Construction and validation of a LASSO regression-based nomogram prediction model for active tuberculosis

Abstract

Objective To develop and validate a nomogram prediction model for the early diagnosis of active pulmonary tuberculosis (ATB) based on LASSO regression. Methods A total of 403 patients with ATB admitted to the Third People's Hospital of Kunming between March 2021 and March 2023 were used as the experimental group, and 175 cases of healthy physical examiners during the same period were selected as the control group. General information and laboratory test results of patients were collected, variables were screened using LASSO regression based on R software, and multivariable logistic regression was performed. A nomogram model of ATB was established based on the results of multivariable analysis, and internal validation was performed using receiver operating curve (ROC) and calibration curves, while clinical utility analysis using clinical decision curves. Results A total of 19 potential risk factors were screened by LASSO regression, namely C-reactive protein (CRP), haptoglobin (HAP), immunoglobulin G (IgG), absolute number of CD4+ lymphocytes (CD4+), ratio of CD4+ lymphocytes to CD8+ lymphocytes (CD4+/CD8+), interleukin (IL)-1β, IL-6, IL-8, IL-10, IL-17, lymphocyte percentage (LYM%), monocyte percentage (MON%), eosinophil percentage (EOS%), mean corpuscular hemoglobin concentration (MCHC), mean red blood cell volume (MCV), platelet count (PLT), red blood cell distribution width-standard deviation (RDW-SD), neutrophil to lymphocyte ratio (NLR) and platelet large cell ratio (PLCR). Multirariable analysis showed that CRP (OR=1.352, 95%CI: 1.134-1.612), IL-6 (OR=1.165, 95%CI: 1.032-1.315),IL-8 (OR=1.105, 95%CI: 1.019-1.198), IL-10 (OR=1.544, 95%CI: 1.066-2.235), EOS% (OR=1.386, 95%CI: 1.105-1.737), MCV (OR=1.154, 95%CI: 1.051-1.737), PLT (OR=1.025, 95%CI: 1.013-1.037), and MCHC (OR=0.899, 95%CI: 0.854-0.946) were the independent risk factors for ATB. According to the nomogram, the ROC was plotted and showed that AUC of ATB risk predicted by model was 0.982 (95%CI: 0.973-0.991), the calibration curve results showed that the predicting probability of ATB risk by the nomogram model was basically consistent with the actual probability, and the results of decision curve analysis showed that when the probability threshold of the nomogram model predicting the risk of ATB was more than 0.05, the net benefit value of the patients was greater than 0. Conclusion The risk of ATB increases with increasing levels of CRP, IL-6, IL-8, IL-10, EOS%, MCV, PLT and decreasing levels of MCHC. The nomogram prediction model based on the above factors can be used for the early diagnosis of ATB.

Keywords

• active tuberculosis
• early diagnosis
• nomogram
• statistical models
• mean red blood cell volume
• platelet count
• interleukin
• c-reactive protein
• mean corpuscular hemoglobin concentration