3
Comparative Analysis of Two Brain Age Algorithms Across Clinical Pain and Psychological Measures in Individuals with and Without Knee Pain
Kimberly Sibille, University of Florida, Gainesville, United States
Brittany Addison, University of Florida, Gainesville, United States
Kenia Rangel, University of Florida, Gainesville, United States
Angela Mickle, University of Florida, Gainesville, United States
Cynthia Garvan, University of Florida, Gainesville, United States
Song Lai, University of Florida, Gainesville, United States
Roland Staud, University of Florida, Gainesville, United States
Burel Goodin, Washington University, St. Louis, United States
Roger Fillingim, University of Florida, Gainesville, United States
Catherine Price, University of Florida, Gainesville, United States
Tanner Jared, University of Florida, Gainesville, United States
Several algorithms have been developed that utilize machine learning techniques to estimate an individual's neurological age (i.e., “brain age”) based on MRI data, serving as a surrogate metric for neurological health. The concept of "Brain Age Gap" (BAG), calculated as the difference between estimated and chronological age, is used as an index for accelerated or decelerated aging. The study examined the validity and clinical relevance of two brain age estimation algorithms: Kaufmann et al.'s (2019) method and DeepBrainNet. Specifically, the study assessed how these algorithms correlated with clinical measures that assess pain intensity, physical functioning, depressive symptoms, and cognitive performance in individuals with and without osteoarthritis-associated knee pain.
Cross-sectional data were analyzed from 203 adults between 45 80 years of age, consisting of 49% non-Hispanic Blacks and 51% non-Hispanic Whites adults, with or without chronic knee pain. Participants completed a Health Assessment and brain MRI.
Kaufmann's brain age significantly positively correlated with chronological age (r=0.62, 95%CI=[0.55, 0.68], p<1.52e-21) and Kaufmann’s BAG showed a relationship with chronological age (r=-0.45, 95%CI=[-0.56, -0.34], p = 1.71e-10). DeepBrainNet’s brain age exhibited a stronger correlation with chronological age (r=0.73, 95%CI=[0.67, 0.78], p<2.58e-33; z-diff=-3.29) and no significant relationship between the derived DeepBrainNet BAG and age.
In correlation analyses with clinical measures, Kaufmann's BAG correlated with GCPS CPI (r=0.19, p=0.048) but not GCPS disability. DeepBrainNet's BAG is associated with both GCPS CPI and disability (r=0.24 and r=0.25, p=0.003). Both Kaufmann's and DeepBrainNet's BAGs correlated with PROMIS Depression T-scores (r=0.26 and r=0.25, p=0.0029 and p=0.0032). Neither showed a significant correlation with MoCA after multiple comparison corrections.
Both brain age estimation methods demonstrated relationships with clinical variables in our sample. However, given the observed future research should focus on algorithm refinement and cross-validation using different clinical populations. This would allow for a more robust assessment of each algorithm's reliability and validity.
Grant Funding: NIA: R01AG054370 (KTS); R01AG054370-05S1 (JJT, KTS), R37AG033906 (RBF); RCCN NIA Pilot Funding 2021-2022
Keyword 1: aging (normal)
Keyword 2: neuroimaging: structural
Keyword 3: chronic pain