Clinical impact of amyloid PET using ¹⁸F-florbetapir in patients with cognitive impairment and suspected Alzheimer’s disease: a multicenter study

Amyloid positron emission tomography (PET) can reliably detect senile plaques comprising amyloid β peptides, one of the hallmarks of Alzheimer’s disease (AD), and fluorinated ligands are approved for clinical use in several countries. In Japan, one of the approved tracers for amyloid PET imaging is ¹⁸F-fluorobetapir [1, 2]. This radiopharmaceutical, delivered as a final product to a clinical facility, has proven efficacy in the visualization of amyloid β plaques in the brains of patients with cognitive impairment and suspected AD (https://​www.​info.​pmda.​go.​jp/​go/​pack/​43004A2A1029_​2_​01/​). However, the diagnostic impact and clinical utility of amyloid PET imaging are still under investigation. Although negative findings on amyloid PET can almost entirely rule out AD [3], the prevalences of positive scans in one meta-analysis were 88% in patients with AD, 51% in patients with dementia with Lewy bodies, 30% in patients with cerebrovascular disease, 12% in patients with frontotemporal dementia, 38% in patients with corticobasal degeneration, and 24% in healthy elderly individuals serving as controls [4]. In contrast, in other meta-analyses of the clinical impact of amyloid PET, the overall rates of changes in diagnosis and patient management after amyloid PET varied widely from 19 to 55% [5‐13] and from 37 to 87% [8, 9, 11‐16], respectively, from study to study. Consequently, due to the lack of definitive evidence supporting its clinical impact, amyloid PET is still not reimbursed by the Japanese health insurance system.

Visual interpretation is utilized to determine qualitatively if amyloid PET is positive or negative when it is employed in clinical practice. Equivocal results are unavoidable in this binary classification and cause inter-rater variability in visual interpretation [17]. The addition of quantitative analysis to visual interpretation has thus been suggested [18]. The standardized uptake value ratio (SUVR) has been extensively employed in the quantitative study of amyloid PET. Additionally, the Centiloid (CL) scale has recently been adopted [19, 20], which harmonizes the quantitative amyloid imaging measurements by standardizing the results of each analytical method or PET ligand.

The aim of this multicenter study was to evaluate the diagnostic impact and clinical utility in patient management of amyloid PET using ¹⁸F-florbetapir in patients with cognitive impairment and suspected AD. We also aimed to determine the cutoffs for amyloid positivity for the SUVR and CL in quantitative analysis by investigating the agreement between the quantitative and visual assessments of ¹⁸F-florbetapir PET.

No adverse events were observed after the administration of the tracers or immediately after the PET scan in all subjects.

The mean ± SD MMSE score of the 99 patients was 24.6 ± 3.2 (range 20–30). No significant differences were found in demographic characteristics between the 54 patients with a pre-scan AD diagnosis and the 45 patients with a pre-scan non-AD diagnosis (Table 2). The patients with a pre-scan AD diagnosis showed significantly lower MMSE scores compared with those with a pre-scan non-AD diagnosis (p < 0.001).

Visual interpretation obtained 48 positive and 51 negative PET scans. The prevalence of amyloid-positive scans was not significantly different (χ² = 0.539, p = 0.462) between patients with a pre-scan AD diagnosis (29 of 54, 53.7%) and those with a non-AD diagnosis (19 of 45, 42.2%). Amyloid PET results led to changes in the AD and non-AD diagnoses in 39 of 99 patients (39.3%), with significantly higher rates in patients with pre-scan AD (26 of 54, 48.1%) than in patients with non-AD diagnosis (13 of 45, 28.8%) (χ² = 5.334, p = 0.0209). Details of the pre-scan and post-scan diagnostic changes are shown in Fig. 1. The diagnostic confidence of AD significantly increased for patients with an unchanged diagnosis of AD (Δ = 22.7% ± 13.2%, p < 0.0001) and for patients whose diagnosis changed from non-AD to AD (Δ = 46.1% ± 19.4%, p < 0.0001) after the disclosure of the amyloid PET results. Meanwhile, the diagnostic confidence of AD significantly decreased in patients whose diagnosis changed from AD to non-AD (Δ =  − 53.2% ± 13.5%, p < 0.0001) and in patients with an unchanged diagnosis of non-AD (Δ =  − 13.2% ± 30.2%, p = 0.0094) after amyloid PET (Table 3). Overall, 41 of the 48 patients (85.4%) with amyloid-positive results received a post-scan diagnosis of AD, and the remaining seven patients with amyloid-positive results received a post-scan diagnosis of MCI in six patients and primary progressive aphasia in one patient. All of the 51 patients with amyloid-negative results received a post-scan diagnosis of non-AD.

Amyloid PET results led to at least one change in the patient management plan in 42 of 99 patients (42%), without significant differences between patients with pre-scan AD and those with non-AD diagnosis (χ² = 0.001, p = 0.970, Table 4). Amyloid PET results thereafter led to changes in medication, prescription dosage, additional diagnostic tests, and care planning in 20 (20%), 5 (5%), 4 (4%), and 25 (25%) patients, without significant differences in medication (χ² = 0.301, p = 0.583), additional diagnostic tests (χ² = 3.474, p = 0.0624), and care planning (χ² = 2.854, p = 0.091) but with a significant difference in prescription dosage (χ² = 4.388, p = 0.0362) between patients with pre-scan AD and those with a non-AD diagnosis.

In a visual interpretation of the positive or negative findings of amyloid PET, four readers completely agreed in 78 of 99 scans of the patients (79%). The Cohen’s kappa agreement between two of each of the four readers ranged from 0.718 to 0.778 (0.743 ± 0.026). SUVR_40–60 values were 1.00 ± 0.05, 1.04 ± 0.08, and 1.41 ± 0.15 for scans of young healthy controls, visually negative scans, and visually positive scans, respectively, whereas CL_{40_60} scales were − 7.6 ± 8.4, − 0.7 ± 15.3, and 66.2 ± 24.9, respectively (Fig. 2). Significant differences in the SUVR_40–60 and CL_40–60 scales were observed between visually positive scans and visually negative scans and between visually positive scans and the scans of young healthy controls (p < 0.0001). There were no significant differences in these values between the scans of young healthy controls and visually negative scans (p > 0.05).

ROC analysis determined the best agreement of quantitative assessments and visual interpretation of ¹⁸F-florbetapir PET scans obtained 40–60-min post-injection to have an area under the curve of 0.993 at an SUVR of 1.19 and CL of 25.9. If visual interpretation was considered the standard of truth, quantitative assessment demonstrated 97.9% sensitivity, 94.1% specificity, and 95.9% accuracy. Using these cut-off values, there was strong agreement between them (Cohen’s kappa = 0.92). The SUVR and CL values of the four discordant cases between quantitative assessments and visual interpretation ranged from 1.15 to 1.24 and from 19.2 to 35.3, respectively. These discordant cases were classified into two patterns. Diffuse elevation of cortical activity was regarded as visually amyloid-negative despite a relatively high SUVR or CL in three cases. In contrast, mildly focal elevation of cortical activity in two areas was regarded as visually amyloid-positive despite a relatively low SUVR or CL in one case.

The SUVR_40–50, SUVR_50–60, and SUVR_40–60 of all 99 patients and 20 young healthy controls were 1.17 ± 0.22, 1.19 ± 0.23, and 1.18 ± 0.22, respectively, while the CL_40–50, CL_50–60, and CL_40–60 scales of all subjects were 23.6 ± 37.8, 26.7 ± 40.4, and 25.1 ± 38.9, respectively. In a Bland–Altman plot, there were significant differences among the SUVR_40–50, SUVR_50–60, and SUVR_40–60 and among the CL_40–50, CL_50–60, and CL_40–60 scales, while Spearman correlation analysis identified a significant association between the difference among the SUVR_40–50, SUVR_50–60, and SUVR_40–60 and SUVR load and among the CL_40–50, CL_50–60, and CL_40–60 scales and CL load (all p < 0.001, Fig. 3a–f). Representative PET images obtained 40–50 min, 50–60 min, and 40–60 min post-injection are shown in Fig. 4 along with their respective CL scales.

The current multicenter study examined the clinical impact of amyloid PET on the diagnosis and management of cognitively impaired patients with probable AD and of those with possible AD, but other disease was more likely. This is the first report of a clinical impact study of amyloid PET using ¹⁸F-florbetapir in Japan. The details of the diagnosis changed by amyloid PET have never been reported before. Amyloid PET results changed the etiologic diagnosis of AD or non-AD in 39.3% of all patients. The change rates were significantly higher for pre-scan AD diagnosis (48.1%) than for pre-scan non-AD diagnosis (28.8%). These change rates are comparable to those in previous multicenter studies.

Grundman et al. [5] reported changes in diagnosis after disclosure of the PET results using ¹⁸F-florbetapir in 125 of 229 patients (54.6%) as well as in 37.2% of patients with a pre-scan AD diagnosis and 61.9% of those with a pre-scan non-AD diagnosis. Another multicenter study using ¹⁸F-florbetapir [11] demonstrated changes in diagnosis after disclosure of the PET results in 62 of 228 patients (27.2%) as well as in 27.9% of patients with a pre-scan AD diagnosis and 25.4% of those with a pre-scan non-AD diagnosis. In the large-scale IDEAS study [13], amyloid PET results led to changes in the etiologic diagnosis from AD to non-AD in 2869 of 11,409 patients (25.1%) and from non-AD to AD in 1201 of 11,409 patients (10.5%). This higher change rate from AD to non-AD diagnosis compared with that from non-AD to AD agreed well with the results of the present study. The higher rate of change in pre-scan AD may be because AD can be ruled out when amyloid PET is negative. Furthermore, focusing on the details of the diagnosis, while pre-scan AD and MCI diagnoses decreased after the amyloid PET, post-scan diagnoses became further subdivided, increasing from 13 different diagnoses to 21 different diagnoses. This subdivision suggests the contribution of amyloid PET to more confirmatory diagnoses. In contrast, pre-scan non-AD diagnosis of 13 patients (6 MCI, 2 primary age-related tauopathy, 3 depression, 1 bvFTD, and 1 diabetic dementia) was changed to post-scan AD diagnosis. The considerable amyloid positivity with 59.3 ± 25.9 of CL scales led to the elevation of diagnostic confidence of AD.

These diagnostic changes led to changes in management in 42% of the patients, mainly in medication and care planning. These change rates were also comparable to those from previous investigations. The most common reported change in patient management due to amyloid PET results is a change in medication, ranging from 20 to 60% of cases [8, 11, 12, 14, 16]. In two previous studies, the care plan was changed in 10.9% [8] and 46.4% [9] of cases, respectively. The amyloid PET results also led to changes in prescription dosage and additional diagnostic tests in a small number of patients with a pre-scan AD diagnosis, as demonstrated in previous reports [7, 14, 16]. Thus, amyloid PET had a considerable impact on change in diagnosis and patient management by improving diagnostic confidence.

The inter-rater agreement of visual interpretation for amyloid positivity using ¹⁸F-florbetapir was similar to that in previous reports, where κ ranged from 0.69 to 0.71 [18, 27]. The high agreement of 95.9% of quantitative measures with the final decision regarding the visual interpretation may indicate the usefulness of CL scales as an adjunct to visual interpretation. The optimal CL cut-off values for amyloid positivity have been published by numerous investigations. A cut-off of 12.2 CL detected moderate-to-frequent CERAD neuritic plaques, while a cut-off of 24.4 CL identified intermediate-to-high AD neuropathological changes, according to a multicenter study using ¹¹C-PiB [28] that looked at the relationship between antemortem amyloid PET and standard postmortem measures of AD neuropathology. Similar research employing ¹¹C-PiB or ¹⁸F-florbetaben [29] found that the optimal threshold for finding moderate-to-frequent CERAD neuritic plaques was 20.1 CL, while the best cut-off for excluding neuritic plaques was a CL of 10. A favorable visual interpretation showed good agreement with results over 26 CL, according to that study's report. As for ¹⁸F-florbetapir, Clark et al. [30] demonstrated that an SUVR cut-off of 1.10 distinguished negative and positive ¹⁸F-florbetapir uptake relative to autopsy comparison of sparse/none versus moderate/frequent amyloid plaques. Royse et al. [31] also reported a similar SUVR cut-off of 1.11. In a comparative study of ¹¹C-PiB and ¹⁸F-florbetapir, Navitsky et al. [20] translated an SUVR threshold for amyloid positivity to 24.1 CL. This cut-off value is very close to the 25.9 obtained in the present study as a comparison with visual interpretation. It should be noted here, however, that the start time of 40 min post-injection and imaging time of 20 min for the PET images used in the present visual interpretation differed from those of the study by Navistky et al. [20], which started 50 min post-injection with a 10-min data acquisition. The present study demonstrated slight but significant CL changes depending on start time and imaging time. The CL_{50_60} value is approximately 8% higher than the CL_40–60 value, and the difference is more prominent for higher CL values. Thus, when a CL_50–60 value is applied to ROC analysis, the threshold for amyloid positivity slightly elevates to 27.7.

In the present study, quantitative values were also obtained from young healthy controls who were younger than 50 years of age, because the presence of amyloid β deposition in postmortem studies was found to be relatively rare before 50 years of age [32]. Although the SUVR and CL scales in this young control group were lower than those in visually amyloid-negative patients, the differences were not statistically significant. These amyloid-negative PET images from young healthy controls may be further applied to a software program for Z-score analysis [25] as a negative control database to localize the significant amyloid accumulation. In contrast with an AD-related increase in amyloid deposition in the posterior cingulate gyrus, precuneus, and frontal cortex, an age-related increase in amyloid deposits was specifically observed in the temporal neocortex [33].

This study has several limitations. First, the nonrandomized design and lack of a control group limit the direct attribution of changes in management to PET. Second, the observed changes in diagnosis and management represent the behavior of specialized physicians rather than an evidence-based standard of care. Third, because no postmortem data were available, the lack of a gold standard hampered our ability to relate the findings to the underlying neuropathology. Fourth, the sample size was not particularly large.

The present multicenter study suggested that ¹⁸F-florbetapir PET can exert a considerable clinical impact on AD and non-AD diagnosis and on patient management, particularly for medication and care planning, by improving the diagnostic confidence of AD. CL scale measures may help in the visual interpretation of amyloid positivity.