CDLB criteria for probable disease. Requiring three major or minor criteria provided excellent specificity (92%) at the price of poor sensitivity (50%). The PPVs (the proportion of individuals who screen positive who have disease) were all modest. The highest PPV (83%) was achieved at the price of very low sensitivity (28%). In other words, to define a group with high rates of DLB we needed criteria that excluded almost three-fourths of cases of DLB. High NPV was achieved both if CDLB criteria (85–91%) or if a combination of any major or minor criteria were used (89–92%). The criteria were most useful in excluding patients with DLB from a group.

Table 3 displays the data for 9 patients with pure DLB compared with 85 patients without pure DLB (76 with other dementias and 9 with DLB and concomitant AD). If analyses are confined to pure cases of DLB, clinical criteria do somewhat better. For example, CDLB criteria for probable DLB (two out of three major criteria) have a sensitivity of 78% and specificity of 81%. This definition still is associated with a modest PPV of 30%, indicating that only one-third of individuals who meet criteria has disease. The NPV for the CDLB criteria is excellent (94–97%).

Because extrapyramidal signs become increasingly common as severity of AD increases, we assessed the validity of CDLB criteria in cases with mild or moderate dementia (Blessed test score ≤ 20; (tables 4 and 5⇓). Diagnosis and treatment are important issues in this less severely affected group. CLDB clinical criteria for probable DLB perform better when limited to cases with mild to moderate dementia (see table 4) than when they are applied to all cases regardless of dementia severity. Limiting criteria to cases without severe dementia improves sensitivity to 67%, specificity to 82%, PPV to 47%, and NPV to 91%. In the pure DLB cases with mild to moderate disease (see table 5) there is an increment in sensitivity to 88% with little change in specificity.

In a clinical or research setting it is important to isolate pure diagnostic groups, and hence we compared the clinical criteria in the 9 cases of pure DLB and 33 cases of pure AD (table 6). The CDLB criteria for probable DLB had fair sensitivity (78%) and specificity (64%). In the analysis restricted to mild to moderate disease, the combination of any three major or minor criteria had the best combination of sensitivity (75%), specificity (93%), and PPV (75%).

We explored alternatives to the CDLB definitions in various diagnostic groups. A definition requiring any three of the major or minor criteria did well in the analyses restricted to mild to moderate disease. For the group including any DLB, sensitivity was 50%, specificity 96%, PPV 75%, and NPV 89%. For the pure DLB group with mild to moderate disease sensitivity was 75%, specificity 96%, PPV 75%, and NPV 96%. Neuroleptic sensitivity had the best PPV among all criteria in the mild to moderate pure DLB group (100%) and the group with any DLB (91%).

We also used these data to compare clinical characteristics of groups with pure DLB and DLB-AD. We first examined clinical criteria irrespective of severity of dementia. Although 16 of 18 cases with DLB met at least one major criterion, two or more criteria were much more likely in patients with pure DLB than in those with DLB-AD (p < 0.02). If analyses were confined to mild to moderate cases, the eight pure DLB had a total of 24 out of 48 possible abnormal signs or symptoms (i.e., each patient could have visual hallucinations, neuroleptic sensitivity, etc. for up to six symptoms or signs × eight patients). The four patients with DLB-AD had 6 out of 24 possible abnormal signs or symptoms (p < 0.04).

We tested whether any combination of clinical criteria could distinguish cases of pure DLB from cases of DLB-AD. Neuroleptic sensitivity occurred in six of nine patients with pure DLB, but in only one of nine patients with DLB-AD (p < 0.05). If dementia severity was not used to limit cases, no other clinical criterion or combination of criteria could distinguish between the pure and mixed DLB groups. If the analysis was confined to the 12 cases with mild to moderate dementia, the combined criteria of meeting any three major or minor criteria separated the groups. Thus none of the four DLB-AD cases met this criteria, compared with six of eight cases with pure DLB and mild to moderate dementia (p < 0.06).

Discussion.

Our study showed that if CDLB criteria were applied to the 18 cases with DLB and 76 cases with non–Lewy body dementias, the definition for possible DLB had excellent sensitivity (89%) but poor specificity (28%), and the criteria for probable DLB (two out of three major criteria) had moderate sensitivity (61%) and good specificity (84%). In this sample, using CDLB criteria, we can identify most people with DLB (using one major criterion) but only at the price of including many people without DLB. If only the 9 cases with pure DLB were compared with the 85 cases without pure DLB, CDLB criteria for probable DLB yielded a sensitivity of 78% and a specificity of 85%. PPVs remained under 50% in both diagnostic groups, indicating that most people who meet CDLB criteria do not have DLB. Accuracy was improved if analyses were confined to patients with mild or moderate dementia, though for all groups CDLB criteria for probable DLB had PPVs under 50% in this sample.

Our study shows that the proposed CDLB major criteria have high NPVs, both alone and in combination. Thus in a study of AD one can readily exclude patients with DLB. High PPVs can be achieved only if criteria are confined to cases with mild to moderate dementia. In that case, the criterion of any three major or minor criteria has a PPV of 75% and might be useful in identifying cases for studies highlighting natural history or treatment response. Limitations of this suggestion are discussed subsequently.

Two previous published papers have reported sensitivity and specificity of clinical criteria for diagnosing DLB.9,10 Our sensitivity and specificity were somewhat higher than those reported by Mega et al.9 Sixteen of the 18 patients in their study met pathologic criteria for AD. Data are not presented to determine how many of these would have been classified as DLB without concomitant AD if neocortical neuritic pathology also was required for the diagnosis of AD. Sensitivity and specificity of particular clinical features were not described. Cases in the study of Ala et al.10 were derived from a large brain-bank series. Prospective neurologic examinations were not part of the criteria for inclusion in their study. Accordingly, the frequency of signs of DLB was low in their series, and this may account for the lower sensitivity and specificity.

At the sixth International Conference on Alzheimer’s Disease and Related Disorders in July 1998, three papers examined the sensitivity and specificity of CDLB criteria. Because these data still are available only in abstract form, interpretation of these results must be viewed with caution. The Newcastle group14 analyzed data on 50 prospectively evaluated demented patients (21 with DLB) and found a sensitivity of 83% and a specificity of 92%. However, other investigators found very poor sensitivity but perfect specificity for clinical diagnostic criteria of DLB.11 Another group reported a sensitivity of 80% and specificity of 55% in 102 demented patients, nine with cortical Lewy bodies.12 If cingulate Lewy bodies fulfilled pathologic criteria for DLB, sensitivity improved to 89% and specificity to 70%. In part because of the brevity of these abstracts, it is not possible to determine how many DLB cases were of “pure” DLB and how many were DLB-AD. These authors did not report on how severity of dementia influences accuracy of diagnostic criteria.

Diagnostic accuracy has implications for treatment trials. Most clinical trials do not state explicitly whether or not patients felt to have AD with DLB were included in their study.1,2 As both DLB and AD can be associated with parkinsonism, the results of studies investigating the beneficial effects of various neuroprotective agents may be confused.2 NINCDS-ADRDA criteria26 are ambiguous as to whether or not patients with AD and DLB would meet criteria for probable or possible AD. The criteria require “the absence of other brain diseases that . . . could account for the progressive deficits in memory and cognition.”26 Nonetheless, with many investigators calling cases with mixed DLB-AD the “Lewy body variant of AD,”27 many cases with DLB are included in clinical studies of AD.1,2 Such lumping of cases of pure AD and DLB-AD may add undesirable heterogeneity to the study samples. The presence of concomitant stroke with AD (17 in our series) can compound further this heterogeneity. Such heterogeneity can be especially troubling if the therapeutic agent may have differential influences on the disease process leading to AD or DLB.

We did not examine the validity of the CDLB pathologic criteria. Unlike the Kosaka criteria,25 which utilized a quantitative approach (five or more Lewy bodies per 100× field) or the criteria proposed for the Lewy body variant of AD27 (Lewy bodies in the brainstem and cortex in addition to AD pathology), operational criteria for senile dementia of Lewy body type required the presence of “moderate numbers of LB in the limbic cortex.”28 The CDLB employs a semiquantitative scale to classify cases into brainstem-predominant, transitional, and neocortical types.8 The presence of different pathologic criteria makes strict comparisons between different studies difficult. For instance, Mega et al.9 point out that using a less restrictive definition, the specificity of the CDLB pathologic criteria improves at the expense of a marked drop in sensitivity. Pathologic diagnoses provide the means for separating pure cases from those mixed with varying amounts of Alzheimer’s pathology. In the past, we have made a distinction between the pathologic diagnoses of diffuse Lewy body disease with aging changes (diffuse Lewy body disease/pathologic aging), pure diffuse Lewy body disease, and diffuse Lewy body disease with Alzheimer’s disease.16

It may seem surprising that there was such a high prevalence of vascular dementia among the 76 demented cases without cortical Lewy bodies. There are two explanations for this finding. First, we used the relatively restrictive National Institute of Aging—Reagan institute pathologic criteria for AD. Many of these cases would have been classified as AD with incidental strokes had we used less restrictive pathologic criteria. In fact, with Khachaturian criteria, 12 of the 19 cases with vascular dementia would have been classified as AD with incidental strokes. Second, our sample was very old (mean age of all autopsied demented patients was 83.2 ± 10.5 years). Several studies have suggested that vascular dementia and also mixed dementias with minimal AD pathology and vascular lesions are most common in the oldest old.29-32

Our study has several limitations. Though the sample is substantial for a clinical–pathologic series, it is modest in size given our objectives. As a consequence, our assessment of various DLB groups (pure DLB versus DLB-AD), disease severity, and alternative case definitions must be viewed as exploratory. These definitions should be assessed in an independent sample.

We also are limited by the retrospective method of rating of various clinical criteria, which is not the best method to test diagnostic criteria. In addition, we studied the sensitivities and specificities of individual major and minor clinical criteria for DLB alone and in various combinations. Given the small size of the sample, these analyses are intended primarily to generate alternative definitions to be tested in other samples. We regard our any three major or minor clinical criteria for DLB as promising, particularly for patients with mild to moderate dementia (sensitivity 50%, specificity 96%, PPV 75%) and if the objective is to detect pure mild to moderate DLB (sensitivity 75%, specificity 96%, PPV 75%). We obtained comparable results in cases of pure mild to moderate AD (sensitivity 75%, specificity 93%, PPV 75%) as well.

Because our data collection antedated the recommendations for operational criteria for senile dementia of Lewy body type,33 we may have underestimated the frequency of fluctuations and hallucinations. The interrater reliability studies indicate that fluctuations are the most difficult to identify. We also were not able to reliably ascertain the presence of other supportive features such as syncope or transient loss of consciousness, which may have improved the diagnostic accuracy in mild to moderate disease. Because of the retrospective nature of our study, we may have underestimated the presence of clinical features in patients who were first seen with moderate to severe dementia.

The optimal clinical criteria for DLB may depend upon the objectives of the diagnostician. If the objective is to identify a sample highly likely to have DLB for a treatment trial or to observe natural history, then optimizing PPV is critical to ensure that research samples have individuals with DLB. Requiring all three major CDLB clinical criteria or perhaps a combination of three major or minor criteria in patients with early disease may be optimal. If the objective is to exclude cases with pure DLB from a treatment trial, then CDLB clinical criteria for probable DLB with a NPV of 97% are adequate for most purposes. Among the minor clinical criteria, neuroleptic sensitivity is the main feature that contributes to improved predictive value. Although a history of previous neuroleptic sensitivity may be useful, it is improper to use neuroleptic sensitivity in a prospective fashion as a “therapeutic test” to improve diagnostic efficiency. In fact there are recommendations that neuroleptics should be avoided in demented patients in whom DLB is suspected.6 This limits the practical use of this feature in the diagnostic process.

Our study shows that CDLB clinical criteria work better at identifying cases with “pure” DLB than cases with DLB-AD. Because pathologic criteria for DLB and AD are also uncertain, clinical pathologic studies are needed to define homogenous groups based on clinical and pathologic features. As criteria are refined, our understanding of DLB will evolve and clinical diagnosis will improve.

Note added in proof. Since this paper was submitted, a summary statement of the second dementia with Lewy bodies international workshop has been published.34 The report recommends that the CLBD clinical criteria continue to be used in their current format with research efforts focusing on increasing sensitivity of case detection.