## a randomized, placebo-controlled, dose-response exploratory study

Trang Lê

Biomedical Researcher
Institute for Biomedical Informatics
University of Pennsylvania

Laureate Institute for Brain Research, Tulsa, OK

extract image-related features

train a machine learning model on a large normative dataset
to predict age

obtain predicted ages on a separate dataset of interest with the trained model

compute each subject’s Brain Age Gap Estimate (BrainAGE)

make inferences about accelerated or delayed aging

T1-weighted structural

diffusion tensor

functional MRI

\textrm{BrainAGE}_i = \widehat\textrm{age}_{\textrm{brain, i}} - \widehat\textrm{age}_{\textrm{chrono, i}}

SVR

RVR

GPR

\widehat\textrm{age}_{\textrm{brain, i}}

medial prefrontal cortex: strong age predictor

### Gray matter density 3334 course voxels

r = 0.76, MAE = 5.34 years

## SVR 10-fold CV:   , C

\epsilon

20 healthy participants (10 males)

ICC =                  = 0.96

(95% CI=(0.92, 0.98)

\frac{\hat\sigma^2_B}{\hat\sigma^2_B + \hat\sigma^2_W}

F=5.83, p=0.006

BrainAGE was reduced by

• 200mg: 1.15 year (t=-2.95, p=0.005)
• 600mg: 1.18 year (t=-2.97, p=0.005)

Cohen's D (95% CI):

• D200mg-placebo: 0.57 (0.27, 0.95)
• D600mg-placebo: 0.69 (0.39, 1.15)
\textrm{BrainAGE} = (\beta_0 + u_i) + \beta_1(\textrm{ibu 200mg}) + \beta_2(\textrm{ibu 600mg}) + \epsilon
u_i \sim N(0, \sigma^2_B)
\epsilon \sim N(0, \sigma^2_W)

random intercept

within-subject variation

@trang1618

Thank you!

Funding: The William K. Warren Foundation

LIBR collaborators:

• Rayus Kuplicki
• Henry Yeh
• Martin Paulus

By Trang Le

• 537