[Multilevel interpretable modeling and analysis of changes in dual pain dimensions in knee osteoarthritis].
文献摘要
Objective Based on real-world clinical data, aimed to systematically characterize the key factors, decision-making pathways, and structural relationships influencing pain, and to investigate the differences and associated structural effects of changes across two pain dimensions in patients with knee osteoarthritis(KOA). Methods Using the visual analog scale (VAS) and Traditional Chinese Medicine(TCM) pain symptom score as bimodal pain dimensions, XGBoost models were constructed under a unified feature system. Multi-layer interpretability analysis was conducted through SHAP, rule extraction, and Bayesian networks to examine the observational data from 187 cases of KOA. Results The model identified age, disease duration, body mass index (BMI), administration time and local joint symptoms as key factors influencing changes in the TCM pain symptom score, while baseline VAS and medication use were key factors for changes in the VAS. Decision path analysis revealed that the primary pathway for changes in TCM pain symptoms was "age -2 , and joint swelling score -2 , and joint swelling score Conclusion Under a unified model framework, an analytical approach for examining differential changes in related indicators was explored. Through multi-layered interpretability analysis encompassing "feature importance assessment-decision rule extraction-Bayesian network inference, " it was found that the dual pain dimensions share common influencing factors in reflecting pain changes in knee osteoarthritis, yet also exhibit structural association differences. These findings provide a reference basis for identifying pain change characteristics and selecting pain assessment tools in KOA patients, and offer a replicable methodological paradigm for similar analyses.