# Prognostic model of HIV-associated talaromycosis in south China: A large-scale retrospective study

## TL;DR

This study creates a reliable tool to predict outcomes for HIV patients with talaromycosis using common clinical data, helping prioritize care in resource-limited areas.

## Contribution

A dynamically validated prognostic model for HIV-associated talaromycosis using routinely available clinical variables and tested across multiple time points.

## Key findings

- Eight independent clinical predictors were identified for poor outcomes in HIV-talaromycosis patients.
- The nomogram showed strong performance with AUC values of 0.905, 0.863, and 0.838 for 7-, 14-, and 28-day outcomes.
- The model demonstrated excellent calibration and clinical utility in both internal and external validation sets.

## Abstract

HIV-associated talaromycosis causes substantial mortality despite available therapies. Early identification of high-risk patients remains challenging, particularly in resource-limited settings. We aimed to develop and validate a dynamic prognostic model for rapid risk stratification.

This retrospective cohort study analyzed 1,892 HIV-talaromycosis patients admitted to Guangzhou Eighth People’s Hospital (2011–2023). Poor outcome (in-hospital death or deterioration-related discharge) was the primary endpoint. A nomogram was developed using Cox regression on admission variables in a training set (2011–2020, N = 1,435), with internal validation set (2011–2020, N = 431) and independent testing set (2021–2023, N = 457). Performance was assessed via time-dependent AUC, C-index, calibration, and decision curve analysis.

Poor outcomes occurred in 14.1% of cases (266/1,892), with 86.5% of these events happening within 28 days. Winter admissions exhibited the lowest case volume but the highest poor outcome rate. Multivariable analysis revealed eight independent readily available predictors: absence of lymphadenopathy (aHR: 0.581, 95%CI: 0.396-0.852, P = 0.005) and hepatosplenomegaly (aHR: 0.347, 95%CI: 0.232-0.519, P < 0.001), respiratory rate (aHR: 1.041, 95%CI: 1.007-1.076, P = 0.016), white blood cell count (aHR: 1.089, 95%CI: 1.049-1.132, P < 0.001), platelet count (aHR: 0.995, 95%CI: 0.992-0.997, P < 0.001), albumin level (aHR: 0.911, 95%CI: 0.872-0.952, P < 0.001), lactate dehydrogenase (aHR: 1.000, 95%CI: 1.000-1.000, P < 0.001), and blood urea nitrogen (aHR: 1.087, 95%CI: 1.068-1.106, P < 0.001). The above indicators were stratified according to predefined classifications and used to established a nomogram. The nomogram demonstrated strong discriminatory performance for 7-, 14-, and 28-day outcomes (AUC 0.905/0.863/0.838 in development; 0.851/0.832/0.807 in independent testing; C-index 0.813-0.841). Calibration curve analysis demonstrated that the nomogram exhibited excellent predictive accuracy and decision curve analysis indicated substantial clinical benefit. The model could effectively differentiate between high-risk and low-risk populations.

This study provides a dynamically validated prognostic tool for HIV-associated talaromycosis, enabling risk stratification using readily available clinical data. Its integration into electronic health systems could off an opportunity to optimize resource allocation and improve outcomes in endemic regions.

HIV-associated talaromycosis continues to cause significant mortality despite current treatment options. Identifying high-risk patients early remains difficult, especially in resource-constrained settings. While prior research has identified certain prognostic factors, these studies typically involved limited sample sizes and relied solely on internal validation without external verification. Existing models also fail to evaluate mortality risk at varying time points, restricting their clinical utility. Our study examined in-hospital prognosis in HIV-associated talaromycosis patients, identifying absence of lymphadenopathy and hepatosplenomegaly, elevated respiratory rate, and abnormal white blood cell count, platelet count, albumin level, lactate dehydrogenase, and blood urea nitrogen as independent predictors of poor outcomes. Using these routinely measured clinical parameters, we developed a dynamic prediction model for risk stratification. Both internal and independent testing showed robust discrimination, high accuracy, and meaningful clinical applicability. This validated prognostic tool could effectively distinguish between high- and low-risk patients, hoping to offer a practical support for clinical decision-making and resource prioritization in endemic areas.