Systematic review and meta-analysis of risk prediction models for HIV testing in key populations
Xin Xie 1, 2, Chaowen Zhang 3, Shuyu Han 3, Yirong Shi 4, Jinyang Ming 3, Weimei Chen 4, Jingxian Hu 2, Bo Zhou 3, Lili Zhang 2
1 School of Nursing, Capital Medical University, Beijing, China; 2 Department of Nursing, Beijing Youan Hospital Affiliated with Capital Medical University, Beijing, China; 3 Department of Nursing, School of Nursing, Peking University, Beijing, China; 4 Department of Infection and Immunology, the Third People’s Hospital of Shenzhen, Shenzhen, China
*Correspondence: Shuyu Han. Email: 2116393033@bjmu.edu.cn
Abstract
HIV testing is a critical tool for preventing HIV transmission, with early identification in key populations reducing onward spread. This study aims to evaluate risk prediction models, identify factors influencing HIV testing, and provide recommendations to enhance testing among key populations. Electronic databases were searched for peer-reviewed and gray literature published in English and Chinese from January 1, 1996, to November 14, 2025. Two reviewers independently assessed methodological quality and extracted data. The prediction model risk of bias assessment tool was used to evaluate bias and applicability. Of 3693 initially identified studies, seven met the inclusion criteria. Reported area under the curve (AUC) values ranged from 0.72 to 0.82, and the pooled AUC of validated models was 0.77 (95% confidence interval: 0.70-0.84), indicating moderate discriminative performance. However, most of the studies were assessed as having a high risk of bias, primarily due to insufficient reporting in the analysis domain, limiting the reliability of existing models for clinical or public health applications. Predictors of HIV testing were broadly grouped into sociodemographic, behavioral, knowledge-related, and structural factors, and predictors of HIV testing varied considerably across models, reflecting differences in study populations, behavioral characteristics, and contextual factors across settings. Overall, although existing models demonstrate moderate predictive ability, their methodological limitations and lack of external validation restrict their generalizability and practical utility. Therefore, reliable prediction models remain limited. Future research should develop high-quality models with larger sample sizes, robust designs, and multi-center external validation to support clinical application, improve practical relevance, and inform strategies to advance HIV testing. Beyond HIV testing prediction, this study highlights the need for broader prevention approaches, including sexuality education, risk awareness, and safe sexual behaviors, alongside targeted testing interventions.
