Construction of a comprehensive prediction and visualization system for drug resistance in pulmonary tuberculosis patients based on an improved machine learning model
10.12092/j.issn.1009-2501.2025.05.011
- VernacularTitle:基于改良机器学习模型的肺结核患者耐药综合预测及可视化系统搭建
- Author:
Feng WANG
1
;
Luhua LIANG
;
Fei ZHAI
;
Xiaoling LUO
;
Rongwu XIANG
Author Information
1. 沈阳药科大学医疗器械学院,沈阳 110016,辽宁
- Publication Type:Journal Article
- Keywords:
pulmonary tuberculosis;
drug resis-tance;
improve machine learning models;
predic-tion;
visualization system
- From:
Chinese Journal of Clinical Pharmacology and Therapeutics
2025;30(5):673-682
- CountryChina
- Language:Chinese
-
Abstract:
AIM:To analyze the clinical value of predicting drug resistance in pulmonary tuberculo-sis patients based on improved machine learning models,and to build a visualization system for veri-fication.METHODS:Retrospectively selected 1 025 pulmonary tuberculosis patients hospitalized in Zhuhai Sixth People's Hospital from March 2019 to March 2024 with drug sensitivity test results as the research object.According to the definition of drug-resistant tuberculosis,the patients were divided in-to 631 sensitive groups(drug sensitivity test results showed no drug resistance),271 RR/MDR groups(meeting the definition of rifampicin resistant tu-berculosis or multi drug resistant tuberculosis,but no drug resistance to any kind of fluoroquino-lones),and 123 pre XDR groups(on the basis of multi drug resistant tuberculosis,and at the same time,drug resistance to any kind of fluoroquino-lones).Analyze clinical data based on the improved machine learning model,help build a drug resistant tuberculosis prediction model,synchronously com-plete feature screening,conduct value analysis on the screened features,and build a visual system for verification.RESULTS:Three groups of patients with baseline data comparison shows:Age,Body mass index(BMI),basic treatment of classification,lung diseases,haemoptysis,second-line drug use history,damage to lung,with empty in all statisti-cally significant difference between the three groups(P<0.05);Based on the modified ma-chine learning model,8 variables were screened,which were history of second-line drug use,BMI,treatment classification,destructive lung,underly-ing lung diseases,cavitation,hemoptysis,and age.The modified machine learning model had the high-est prediction accuracy compared with the tradi-tional model,with AUC values of 0.9322(RR/MDR prediction was positive class)and 0.9545(pre-XDR prediction was positive class).CONCLUSION:The application of the improved machine learning mod-el can help predict the occurrence of drug-resistant tuberculosis and assist the clinical formulation of more effective treatment plans.
