Abstract:

Objective: To establish and verify the early diagnosis model of knee joint tuberculosis based on LASSO regression. Methods: One hundred and thirty-six patients with knee joint tuberculosis admitted to Public Health Clinical Center of Chengdu from January 2019 to January 2022 were selected as the case group; 136 patients with non-tuberculous knee disease in the same period were selected as the control group for modeling. In addition, 72 patients with suspected knee joint tuberculosis from February to October 2022 were selected as the validation group, 13 of whom were pathologically confirmed as having it. The general information, laboratory examination and MRI examination results of those patients were collected. The indicators of patients in the two groups were compared, and LASSO regression was used to screen factors that might be associated with knee joint tuberculosis and multivariable logistic regression was conducted to establish a nomogram model which then was verified internally. Results: LASSO regression model screened out 11 potential diagnostic factors (gender, age, IFN-γ release, LAM antibody, GeneXpert MTB/RIF result, bone marrow edema, meniscus injury, cartilage injury, swelling of surrounding tissue, bone destruction and periarticular abscess formation). The results of multivariable analysis showed that age (OR=0.977, 95%CI: 0.955-0.999), IFN-γ release level (OR=1.005, 95%CI: 1.001-1.009), LAM antibody positive (OR=15.348, 95%CI: 6.344-37.130), GeneXpert MTB/RIF (OR=21.073, 95%CI: 8.281-53.628), bone marrow edema (OR=2.996, 95%CI: 1.165-7.702), meniscus injury (OR=5.007, 95%CI: 1.868-13.425), cartilage injury (OR=4.117, 95%CI: 1.649-10.274), surrounding tissue swelling (OR=5.389, 95%CI: 2.059-14.102) and periarticular abscess formation (OR=7.570, 95%CI: 1.876-30.546) were independent influencing factors for knee joint tuberculosis. A nomogram model was established according to the results of multivariable analysis and a ROC curve was drawn according to the data of the validation group. Results showed that the AUC of the nomogram model for predicting the risk of knee joint tuberculosis was 0.927 (95%CI (0.898-0.957)); calibration curve analysis showed that the predicted risk probability of knee joint tuberculosis by the nomogram model was basically consistent with the actual probability. Decision curve analysis showed that when the probability threshold of nomograph model to predict the risk of knee joint tuberculosis was 0.15-0.90, the net profit rate of patients was greater than 0. Conclusion: With the increase of age, the risk of tuberculosis of knee joint decreases, IFN-γ release increasing, LAM antibody positive, GeneXpert MTB/RIF positive, as well as bone marrow edema, meniscus injury, cartilage injury, swelling of surrounding tissue and formation of periarticular abscess, increase the risk of knee tuberculosis. The nomograph prediction model established based on the above factors could be used for early diagnosis of knee joint tuberculosis.

Key words: Tuberculosis, Knee joint, Early diagnosis, Models, statistical