Accuracy of biomarker testing for neuropathologically defined Alzheimer disease in older adults with dementia: A systematic review

Background: Biomarker accuracy for Alzheimer disease (AD) is uncertain. Purpose: To summarize evidence on biomarker accuracy for classifying AD in older adults with dementia. Data Sources: Electronic bibliographic databases (searched from January 2012 to November 2019 for brain imaging and cerebrospinal fluid [CSF] tests and from inception to November 2019 for blood tests), ClinicalTrials.gov (to November 2019), and systematic review bibliographies. Study Selection: English-language studies evaluating the accuracy of brain imaging, CSF testing, or blood tests for distinguishing neuropathologically defined AD from non-AD among older adults with dementia. Studies with low or medium risk of bias were analyzed. Data Extraction: Two reviewers rated risk of bias. One extracted data; the other verified accuracy. Data Synthesis: Fifteen brain imaging studies and 9 CSF studies met analysis criteria. Median sensitivity and specificity, respectively, were 0.91 and 0.92 for amyloid positron emission tomography (PET), 0.89 and 0.74 for ¹⁸F-labeled fluorodeoxyglucose (¹⁸F-FDG) PET, 0.64 and 0.83 for single-photon emission computed tomography, and 0.91 and 0.89 for medial temporal lobe atrophy on magnetic resonance imaging (MRI). Individual CSF biomarkers and ratios had moderate sensitivity (range, 0.62 to 0.83) and specificity (range, 0.53 to 0.69); in the few direct comparisons, β-amyloid 42 (Aβ42)/phosphorylated tau (p-tau) ratio, total tau (t-tau)/Aβ42 ratio, and p-tau appeared more accurate than Aβ42 and t-tau alone. Single studies suggested that amyloid PET, ¹⁸F-FDG PET, and CSF test combinations may add accuracy to clinical evaluation. Limitations: Studies were small, biomarker cut points and neuropathologic AD were inconsistently defined, and methods with uncertain applicability to typical clinical settings were used. Few studies directly compared biomarkers, assessed test combinations, evaluated whether biomarkers improved classification accuracy when added to clinical evaluation, or reported harms. Conclusion: In methodologically heterogeneous studies of uncertain applicability to typical clinical settings, amyloid PET, ¹⁸F-FDG PET, and MRI were highly sensitive for neuropathologic AD. Amyloid PET, ¹⁸F-FDG PET, and CSF test combinations may add accuracy to clinical evaluation.