Prevalence, associated factors, and machine learning classification of low back pain among postmenopausal women attending rehabilitation and physiotherapy facilities in Bangladesh

Background Low back pain (LBP) is a leading cause of disability worldwide and disproportionately affects women, particularly after menopause. However, evidence on the prevalence and associated factors of LBP among postmenopausal women in low- and middle-income countries remains limited. This study aimed to estimate the past-year prevalence of LBP among postmenopausal women attending selected rehabilitation and physiotherapy facilities in Bangladesh, examine factors associated with LBP, and evaluate the internal classification performance of machine learning models. Methods A clinic-based cross-sectional study was conducted among 566 postmenopausal women attending two specialized rehabilitation centers in Bangladesh. Sociodemographic, clinical, lifestyle, and socioeconomic data were collected using structured questionnaires. Multivariable logistic regression was used to estimate adjusted associations with LBP. Machine learning models (Logistic Regression, Support Vector Machine with radial basis function (SVM-RBF), Random Forest, Bernoulli Naïve Bayes, and Extreme Gradient Boosting) were developed to classify LBP. Model development used nested cross-validation, and final discriminative performance was evaluated using a held-out test set. Model interpretability was assessed using SHapley Additive exPlanations. Results The prevalence of past-year LBP among the postmenopausal women was 46.6%. Multivariable analysis showed lower odds of LBP among women aged 51–57 years than those aged 43–50 years (AOR = 0.50, 95% CI: 0.31–0.81), and higher odds among housewives (AOR = 2.20, 95% CI: 1.33–3.64), women with hypertension (AOR = 1.60, 95% CI: 1.09–2.35), respiratory disease (AOR = 2.46, 95% CI: 1.28–4.71), and selected socioeconomic groups, while heart disease and BMI category were not clearly associated after adjustment. Among the evaluated machine learning models, the SVM-RBF classifier achieved the highest accuracy during nested cross-validation (0.772), while random forest achieved comparable accuracy (0.768) and the highest F1-score (0.744). On the held-out test set, SVM-RBF achieved the highest ROC-AUC (0.890), followed by random forest (0.885). SHAP analysis identified sunlight exposure, hypertension, and socioeconomic status as the most influential contributors to model classification. Conclusions LBP is commonly reported among postmenopausal women seeking rehabilitation and physiotherapy care in Bangladesh and is associated with multiple social, occupational, and health-related factors. Machine learning models demonstrated relatively strong internal classification performance; however, external validation is required before broader clinical use can be considered. Longitudinal community- and clinic-based studies are needed to clarify temporal relationships and assess the generalizability of these findings.