Clinical impact of [18F]flutemetamol PET among memory clinic patients with an unclear diagnosis

Purpose To investigate the impact of amyloid PET with [18F]flutemetamol on diagnosis and treatment management in a cohort of patients attending a tertiary memory clinic in whom, despite extensive cognitive assessment including neuropsychological testing, structural imaging, CSF biomarker analysis and in some cases [18F]FDG PET, the diagnosis remained unclear. Methods The study population consisted of 207 patients with a clinical diagnosis prior to [18F]flutemetamol PET including mild cognitive impairment (MCI; n = 131), Alzheimer’s disease (AD; n = 41), non-AD (n = 10), dementia not otherwise specified (dementia NOS; n = 20) and subjective cognitive decline (SCD; n = 5). Results Amyloid positivity was found in 53% of MCI, 68% of AD, 20% of non-AD, 20% of dementia NOS, and 60% of SCD patients. [18F]Flutemetamol PET led, overall, to a change in diagnosis in 92 of the 207 patients (44%). A high percentage of patients with a change in diagnosis was observed in the MCI group (n = 67, 51%) and in the dementia NOS group (n = 11; 55%), followed by the non-AD and AD (30% and 20%, respectively). A significant increase in cholinesterase inhibitor treatment was observed after [18F]flutemetamol PET (+218%, 34 patients before and 108 patients after). Conclusion The present study lends support to the clinical value of amyloid PET in patients with an uncertain diagnosis in the tertiary memory clinic setting.


Introduction
Reliable biomarkers are a prerequisite for the early and accurate diagnosis of Alzheimer's disease (AD) and other dementia disorders. Positron emission tomography (PET) imaging using the metabolic tracer 2-deoxy-2-[ 18 F]fluoro-D-glucose ([ 18 F]FDG) is a well-established method for the evaluation of functional changes in the brain of patients with dementia disorders, and has been shown to be useful in discriminating the typical pattern of hypometabolism seen in AD from those seen in other dementia disorders [1]. [ 18 F]FDG PET shows low accuracy, however, in identifying patients with mild cognitive impairment (MCI) who will convert to different forms of dementia [1,2]. According to the recent International Working Group 2 criteria [3], [ 18 F]FDG PET should be considered a marker of disease progression, in contrast to amyloid-β (amyloid) PET, which is considered a pathophysiological marker.
Since the publication of the first in vivo PET study showing selective imaging of amyloid plaques in AD [4], amyloid PET has proven to be instrumental as a research tool [5]. Beyond this application, amyloid PET holds great potential as a diagnostic aid because of its ability to directly detect a core neuropathological feature of AD, a disease diagnosed with only moderate accuracy, even by expert clinicians [6]. Antoine Leuzy and Irina Savitcheva contributed equally to this work.
On the basis of convincing findings from phase I-III studies, several fluorine-18 amyloid tracers, including [ 18 F]florbetapir ( A m y v i d ® ) , [ 1 8 F ] f l o r b e t a b e n (N eu r ac e q ®) a n d [ 18 F]flutemetamol (Vizamyl®), were commercially developed and approved for use in the exclusion of AD as an underlying cause of cognitive impairment. Although interest in their incorporation into daily practice has been shown to be high among dementia specialists [7] relatively few of the studies addressing their clinical impact [8,9] have been prospective in nature or performed in the tertiary setting [10][11][12][13][14][15], with only a subset including patients with MCI in sizeable numbers [11,[15][16][17]. Moreover, few studies have included CSF biomarkers [10,13,18]. The objective of the present study was thus to investigate the impact of [ 18 F]flutemetamol PET imaging, in terms of changes in diagnosis and treatment, in patients whose diagnosis remained unclear following clinical assessment at a tertiary memory clinic.

Participants
The study population consisted of 207 patients attending the Clinic for Cognitive Disorders, Theme Aging, Karolinska University Hospital, Stockholm, Sweden. Patients were seen between 2014 and 2018 and had been mainly referred by primary care physicians (GPs), but also from different hospital clinics, owing to different forms of cognitive problems. Some patients were referred from other memory clinics in Sweden to seek a second opinion. Many patients were relatively young (mean age <65 years). Most patients referred by GPs had undergone cognitive testing (e.g. Mini-Mental State Examination, MMSE), structural imaging (CT, or in a few patients, MRI) and blood analysis, while patients referred from other clinics had often undergone less thorough assessments.
At their first visit, patients underwent physical, neurological, psychiatric and cognitive assessments, and a detailed medical history was recorded. Most patients were accompanied by a close relative or friend. Patients were referred for neuropsychological testing, CT/MR imaging and CSF sampling, and some underwent apolipoprotein E (APOE) genotyping, electroencephalography, speech/language testing and [ 18 F]FDG PET. Diagnoses were based on a consensus meeting between specialists in cognitive disorders, clinical neuropsychologists and specialist nurses. Diagnostic categories included MCI [19,20], AD [21,22], dementia of unclear aetiology (not otherwise specified, dementia NOS) [22], dementia due to a non-AD disorder, including dementia with Lewy bodies (DLB) [23], frontotemporal dementia (FTD) [24] and vascular dementia [25], and subjective cognitive decline (SCD) [26].
Patients were referred for amyloid imaging with [ 18 F]flutemetamol PET because of an uncertain diagnosis despite extensive cognitive and biomarker-based assessments, and [ 18 F]flutemetamol PET was generally performed either directly after initial diagnosis or later, during clinical followup. Following the [ 18 F]flutemetamol PET scan, patients revisited the Clinic for Cognitive Disorders and were informed of their results, any possible changes in diagnosis and the management plan by a specialist in dementia disorders. The diagnostic categories before and after [ 18 F]flutemetamol PET were the same, except for the addition of prodromal AD in a subset of patients with MCI [3].

Neuropsychological assessments
Most patients completed a large battery of neuropsychological tests covering different cognitive domains [27]. These included the MMSE as well as components of the Wechsler Adult Intelligence Scale, Revised (WAIS-R; information and similarities, logical memory, block design and digit symbol), figure classification, subtest of the Synonyms Reasoning Block Test (SRB2), Rey Auditory Verbal Learning Test (RAVLT), copying and memory subtests of the Rey-Osterrieth Complex Figure Test (ROCFT), parts A and B of the Trail Making Test (TMT), and/or the Verbal Fluency Test (FAS).

Structural imaging
Structural CT or T1-weighted MR imaging was performed at various radiology departments in Stockholm using different platforms and protocols. Cerebral atrophy and white matter lesions were assessed clinically by experienced neuroradiologists at the Department of Radiology, Karolinska University Hospital, according to standard visual rating scales. Atrophy of the medial temporal lobe was evaluated using the medial temporal atrophy (MTA) scale [28]. Overall cortical atrophy was assessed using the global cortical atrophy (GCA) scale [29]. White matter hyperintensities were scored using the Fazekas scale [30].

CSF biomarkers
CSF samples were collected via lumbar puncture from 152 out of the 207 patients (73%) under nonfasting conditions as part of routine memory assessment. The lack of CSF biomarkers in 55 patients was due to the use of anticoagulants in 18 patients, spinal stenosis or related problems in 6, refusal to undergo CSF sampling in 12, and technical problems in 19. The CSF samples were routinely analysed at the Clinical Neurochemistry Laboratory, Sahlgrenska University Hospital, Mölndal, Sweden, for Aβ 1-42 , t-tau and p-tau, using commercially available ELISAs (INNOTEST; Fujirebio, Ghent, Belgium). Internal cut-off values of 550 ng/L for Aβ  , 400 ng/L for t-tau and 80 ng/L for p-tau were used.  According to clinical routine assessment, summation images were visually analysed, supported by a semiquantitative analysis in the form of standardized uptake value ratios (SUVR) based on automatically generated regions of interest (ROI; cortical and subcortical regions, normalized to whole brain) and voxelwise Z-score stereotactic surface projection images (Siemens syngo.via software). Metabolic patterns were classified as typical of AD, as suggestive but not typical of AD, as consistent with other non-AD neurodegenerative diseases (e.g. FTD or DLB) or as nonspecific in the case of minor patchy focal changes.  [31]. In cases of disagreement between raters, a consensus was achieved between readers during the re-read session.

[ 18 F]Flutemetamol PET
In addition to visual reads, semiquantitative analysis of [ 18 F]flutemetamol uptake was performed using an automated ROI-based approach (Hermes Medical Solutions) [32], with SUVR calculated using an isocortical composite ROI, comprising brain regions typically associated with high amyloid load in AD (frontal, lateral temporal, cingulate and parietal cortices), using the pons as the reference region. Amyloid-positivity was defined using an a priori SUVR cut-off value of 0.60, based on separation from cognitively normal controls [33].

Statistical analysis
Characteristics were compared between diagnostic groups using Kruskal Wallis analysis of variance (ANOVA) and the Wilcoxon signed ranks test for continuous variables. Categorical variables were assessed using Fisher's exact test. A g r e e m e n t b e t w e e n v i s u a l a n d S U V R -b a s e d [ 18 F]flutemetamol classifications, as well as interrater agreement for visual readings, were assessed using percentage agreement and Cohen's kappa. The proportions of patients showing changes in diagnosis and drug treatment were assessed using a one-sample proportions test. All statistical tests were performed using R v. 3.3.2 (The R Foundation for Statistical Computing), with two-sided p values <0.05 considered to indicate significance.

Results
The demographic, clinical and biomarker findings in the 207 patients in whom [ 18 F]flutemetamol PET was performed due to diagnostic uncertainty are presented in Table 1. Most patients received an initial diagnosis of MCI (131, 63%), followed by AD (41, 20%), dementia NOS (20, 10%), non-AD (10, 5%) and SCD (5, 2%). Figure 1 shows the structural imagingbased ratings for atrophy and white matter changes in the different patient groups before amyloid PET. MTA scores of 0 or 1, indicating no or minimal atrophy, respectively, were predominant across the MCI, AD, dementia NOS and SCD groups. In contrast, MTA scores indicating mild atrophy were noted in half of the patients with non-AD disorders. In terms of GCA and white matter lesions, most patients showed mild changes. [ 18 F]FDG PET imaging was performed in 78 of the 207 patients (38%; Table 1, Fig. 1d). A metabolic pattern suggestive but not typical of AD was the most common finding in those with a diagnosis of MCI or AD prior to [ 18 F]flutemetamol PET (16 patients, 36%, and 9 patients, 56%, respectively); in the remaining groups, patterns not typical of AD or suggestive of a non-AD disorder were predominant (non-AD, 4 patients, 100%; dementia NOS, 10 patients, 91%), with only minor patchy changes seen in patients with SCD. Similar findings were obtained using the diagnoses obtained after [ 18 F]flutemetamol PET, with only 11 AD patients (14%) showing a typical metabolic pattern.
CSF was sampled in 152 patients according to the Swedish clinical practice guidelines for memory assessment at specialist memory clinics. Overall, 103 patients (68%) showed abnormal CSF biomarkers: 24% showed only abnormal Aβ 1-42 , 21% showed abnormal Aβ 1-42 in combination with elevated ttau or p-tau, and 32% showed negative CSF; a further 23% showed only abnormal tau. The main reasons for performing amyloid PET were a clinical suspicion of AD in combination with either a negative or an unclear (i.e. isolated positive or borderline Aβ 1-42 or tau) CSF profile (117 patients, 57%) or   Cohen's kappa = 0.92). In 37 visual readings, however, one of the two readers had some difficulty in defining the [ 18 F]flutemetamol PET scan as positive or negative; this finding was similarly distributed between experienced and less experienced readers (20 and 17 readings, respectively). In four scans, both readers were uncertain. Overall, some uncertainties were observed in 11% of readings. Agreement between visual and semiquantitative assessment approaches was achieved in all cases. Using both visual and SUVR-based classifications, a higher proportion of patients were rated as [ 18 F]flutemetamolpositive in the preamyloid PET MCI and AD patients than in the dementia NOS patients (p < 0.01), and in the AD patients as compared with the non-AD patients (p < 0.05).
On visual evaluation of the [ 18 F]flutemetamol PET scans, amyloid positivity was found in 69 of 131 patients (53%) with an initial diagnosis of MCI, in 28 of 41 (68%) with AD, in 2 of 10 (20%) with a non-AD disorder, in 4 of 20 (20%) with dementia NOS and in 3 of 5 (60%) with SCD (Table 1). Amyloid status in the diagnostic groups before and after [ 18 F]flutemetamol PET as well as changes in diagnosis are summarized in Table 3

Discussion
The rapid development of molecular imaging techniques has enabled the in vivo study of the pathophysiology underlying different neurodegenerative diseases, including AD. Amyloid imaging has reached clinical use in memory clinics. However, most research studies published to date have included selected research populations. The present study included an unselected cohort of 207 patients, in whom extensive neuropsychological and, in various subsets, biomarker-based assessments could not provide a sufficiently certain clinical diagnosis. The main reasons for amyloid PET in our cohort were the clinical suspicion of AD accompanied by unclear or negative CSF findings, the absence of CSF biomarkers (due to contraindications, patient refusal or technical difficulties) and an ambiguous clinical presentation in the context of a CSF profile indicative of AD (low Aβ 1-42 and elevated tau).
Despite the fact that [ 18 F]FDG PET is a well-established diagnostic tool in the work-up of dementia disorder patients, it did not provide an adequate differential diagnosis in the present cohort. We estimate that this cohort of patients, requiring complementary amyloid PET after an extensive clinical workup including neuropsychological testing, MRI, CSF biomarker analysis and other additional examinations, represents approximately 10% of all new referrals to our academic memory clinic (approximately 500 to 600 per year).
In this study, we assessed the incremental value of amyloid PET in the work-up of patients being followed in a tertiary specialist setting due to cognitive impairment. The largest patient group was MCI patients of whom 69 (53%) were Overall, amyloid PET led to a significant change in diagnosis in 44% of patients and an increase in the use of AD drug treatment, from 34 to 109 of 207 patients (16% to 53%). On the basis of our study design, in which amyloid PET results were used to revise initial diagnoses and treatment plans, our findings provide an estimate of the incremental value of this type of investigation in the clinical setting. Thus, this study fulfils the requirements of a phase 4 study as defined by the five-phase biomarker validation framework recently introduced to the field [34,35]. This type of study is vital to assess the utility of amyloid PET and to accelerate the development of evidence-based guidelines for its clinical use [36].
In the present study, the concordance rates observed between [ 18 F]flutemetamol PET and CSF Aβ 1-42 were lower than those reported to date from the ADNI (Alzheimer's Disease Neuroimaging Initiative) and the Swedish BioFINDER (Biomarkers for Identifying Neurodegenerative Disorders Early and Reliably) studies [37][38][39][40][41], in which the composition of the cohorts largely resembles that found in clinical trials. The difference in observations is due to the fact that in the present study only patients with an unclear diagnosis following extensive assessment were recruited for amyloid imaging. Although our study, by design, was predisposed to discordance, it draws attention to potential complications in implementing the recent AD biomarker classification scheme [42]: based on categorical and biomarker analyses, this framework assumes that measures of amyloid based on CSF analysis and PET are interchangeable. Although disagreement between imaging and CSF analysis may be a reflection of how cut-off values are established-with European laboratories generally using Aβ 1-42 cut-off values that are too low [18]the use of a more lenient cut-off value (647 ng/L) derived from a recent study using the same CSF INNOTEST assay [40] resulted in the same overall level of concordance; though an increase in concordant-positive subjects was seen, this was accompanied by an increase in the number of subjects showing isolated CSF positivity. There is evidence, however, to suggest that higher cut-off values for Aβ 1-42 may prove appropriate. Further studies addressing this, including comparison with Aβ 1-42 /tau ratios, are warranted. While many subjects may show concordance in the long term [42], more work is required to integrate and compare CSF analysis and amyloid PET so as to develop more refined guidelines for how these biomarkers are to be used and interpreted in the diagnostic work-up of patients with dementia disorders [43], including considerations related to differences in amyloid processing and neurodegeneration seen across atypical forms of AD [44] and the value of Aβ 1-42 ratios with shorter isoforms [45,46] and tau/Aβ 1-42 [47].
The percentage of patients with a change in diagnosis observed in this study (44%) was higher than the average of those found in previous studies [48], although somewhat lower than in other studies [10,11]. This most likely reflects differences in the composition of the study cohorts; in the present study, diagnostic uncertainty remained despite extensive neuropsychological testing, imaging and biomarker investigations. Our study cohort therefore most likely included a higher proportion of patients with atypical disease, an assertion supported by the number of MCI and dementia NOS patients with an unchanged diagnosis. We consider, however, that the most important finding was the high percentage of amyloid-positive MCI patients who received a change in diagnosis to prodromal AD or AD, and thus began treatment with ChEIs. Although currently not included in the appropriate use criteria for amyloid PET [49], patients with SCD are increasingly discussed as an at-risk population given the association between SCD and cognitive decline in the context of biomarker evidence for AD [50,51]. Indeed, our findings that most such patients showed abnormal CSF findings and clinical progression at follow-up suggest that this group is important in the context of best practice guidelines.
In our cohort, amyloid PET led to significant changes in the management of patients, including treatment with ChEIs, with a more than threefold increase in the number of patients receiving ChEI treatment. This increase was primarily accounted for by amyloid-positive patients with a diagnosis after [ 18 F]flutemetamol PET of prodromal AD/ AD. Similarly, amyloid-positive patients with a diagnosis after [ 18 F]flutemetamol PET of AD comprised the majority of those receiving ChEIs prior to undergoing amyloid PET. The small number of [ 18 F]flutemetamol-positive MCI and non-AD (DLB) patients receiving cholinergic drugs probably reflects varying adherence by physicians to the diagnostic codes (i.e. MCI, amyloid-positive, versus prodromal AD, a research term not currently recognized in dementia care guidelines) and evidence suggesting that this drug class is of benefit in DLB [52]. Among the few amyloid-negative patients, treatment, whether initiated prior to or after [ 18 F]flutemetamol PET, was related to CSF Aβ 1-42 positivity (seen in close to half of such patients) and/or physicianspecific treatment beliefs.
The present study underlines the value of amyloid PET when CSF biomarkers are not consistent or when CSF sampling is not possible due to the use of anti-coagulants, other medical reasons, or patient refusal of lumbar puncture. In this population of patients it has also been demonstrated that [ 18 F]FDG PET is not always useful, especially in those with MCI. A negative amyloid PET study is important information for an MCI patient and often results in a decease in the use of resources due to fewer ancillary investigations. In patients with aetiologically unclear dementia or non-AD, amyloid PET could provide important diagnostic information, as was illustrated in the present study with a 55% change in diagnosis among dementia NOS patients. Although the fact that the generally rather extensive investigations undergone by the patients in this study help to establish the clinical usefulness of [ 18 F]flutemetamol PET, it must also be underlined that our findings may not be generalizable to older patients or those attending less specialized clinical centres. Although incorporating costeffectiveness parameters associated with amyloid imaging was beyond the scope of the present investigation, further studies addressing this, including the optimal stage (early versus late) for the use of amyloid PET in the diagnostic work-up of patients, are crucial.

Conclusion
In summary, our findings indicate that amyloid PET imaging with [ 18 F]flutemetamol had a significant impact in terms of change of diagnosis, management and drug treatment when added to the work-up of patients with an uncertain diagnosis followed in the setting of a tertiary memory clinical. These findings highlight the clinical value of amyloid PET in patients with cognitive impairment of unclear aetiology. Funding This study was made possible by financial support from Vinnova, the Regional Agreement on Medical Training and Clinical Research (ALF) of Stockholm County Council, the Swedish Research Council (05817, 02695, 06086), the Swedish Brain Foundation, the Swedish Alzheimer Foundation, the KI Foundations, and the Swedish Foundation for Strategic Research (SSF).

Compliance with ethical standards
Conflicts of interest J.L. is an employee of Hermes Medical Solutions, Stockholm, Sweden. All other authors declare no conflicts of interest.
Ethical approval All procedures performed in studies involving human participants were in accordance with the ethical standards of the institutional and/or national research committee and with the principles of the 1964 Declaration of Helsinki and its later amendments or comparable ethical standards.
Informed consent Informed consent was obtained from all individual participants included in the study.