Clinically constrained Bayesian network analysis of immune-metabolic factors associated with bone mineral density in a real-world orthopedic cohort.

OBJECTIVES

Osteoporosis substantially influences surgical decision-making, implant fixation, fracture risk, and postoperative outcomes in orthopaedic practice. However, the systemic biological interactions underlying bone loss remain difficult to characterize using conventional statistical models. We aimed to integrate clinically constrained Bayesian network learning with routine biochemical data to identify immune-metabolic factors associated with bone mineral density (BMD) and to provide a clinically interpretable framework for bone health assessment.

METHODS

In a retrospective real-world orthopedic cohort of 2,880 individuals with concurrent DXA and biochemical data, we constructed a 20-node network comprising BMD, bone metabolism biomarkers, and immune-inflammatory, metabolic, renal, and endocrine indicators. Gaussian graphical models and prior-constrained Bayesian networks were applied, blacklisting edges that pointed to age or menopause status to enforce physiological plausibility. Bridge mediation analysis systematically scanned source-bone biochemistry-BMD pathways, using bootstrap confidence intervals. Analyses were repeated in the postmenopausal subgroup (n = 1,546).

RESULTS

Network hubs included parathyroid hormone (PTH), phosphorus, neutrophil-to-lymphocyte ratio (NLR), high-density lipoprotein (HDL), and triglycerides (TG). Prior constraints eliminated all prespecified prohibited edges. NLR-ALP-BMD emerged as the most robust negative bridging pattern (indirect effect for lumbar spine: -0.045; mediation proportion: 70.5% in the full cohort), with similar findings in postmenopausal women. NLR-PTH-BMD and NLR-phosphorus-BMD were also significant. The NLR-ALP-BMD indirect associations remained significant after hepatic-marker adjustment and in participants without elevated ALP or ALT, whereas the ALB-ALP-BMD signal was attenuated and inconsistent across subgroups. Positive indirect effects involving glucose, TG, and uric acid may reflect unmeasured adiposity and DXA measurement artifacts.

CONCLUSION

Clinically constrained network analysis identified immune-metabolic factors associated with bone mineral density and provided a biologically plausible framework for interpreting bone-immune-metabolic interrelationships in orthopedic patients.