DOI: 10.1002/alz.081653 ISSN: 1552-5260

Biological and methodological factors underlying a continuous amyloid CSF/PET imbalance model and its association with longitudinal cognition

Sophie E Mastenbroek, Arianna Sala, Juan Domingo Gispert, David Vállez García, Mark E Schmidt, Andrew W. Stephens, Rossella Gismondi, Gill Farrar, Giovanni B Frisoni, Mercè Boada, Craig W Ritchie, Pieter Jelle Visser, Marco Bucci, Oskar Hansson, Frederik Barkhof, Rik Ossenkoppele, Agneta K Nordberg, Elena Rodriguez‐Vieitez, Lyduine E. Collij,
  • Psychiatry and Mental health
  • Cellular and Molecular Neuroscience
  • Geriatrics and Gerontology
  • Neurology (clinical)
  • Developmental Neuroscience
  • Health Policy
  • Epidemiology



Discordance between cerebrospinal fluid (CSF) and positron emission tomography (PET) biomarkers of amyloid‐ß (Aß) occurs in up to 20% of individuals. Generally, Aß discordance is defined using binary cut‐offs. Given the methodological limitations of this approach, we propose a continuous Aß CSF/PET imbalance model. We investigated biological and methodological factors underlying discrepancy in these amyloid biomarkers and associations with longitudinal cognition.


We included 383 non‐demented subjects (Nunimpaired = 326, Nmildly‐impaired = 57) from the AMYPAD‐PNHS cohort. Z‐scored CSF Aß42 and amyloid‐PET (Centiloids [CL]) units were fitted to a hyperbolic regression model by minimizing the Euclidian distance of the observed datapoints to the fitted line. Subject‐specific standardized residuals were derived as a measure of continuous imbalance (Zimbalance), with Zimbalance<0 reflecting greater Aß abnormality in CSF relative to PET and Zimbalance>0 greater Aß abnormality in PET relative to CSF (Figure‐1). Linear models adjusted for relevant covariates (Figure‐2) were run between Zimbalance and (1) methodological factors (CSF‐PET interval, CSF kit, PET tracer, and PET injected dose); (2) demographics (age, sex, education, APOE genotype) and CSF p‐tau biomarker; (3) vascular burden (white matter hyperintensities measured as Fazekas scores). Linear mixed models were used to investigate the predictive value of Zimbalance on longitudinal cognition in several domains.


Sample characteristics are shown in Table‐1. Methodological factors were not significantly associated with Zimbalance. Increased CSF p‐tau levels were significantly associated with a higher Zimbalance (i.e. greater Aß abnormality in PET relative to CSF) (ß = 0.06±0.03, p = .033) (Figure‐2). In addition, a lower Zimbalance (i.e. greater Aß abnormality in CSF relative to PET) was associated with a higher Fazekas score (n = 189, ß = ‐0.55±0.26, p = .038) and predictive of faster cognitive decline on a language categorical fluency test (ß = ‐0.03±0.02, p = .038) (Figure‐2).


We found that methodological factors did not contribute to a continuous model of Aß CSF/PET imbalance, suggesting that this imbalance reflects distinct biological profiles. Indeed, continuous CSF/PET imbalance showed differential associations, with greater pathological levels of aggregated Aß relative to soluble Aß being related to increased CSF p‐tau levels and excess pathological levels of soluble Aß relative to aggregated Aß to increased white matter hyperintensities and steeper cognitive decline.

More from our Archive