Abstract
BACKGROUND: Low back pain (LBP) is one of the most common symptoms of osteoporosis (OP), but LBP caused by osteoporosis can easily be masked by other causes, leading to misdiagnosis. However, there are currently no convenient tools available to identify patients with low back pain caused by osteoporosis. METHODS: We consecutively enrolled 769 patients diagnosed with low back pain in our hospital from January 2019 to March 2024. A total of 355 cases were excluded due to relevant missing data, leaving a final analysis cohort of 414 cases. The dataset was randomly divided into a training group and a validation group at a ratio of 7:3 for further analysis. in this preliminary analysis were selected for subsequent multivariate analysis. Least absolute shrinkage and selection operator(LASSO) was employed to identify the associated risk factors for osteoporosis. Independent variables with P<0.05 in univariate analysis were included in the multivariate analysis to construct the prediction model. Once the regression equation was established, a nomogram was utilized to visualize the prediction model, while receiver operating characteristic (ROC) curve was plotted to evaluate its performance, specifically by calculating the area under the curve (AUC) which represents discrimination ability of the model. To assess goodness-of-fit, calibration curve was generated for evaluating calibration accuracy. Furthermore, decision curve analysis (DCA) served to determine clinical application value of this predictive model. Statistical significance level was set at P < 0.05. RESULTS: Building upon the LASSO and multivariate Cox regression, eleven variables were significantly associated with OP (i.e., gender, age, history of fracture, history of alcohol consumption, history of rheumatoid arthritis, hematocrit, red blood cell volume distribution width, lymphocyte percentage, triglyceride, potassium ion, and alanine aminotransferase). In training and validation sets, AUCs and C-indexes of the OP prediction models were all greater than 0.8(AUC: 0.914 for training; 0.833 for validation), which indicated excellent predictability of models. On the whole, the calibration curves coincided with the diagonal in two models. DCA indicated that the models had higher clinical benefit than other risk factors. While confirmed the clinical utility of the model, as it outperformed both the 'treat-all' and 'treat-none' strategies. CONCLUSION: After verification, our prediction models of OP are reliable and can predict the incidence of osteoporosis, providing valuable guidance for clinical prognosis estimation and individualized administration of patients with LBP(a new way for early identification and intervention of patients with osteoporosis).