Application of a machine learning model based on routine clinical parameters for the diagnosis of rheumatoid arthritis with concomitant osteoporosis: a retrospective study

Rheumatoid arthritis (RA) is commonly complicated by secondary osteoporosis (OP), affecting up to 80% of patients. Although dual-energy X-ray absorptiometry (DEXA) is the diagnostic gold standard, its limited accessibility highlights the need for alternative tools. In this retrospective cohort study of 396 hospitalized RA patients, we developed machine learning models using demographic and routine laboratory data to identify concomitant OP. Five classifiers were evaluated and combined <i>via</i> a soft-voting ensemble. The support vector machine achieved the highest area under the curve (AUC) (86.5%), while the random forest showed the highest accuracy (81.5%). The ensemble model demonstrated balanced performance (AUC 83.2%; accuracy 81.1%). SHapley Additive exPlanations (SHAP) analysis indicated age, sex, and body mass index (BMI) as major contributors, whereas albumin and inflammatory markers-including platelet-to-lymphocyte ratio (PLR), neutrophil percentage/albumin ratio (NPAR), white blood cell count (WBC), and neutrophil-to-lymphocyte ratio (NLR)-showed modest but heterogeneous influences on model predictions. These findings suggest that machine learning models incorporating routinely collected clinical data offer a practical and interpretable approach for preliminary OP risk assessment in RA. However, given the single-center design and limited sample size, the results should be considered exploratory, and larger external validation studies are warranted.