Concordance Between Amyloid-PET Quantification and Real-World Visual Reads

Importance With increased payer coverage and the advent of antiamyloid therapies, clinical use of amyloid positron emission tomography (PET) is likely to increase to help guide the diagnosis and treatment of patients with cognitive impairment. However, unlike most previous research studies, in clinical practice, scan acquisition is less standardized, interpretation typically relies purely on visual reads rather than scan quantification, and patients have more frequent comorbidities, all of which might compromise test accuracy. Objective To compare visual interpretation of amyloid-PET in real-world clinical settings to scan interpretation based on central quantification, in order to assess the accuracy of clinical reads. Design, Setting, and Participants This cross-sectional quality improvement study used data from the Imaging Dementia—Evidence for Amyloid Scanning study, collected between February 2016 and January 2018 and analyzed between December 2021 and April 2023. The setting included 294 imaging facilities in the US. Medicare beneficiaries 65 years or older with cognitive decline for whom Alzheimer disease was a diagnostic consideration were recruited by dementia specialists from their clinical practices. Exposures Amyloid-PET with [18F]florbetapir, [18F]florbetaben, or [18F]flutemetamol. Main Outcomes and Measures PET scans were visually interpreted as positive or negative by local radiologists or nuclear medicine physicians following approved guidelines. Independently, scans were centrally processed and quantified using the standardized Centiloid (CL) scale. We applied an a priori autopsy-based threshold of 24.4 CL to quantitatively define scan positivity. Results Of 18 293 participants included in the parent study, scan images were available for 10 774 (59%), of which Centiloids were successfully calculated for 10 361 (96%). Median (IQR) patient age was 75 (71-80) years; 5245 patients (51%) were female, 6500 (63%) had mild cognitive impairment, and 3861 (37%) had dementia). Participants self-reported the following races and ethnicities: 1 Alaska Native (0%), 23 American Indian (0.2%), 188 Asian (1.8%), 316 Black (3.1%), 449 Hispanic or Latino (4.3%), 8 Native Hawaiian or Other Pacific Islander (0.1%), and 9125 White (88.2%). A total of 6332 scans (61%) were visually read as positive, and 6121 (59%) were quantitatively positive. Agreement between visual reads and quantitative classification was 86.3% (95% CI, 85.7%-87.0%; Cohen κ = 0.72; 95% CI, 0.70-0.73). A total of 5519 (53%) scans were positive visually and quantitatively (V+/Q+), 3416 (33%) were negative by both (V−/Q−), 813 (8%) were V+/Q−, and 602 (6%) were V−/Q+. Female sex (female: 4581/5241 [87.4%]; male: 4354/5109 [85.2%]; P =.001), White race (White race: 7900/9125 [86.6%]; non-White race: 1035/1225 [84.5%]; P =.046), and use of [18F]flutemetamol and [18F]florbetaben ([18F]flutemetamol: 559/628 [89.0%]; [18F]florbetaben: 2664/3032 [87.9%]; [18F]florbetapir: 5712/6690 [85.4%]; P <.001), compared with [18F]florbetapir, were associated with higher visual-quantitative concordance. Scans within a 10- to 40-CL borderline positivity zone were more likely to be discordant. Conclusions and Relevance This cross-sectional study found high concordance between local visual reads and central quantification of clinical amyloid-PET scans, supporting the validity of amyloid-PET visual reads in real-world clinical practice.