Intelligent risk assessment of secondary spinal tuberculosis due to pulmonary tuberculosis based on routine test data

WANG Jing-lei; HE Yong-xiong; WANG Zhe; SONG Biao; LI Ming-dong

doi:10.3969/j.issn.2095-252X.2025.10.008

Abstract：Objective To explore the application value of establishing a differential analysis model for spinal tuberculosis under pulmonary tuberculosis using routine test data.Methods Routine test data were retrospectively collected for hospital visiting patients with tuberculosis,spinal TB,and extra-pulmonary TB from January 2011 to February 2024.A total of 346 cases of simple spinal tuberculosis,611 cases of extrapulmonary tuberculosis(with no spinal tuberculosis),233 cases of complicated spinal tuberculosis and 800 cases of tuberculosis were included.The training group and test group were set at the ratio of 8:2.Based on the model of deep learning neural network,two scenarios were modeled:identification between spinal tuberculosis and extrapulmonary TB,and differentiation between spinal tuberculosis and pulmonary TB.The feature contribution of the model was analyzed by the SHapley Additive exPlanations algorithm,and its performance was evaluated using the Receiver Operating Characteristic-Area Under the Curve(ROC-AUC)values,sensitivity,and specificity rate.Results The differentiation model of pure spinal tuberculosis and extrapulmonary tuberculosis showed excellent performance with ROC-AUC value of 0.861,accuracy rate of 81.18%,sensitivity of 72.12%and specificity of 86.34%,indicating that the model can effectively distinguish spinal tuberculosis from other external pulmonary tuberculosis;for the discrimination model of spinal tuberculosis and pure tuberculosis,the ROC-AUC value on test data set reached 0.858 with accuracy of 78.06%,sensitivity of 80.21%and specificity of 77.51%.This indicated that the model was not only more discriminative,but also can achieve accurate risk assessment in patients with different conditions.Conclusions The two machine learning models based on routine test data constructed in this study perform well in the identification of spinal tuberculosis and the risk assessment of secondary spinal tuberculosis from pulmonary tuberculosis.They can be used as important auxiliary tools for clinical decision-making,making up for the defects of traditional means of clinical identification of spinal tuberculosis.Big data technology is applied to the routine test data of pulmonary tuberculosis patients to achieve secondary risk assessment of spinal tuberculosis,forming a high-efficiency and low-cost auxiliary analysis tool.