Biomarkers of iron metabolism facilitate clinical diagnosis in Mycobacterium tuberculosis infection. Issue 12 (14th October 2019)

• Record Type: Journal Article
• Authors: Dai, Youchao; Shan, Wanshui; Yang, Qianting; Guo, Jiubiao; Zhai, Rihong; Tang, Xiaoping; Tang, Lu; Tan, Yaoju; Cai, Yi; Chen, Xinchun
• Is Part Of: Thorax. Volume 74:Issue 12(2019)
• Page Start: 1161
• Page End: 1167
• Publication Date: 2019-10-14
• Subjects: iron homeostasis -- diagnostic biomarker -- neural network model -- tuberculosis -- receive operating characteristic (roc); Chest -- Diseases -- Periodicals; Thorax; Chest -- Diseases; Periodicals; Periodicals; 617.54
• DOI: 10.1136/thoraxjnl-2018-212557
• Languages: English
• ISSNs: 0040-6376

Abstract

Abstract : Background: Perturbed iron homeostasis is a risk factor for tuberculosis (TB) progression and an indicator of TB treatment failure and mortality. Few studies have evaluated iron homeostasis as a TB diagnostic biomarker. Methods: We recruited participants with TB, latent TB infection (LTBI), cured TB (RxTB), pneumonia (PN) and healthy controls (HCs). We measured serum levels of three iron biomarkers including serum iron, ferritin and transferrin, then established and validated our prediction model. Results: We observed and verified that the three iron biomarker levels correlated with patient status (TB, HC, LTBI, RxTB or PN) and with the degree of lung damage and bacillary load in patients with TB. We then built a TB prediction model, neural network (NNET), incorporating the data of the three iron biomarkers. The model showed good performance for diagnosis of TB, with 83% (95% CI 77 to 87) sensitivity and 86% (95% CI 83 to 89) specificity in the training data set (n=663) and 70% (95% CI 58 to 79) sensitivity and 92% (95% CI 86 to 96) specificity in the test data set (n=220). The area under the curves (AUCs) of the NNET model to discriminate TB from HC, LTBI, RxTB and PN were all >0.83. Independent validation of the NNET model in a separate cohort (n=967) produced an AUC of 0.88 (95% CI 0.85 to 0.91) with 74% (95% CI 71 to 77) sensitivity and 92% (95% CI 87 to 96) specificity. Conclusions: The established NNET TB prediction model discriminated TB from HC, LTBI, RxTB and PN in a large cohort of patients. This diagnostic assay may augment current TB diagnostics.