Factors influencing late HIV presentation in China: results from logistic regression and Bayesian network analyses

TL;DR

This study identifies factors that contribute to late HIV diagnosis in China, using statistical and network analyses to inform better testing and intervention strategies.

Contribution

The study combines logistic regression and Bayesian network analysis to reveal complex interrelationships among factors influencing late HIV presentation.

Findings

Age, gender, and sample sources were key factors in late HIV presentation, with age having the most central role.

Education and ethnicity indirectly influenced late diagnosis through their effects on occupation and transient status.

A history of STDs increased the risk of late HIV diagnosis both directly and indirectly through transient population status.

Abstract

Late presentation (LP) of HIV infection remains a major challenge to epidemic control, leading to advanced immunodeficiency, poorer treatment outcomes, and ongoing transmission before diagnosis. Despite expanded testing and awareness efforts, a considerable proportion of people living with HIV (PLHIV) in China are still diagnosed late.

This study analyzed 386,704 newly reported HIV cases (2019–2022) from the National HIV/AIDS Comprehensive Response Information Management System (CRIMS). Logistic regression was used to identify significant predictors of LP, and a Bayesian network was constructed to model the complex interrelationships among variables.

Logistic regression identified several factors associated with LP of HIV. Key factors included being male (aOR = 1.3), over 60 (aOR = 3.36), Han ethnicity (aOR = 1.16), education at below senior high school (aOR = 1.1), being a farmer or worker (OR = 1.04), transient population (aOR = 1.18), engaging in homosexual transmission (aOR = 1.1), being examined at other institutions (hospitals) (aOR = 1.27), having non-marital partners(lifetime history), and a history of STDs (aOR = 1.03) (P < 0.05). Bayesian network analysis revealed that age, gender, and sample sources were the key factors associated with LP of HIV. Among them, age played a central role in the model, directly influencing occupation, transmission routes, education level, transient status, and non-marital partners. Ethnicity indirectly affected LP through occupation, while education influenced LP indirectly by shaping occupation, non-marital partners, and transient status. In addition, a history of STDs not only directly affected sample sources but also indirectly increased the risk of LP through transient population status.

This mixed-model approach demonstrated that demographic, behavioral, and structural factors jointly contribute to LP in China through complex associative pathways. Integrating logistic and Bayesian frameworks provides a more comprehensive understanding of HIV diagnostic delays, informing precision-targeted testing and intervention strategies.

The online version contains supplementary material available at 10.1186/s12879-025-12429-6.

Figures