Interpreting Biomarker Test Results for Alzheimer Disease, Parkinson Disease and Other Neurodegenerative Diseases Without the Autopsy Gold Standard

There is an extensive literature on the difficulty in assessing new diagnostic tests when an accurate gold standard does not exist, is imperfect, or is not available. Relatively few reports, however, address this problem when it confronts researchers reporting tests of potential new biomarkers for Alzheimer disease (AD), Parkinson disease (PD) and other neurodegenerative diseases. This is despite the reality that the vast majority of published studies employ the neurologists clinical diagnoses as the gold standard, despite their well-known inherent inaccuracies relative to the true, gold standard, autopsy neuropathology diagnosis. More recently, biomarkers that have, appropriately, been evaluated against the autopsy gold standard, have then themselves been used as a surrogate gold standard to evaluate other biomarkers, despite their less-than-perfect accuracy against autopsy. The shortcomings of these approaches to neurodegenerative disease biomarker validation are rarely discussed. It is clear from the prior literature that testing the accuracy of a new AD or PD diagnostic test against the clinical diagnosis can in fact lead to both underestimates and overestimates of its true (against autopsy) accuracy. Despite these problems, the clinical diagnoses of AD and PD are still routinely used to assess the accuracy of new biomarkers. A related issue is that when a new biomarker is evaluated against a test type previously validated against autopsy (e.g. amyloid PET tau PET), in effect serving as a surrogate gold standard, it is again clear that this new biomarker might itself be more accurate, as accurate or less accurate than the surrogate gold standard. We here present a method that makes it possible, if there are published data on the accuracy of the clinical diagnosis, or of a surrogate gold standard, relative to autopsy, to at least estimate a range of possible accuracies of a new biomarker using those imperfect gold standards. Our procedure was developed using basic theoretical modeling of conditional probabilities.