LAUSR

Background: The Boston criteria are used widely for the noninvasive diagnosis of sporadic cerebral amyloid angiopathy (CAA) and hence clinical decision-making, as well as research in the field. Yet, their exact diagnostic accuracy and validity remain (paradoxically) poorly studied. We performed a meta-analysis to synthesize evidence on the value and accuracy of the Boston criteria in diagnosing probable CAA patients. Methods: In a systematic literature search, we identified studies with extractable data relevant for sensitivity and specificity of probable CAA diagnosis per the magnetic resonance imaging Boston criteria and neuropathological CAA verification. We included studies that have classified patients according to any version of the Boston criteria, based on available brain magnetic resonance imaging blood-sensitive sequences (index test) and had neuropathologic evaluation for CAA presence from brain tissue samples (diagnostic reference standard). Using a hierarchical (multilevel) logistic regression model, we calculated pooled diagnostic test accuracy for probable CAA diagnosis. Results: Seven studies, including 193 patients, 121 with neuropathologically verified CAA versus 72 non-CAA based on neuropathology definition, were included in the meta-analysis. The studies were of low-to-moderate quality and varied in several methodological aspects. The overall pooled sensitivity for probable CAA diagnosis was 66.7% (95% CI, 45.9%–82.6%) and specificity was 88.2% (95% CI, 68.5%–96.3%). A predefined subgroup analysis of 4 studies on Boston criteria v.1.0 (n=151) demonstrated a pooled sensitivity and specificity of 60% (95% CI, 45.1%–72.9%) and 93.1% (95% CI, 81.8%–97.6%), respectively. Five studies had data on Boston criteria v.1.5 (n=123): the pooled sensitivity and specificity for probable CAA diagnosis was 73.1% (95% CI, 45%–90.1%) and 86% (95% CI, 41.4%–98.1%), respectively. Conclusions: The Boston criteria v.1.0 and v.1.5 appear to have moderate-to-good diagnostic accuracy for probable CAA in symptomatic patients, with high specificity but low-to-moderate sensitivity. Data are based on limited retrospective studies of overall low quality and at high risk of bias.