Risk factors and prediction modeling for drug-resistant tuberculosis in Shenzhen: a 9-year retrospective study

Antimicrobial resistance in Mycobacterium tuberculosis (M.tb) poses a major threat to global tuberculosis (TB) control, limiting treatment options and increasing mortality. The prevalence and risk factors for drug-resistant TB (DR-TB) vary by socioeconomic and demographic context. Shenzhen, a rapidly developing megacity in China with a large migrant population, provides a critical setting for investigating this issue. This study aimed to analyze the DR-TB trends and risk factors in Shenzhen using a nine-year retrospective analysis.

We retrospectively analyzed 16,466 TB patients managed at Shenzhen Third People's Hospital from January 2017 and December 2025. Risk factors for DR-TB were identified using univariate and multivariate logistic regression. Based on these factors, we constructed four prediction models (demographic, demographic+clinical symptom, demographic+clinical symptom+diagnostic, and demographic+clinical symptom+diagnostic+treatment delay), and evaluated their performance using Receiver Operating Characteristic (ROC) curves and calibration curves.

The prevalence of DR-TB decreased from 11.36% in 2017 to 10.42% in 2019, dropped sharply to 7.9% in 2020-2021 (likely due to COVID-19 disruptions), rebounded to 14.63% in 2022 (above pre-pandemic levels), peaked in 2023, and then gradually declined after 2024. Multivariate logistic regression identified multiple factors significantly associated with drug resistance: gender (female), age (25-44, 45-64, ≥65 years), occupation (unemployed, family workers, business/service providers, office workers, students), HIV/AIDS comorbidity, clinical symptoms (cough, sputum production), ≥1 negative etiological test result, re-treatment status, concurrent extra pulmonary TB, time from symptom onset to first medical visit (28-42 days), and time from first visit to diagnosis (≥42 days). The AUC values for the four models were 0.571, 0.599, 0.760, and 0.787, respectively (all p < 0.001). Hosmer-Lemeshow χ ² values were 14.014, 13.402, 9.618 and 8.612, respectively (p > 0.05).

We successfully developed a DR-TB risk prediction model with good discrimination and calibration. Early identification of risk factors is crucial for improving clinical awareness and management of DR-TB. Enhancing early diagnosis and treatment is essential for controlling the disease.

https://www.chictr.org.cn/showproj.html?proj=188950, Identifier [ChiCTR2300067585].