A Substitutional Clinical Nomogram Model for Medical Thoracoscopy to Identify Tuberculous Pleural Effusion

Abstract Background: Tuberculous pleural effusion (TPE) is one of the most prevalent causes of exudative pleural effusion. As it is curable, early diagnosis is essential. Nevertheless, direct examination is not so sensitive while thoracoscopy is invasive and resource-demanding.Objective: To seek a less-invasive but also cost-effective substitute for medical thoracoscopy to identify TPE. Methods: We retrospectively collected 662 patients with pleural effusions who have undergone thoracoscopy as the training cohort, developed a predictive model and then validated it in another independent cohort. Independent predictor was screened by univariate analysis, feature selection and multivariate logistic regression. A predictive nomogram model was then established and evaluated by calibration, discrimination and clinical values. A standard model with a pleural adenosine deaminase (plADA) cut-off value at 40 IU/L was applied for comparison.Results: A lower plADA threshold at 24.5 IU/L was identified to discriminate TPE from other etiologies. Furthermore, serum tumor marker CA153 and Cyfra21.1, pleural lactic acid dehydrogenase and pleural interstitial cells were also identified as independent predictors (p<0.05) in multivariate logistic regression. The multivariate nomogram model presented good calibration, discrimination and clinical values, superior to the standard model. Excellent discrimination was demonstrated by the Harrell’s concordance index of 0.993 in the training cohort and 0.968 in the validation cohort. Furthermore, decision curve analysis demonstrated that the multivariate nomogram model added more benefit to patients with TPE.Conclusion: A multivariate nomogram model involved biochemical examinations and thoracentesis performed well in TPE clinical diagnosis among undetermined pleural effusions with approximate efficacy to medical thoracoscopy.