A Bayesian causal inference model to evaluate shoulder pathology effect on glenoid bone mineral density

Abstract

The effect of shoulder pathologies on glenoid bone mineral density (BMD) remains unclear and can be critical in surgical treatments. It is thus useful to predict this effect and understand how it is influenced by sex, age, and body mass index (BMI), in various glenoid locations. To address this question, we developed a causal model and used do-calculus to identify the minimal adjustment set of covariate variables and developed a varying-intercept varying-slope Bayesian model. We considered two common shoulder pathologies, primary osteoarthritis (OA) and cuff tear arthropathy (CTA), and compared them with normal shoulders (CTRL). Glenoid BMD was automatically measured on computed tomography scans of 93 OA, 53 CTA, and 133 CTRL subjects. OA and CTA subjects had higher BMD than CTRL in subchondral trabecular bone. This difference was affected by sex, increased with age, and was stable with BMI. BMD was higher in OA than CTA, especially on the posterior side. This causal model could be used for surgery planning, outcome prediction, and understanding of the associated pathophysiology.