Abstract
INTRODUCTION: Despite advances in HIV treatment, low- and middle-income countries face challenges due to reduced funding for diagnostic and treatment programs. People living with HIV/AIDS (PLWHA) remain at high risk of severe complications and ICU admission. Prognostic scoring systems, an important benchmark of quality in ICU care, often lack validation for this population and specific clinical timeframes. OBJECTIVE: To evaluate the performance of seven prognostic scoring systems within the first 24 and 72 h of ICU admission among PLWHA. METHODS: This retrospective cohort study was conducted at Fundação de Medicina Tropical Doutor Heitor Vieira Dourado (Manaus, Brazil) and included 103 adult PLWHA. Prognostic models (APACHE II, SAPS III, ODIN, MPM II, SOFA, and MODS) were assessed using ROC curves and the Hosmer-Lemeshow test for calibration. RESULTS: The study cohort had a median age of 34 years (IQR: 28-43), with 76.7% male patients. Most were non-adherent (54.4%) or irregularly adherent (22.3%) to antiretroviral therapy (ART). The median CD4+ count was 57 cells/µL (IQR: 18-179), and 83.1% had a detectable viral load (> 50 copies/mm³). In the first 24 h, five systems demonstrated modest discriminatory power (MPM II 24 h: AUC 0.665; ODIN: 0.649; APACHE II: 0.619). At 72 h, MPM II 72 h had the highest performance (AUC 0.802), followed by MODS (0.780) and SOFA (0.712). All models demonstrated good calibration, with the MPM II 72 h having higher sensitivity and accuracy. CONCLUSION: The systems demonstrated limited performance in the first 24 h, but MPM II 72 h showed better reliability and discrimination. These findings highlight the need for validated prognostic tools tailored to PLWHA in resource-limited settings. CLINICAL TRIAL NUMBER: Not applicable.