[A clinical auxiliary differential diagnostic model for thoracolumbar spinal tuberculosis based on routine laboratory...

Spinal tuberculosis commonly involves the thoracic and lumbar vertebrae, but its clinical differential diagnosis remains challenging. This study included patients with imaging findings suggestive of thoracic or lumbar spinal infection and aims to establish a diagnostic model based on routine laboratory indicators to assist in differentiating thoracolumbar spinal tuberculosis from non-tuberculous spinal infections.

Clinical data were consecutively collected from patients whose imaging findings suggested thoracic or lumbar spinal infection. Patients were divided into a thoracolumbar spinal tuberculosis group and a non-tuberculous spinal infection group according to the final diagnosis. Candidate variables included 40 routine clinical laboratory indicators. First, the Boruta algorithm was used to screen all candidate variables. All cases were then randomly divided into a training set and a validation set at a ratio of 7꞉3, with 70% of cases used for model construction and 30% for model validation. In the training set, the variables selected by the Boruta method were further analyzed using logistic regression to identify independent diagnostic variables with statistical significance and clinical interpretability. An auxiliary differential diagnostic model for thoracolumbar spinal tuberculosis was then constructed based on the final variables included in the model. A nomogram was further developed to visually present the contribution of each predictor to the diagnostic outcome in the form of scores. Receiver operating characteristic curves and the area under the curve were used to evaluate the discriminative ability of the model in the training and validation sets, and the diagnostic performance of the model was internally validated using the validation set. Calibration analysis and decision curve analysis were also performed to evaluate the consistency between predicted probabilities and actual diagnostic results, as well as the potential clinical net benefit of the model.

A total of 275 patients with thoracolumbar spinal infection were included, comprising 104 patients in the tuberculosis group and 171 patients in the non-tuberculous group. The Boruta method selected 11 variables from the 40 candidate variables: Interferon-γ release assay, Mycobacterium tuberculosis antibody, red blood cell count, hemoglobin, lymphocyte count, monocyte percentage, high-density lipoprotein, D-dimer, C-reactive protein, erythrocyte sedimentation rate, and activated partial thromboplastin time. A model was then established using Logistic regression. Five variables with significant differences, namely high-density lipoprotein, tuberculosis antibody, monocyte percentage, lymphocyte count, and interferon-γ release assay, were used to construct the nomogram. Each variable can be converted into a corresponding score according to its value, and the total score can be used to estimate the predicted probability of thoracolumbar spinal tuberculosis. The model showed good discriminative ability in the training set, with an area under the receiver operating characteristic curve of 0.877. It maintained stable diagnostic performance in the validation set, with an area under the curve of 0.875, suggesting good internal validation performance and certain generalizability. Calibration analysis showed good agreement between the predicted probabilities and actual diagnostic results. Decision curve analysis demonstrated that, within a certain range of threshold probabilities, the model provided higher clinical net benefit than the simple assumption that all patients had thoracolumbar spinal tuberculosis or that all patients had non-tuberculous spinal infection, indicating its incremental value for clinical differential diagnosis.

Five routine laboratory indicators, namely high-density lipoprotein, tuberculosis antibody, monocyte percentage, lymphocyte count, and interferon-γ release assay, can be used to construct an auxiliary differential diagnostic model for thoracolumbar spinal tuberculosis. This model has significant net benefit and good ability to distinguish thoracolumbar spinal tuberculosis from non-tuberculous spinal infection. It has potential application value as an auxiliary differential diagnostic model before etiological results become available.