Our data show that the ACE-R, and in particular its total score, verbal fluency and visuospatial subscores, distinguish between PD and PSP, and to a lesser extent between PD and CBD. The differentiation between PD and PSP using verbal fluency alone (specificity 0.92, sensitivity 0.87) outperforms previous reports of diagnostic differentiation achieved using imaging methods alone.^{12 13} In addition, our data suggest that the ACE-R may be helpful for tracking disease progression over time, particularly in CBD. These findings suggest that in the appropriate clinical context, the ACE-R may provide useful information in the diagnosis and tracking of parkinsonian disorders.

The good PPVs and NPVs, low cost and simplicity of these cognitive biomarkers makes them especially useful when the cost and expertise required for routine clinical MRI or positron emission tomography is prohibitive. This includes many regions of the world where imaging is not available.²⁰ Cognitive tests have the potential of being safe, inexpensive and low-tech additions to the diagnostic armoury that can be used in a range of clinical settings—primary, secondary or tertiary care and research studies. The ACE-R has even been validated in a number of languages, although our findings would require replication in different populations to properly assess generalisation.^{17 21 22}

Different parts of the cognitive assessment had different value in the differentiation of PD, PSP and CBD. We focused on the ACE-R rather than the MMSE, as it was superior in differentiating disorders (see ROC curves, figure 2) and was more sensitive to decline over time. The role of verbal fluency was particularly striking. Fluency deficits have been described in PD,²³ PSP^{5 24} and CBD.²⁵ The decline in fluency is not entirely due to bradykinesia, as fluency requires very little motor function. Interestingly, patients with PSP or CBD may generate a small number of low-frequency words (eg, ‘peramubulator’) rather than high-frequency words (eg, ‘put’, ‘people’). This is supported by evidence from functional imaging suggesting that verbal fluency is associated with activation of a diffuse cortical network²⁶ but with an emphasis on Brodmann area 45 for lexical retrieval and Brodmann area 46 in phonological processing during word generation.²⁷ Patients with brain lesions in the left dorsolateral frontal lobe were found to have impaired verbal fluency if there were additional or solitary striatal lesions.²⁸ Neuropathology studies confirm that both of these sites are affected in PSP and CBD.^{29 30} Verbal fluency impairment has previously been reported in the early stages of PD,³ and is consistent with the Braak staging hypothesis that predicts frontal lobe cortical pathology.³¹

Changes in higher visual processing are present in all three disorders, but take different forms in each. Patients with PD may experience hallucinations or misperceptions—usually later in the disease process—that may be related to medication side effects or impaired attentional processing.³² Patients with CBD and, to a lesser extent, those with PSP perform poorly on visuospatial tasks. Patients with PSP are impaired more with ‘object decision’ tasks, thought to be related to impairment in the ventral stream of visuospatial processing, whereas patients with CBD are typically worse on spatial location tasks, thought to relate to the parietal atrophy disrupting the dorsal stream of visuospatial processing.³³ These previous findings are in broad agreement with our results. The ACE-R visuospatial tests rely more on spatial abilities than object recognition and is therefore likely to be more sensitive to the changes seen in CBD and PSP, but is less likely to detect the misperceptions more commonly experienced in PD.

Statistically significant group differences in test scores do not necessarily indicate clinically meaningful differences in test scores, especially in large studies. For many of our results, we suggest that the differences between groups and over time are both statistically significant and clinically meaningful. MMSE scores declined by 2.7 points in the CBD group, which is close to the SD of 2.8 found in a large dementia population,³⁴ suggesting this change is clinically significant. Large absolute differences were seen in PSP and CBD groups for total ACE-R score (PSP 6.0, CBD 10.9, of maximum score 100) and visuospatial subscores (PSP 2.5, CBD 3.8, of maximum 16 points). Although we do not have functional rating scales or measures of activities of daily living, we suggest that the magnitudes of these changes are likely to be clinically significant. In contrast, the PD group showed much smaller changes in total ACE-R score (1.9 points) and the visuospatial subscore (0.5 points), which lie within the range of test variation (inter-rater variance and test–retest) and are less likely to be clinically meaningful. In the context of PD, it has recently been demonstrated that the ACE-R reflects functional changes; the authors could identify the difference between cognitively well patients with PD and those with PD and mild cognitive impairment using a cut-off of 89 (specificity of 84%, sensitivity 69%).³⁵ Our findings would support the use of cognitive measures in diagnostic criteria, although neuropathological correlation would be required to confirm this.

In comparing groups in this study, we are reliant on clinical diagnostic criteria, which may not always reflect underlying pathology. Clinical diagnostic criteria for PSP¹⁸ tend to be specific, with a diagnostic accuracy of 92% in patients meeting criteria.³⁶ They may be insensitive to PSP pathology in patients with a PD-like syndrome³⁷ or no clinical disease.³⁸ In our cohorts, many patients with PSP had consented to brain donation in a regional brain bank, providing neuropathological confirmation in nine out of nine cases, in keeping with the high clinicopathological correlations in other centres.

There are several current clinical diagnostic criteria for CBD, recently reviewed by Mathew et al.⁸ They applied clinical criteria in early CBD, finding sensitivities of 23.5–67.5%, increasing to 87.5–92.5% in later stage disease depending on the criteria applied. Diagnostic uncertainty is further reflected in the ante-mortem misclassification of CBD from post-mortem studies.^{39 40} Our findings therefore support the use of cognitive measures in diagnostic criteria for PD, PSP and CBD, although further neuropathological correlation would be helpful to investigate the link between cognition and histological disease changes in more detail, and evaluating cognitive differentiation of atypical or intermediate syndromes which combine features of PD, PSP and CBD.

In principle, one could apply more detailed neuropsychological tests. However, an advantage of the ACE-R is ease of use, requiring short training for those administering it. Usually a full test will take around 10 min in cognitively well participants without akinesia, although it can be up to half an hour in severely affected patients with PSP and CBD. Verbal fluency is an even briefer test, taking just 2 min of testing time. The ACE-R and verbal fluency could be assessed routinely in a general neurology or movement disorders clinic and its utility has been shown in this study.

There are however several limitations to our study. The number of patients and follow-up time was limited. Furthermore, the duration of symptoms was shorter in the PD group, although correcting for this by comparing baseline PSP and CBD scores with follow-up PD scores led to the same pattern of results. Delayed diagnosis in PSP and CBD is common, with patients often receiving an alternative diagnosis for 1–2 years, therefore recruiting patients with PSP and CBD at an earlier stage is challenging. We cannot exclude the possibility that cognitive profiles would be different and between-group differences smaller at an earlier stage in the disease processes. Post-mortem confirmation of diagnosis was not available in all patients and so it is possible that for a small number of patients the diagnosis may be inaccurate. Patients with PD were seen for follow-up slightly later than the other two groups. However, correcting for this difference would augment rather than diminish our findings given the largest differences over time were seen in the CBD group who had the shortest mean follow-up.

Low scores may arise for different reasons in the three groups. For example, it is possible that apraxia or other features of a movement disorder impaired task performance differentially between groups, particularly on visuospatial drawing tasks and in patients with CBD. However, the majority of subjects were relatively early in the disease course, able to hold a pencil and draw sufficiently to reveal the cognitive elements of the test.

There are also caveats to our use of the UPDRS III motor subscale to measure motor severity. This scale does not capture all signs in the CBD phenotype, omitting apraxia, alien limb phenomena, dysarthrophonia and mirror movements. Although the principal axial and appendicular motor signs of PSP are included in the UPDRS III, it should be borne in mind that the test was developed for use in PD and that it may not be adequately balanced to sensitively grade the progressive severity of PSP or CBD. The use of the UPDRS should therefore be seen as descriptive and not diagnostic of our study populations.