Predicting Brain Age for Veterans with Traumatic Brain Injuries and Healthy Controls: An Exploratory Analysis

John Philip Coetzee^1,2*Xiaojian Kang²victoria Liou-Johnson^1,2Innes Luttenbacher³Srija Seenivasan⁴Elika Eshghi⁵Daya Grewal⁶Siddji Shah²Frank Hillary⁷Emily L Dennis⁸Maheen Mausoof Adamson^1,2

• ¹Stanford University, Stanford, United States
• ²VA Palo Alto Health Care System, Veterans Health Administration, United States Department of Veterans Affairs, Palo Alto, California, United States
• ³University of Amsterdam, Amsterdam, Netherlands
• ⁴Uniformed Services University of the Health Sciences, Bethesda, Maryland, United States
• ⁵Icahn School of Medicine at Mount Sinai, New York, New York, United States
• ⁶Palo Alto University, Palo Alto, California, United States
• ⁷The Pennsylvania State University (PSU), University Park, Pennsylvania, United States
• ⁸School of Medicine, The University of Utah, Salt Lake City, Utah, United States

Traumatic brain injury (TBI) is associated with increased dementia risk. This may be driven by underlying biological changes resulting from the injury. Machine learning algorithms can use structural MRIs to give a predicted brain age (pBA). When the estimated age is greater than the chronological age (CA), this is called the brain age gap (BAg). We analyzed this outcome in men and women with and without TBI.Objective: To determine whether factors that contribute to BAg, as estimated using the brainageR algorithm, differ between men and women who are US military Veterans with and without TBI. Methods: In an exploratory, hypothesis-generating analysis, we analyzed data from 85 TBI patients and 22 healthy controls (HCs). High-resolution T1W images were processed using FreeSurfer 7.0. pBAs were calculated from T1s. Differences between the two groups were tested using the Mann-Whitney U. Associations between the BAg and other factors were tested using partial Pearson’s r within groups, controlling for CA, followed by construction of regression models. Results: After correcting for multiple comparisons, TBI patients and HCs differed on PCL score (higher for TBI patients) and cortical thickness (CT) in both hemispheres (higher for HCs). Among women TBI patients, BAg was correlated with pBA and hippocampal volume (HV), and among men TBI patients, BAg was correlated with pBA and CT. Among both men and women HCs, BAg was correlated only with CA. Four hierarchical regression models were constructed to predict BAg in each group, which controlled for CA and excluded pBA for multicollinearity. These models showed that HV predicted BAg among women with TBI, while CT predicted BAg among men with TBI, while only CA predicted BAg among HCs.Interpretation: These results offer tentative support to the view the factors associated with BAg among individuals with TBI differ from factors associated with BAg among HCs, and between men and women. Specifically, BAg among individuals with TBI is predicted by neuroanatomical factors, while among HCs it is predicted only by CA. This may reflect features of the algorithm, an underlying biological process, or both.