Machine Learning-Based HIV Risk Estimation Using Incidence Rate Ratios.

Area of Science:

• Medical informatics
• Machine learning applications in healthcare
• Epidemiology of infectious diseases

Background:

• HIV/AIDS remains a global pandemic, necessitating early detection for improved patient outcomes and infection prevention.
• Point-of-care diagnostics and risk estimation tools are crucial for timely interventions and broader population reach.
• Objective HIV risk assessment can guide healthcare providers in decisions regarding HIV testing and pre-exposure prophylaxis (PrEP).

Purpose of the Study:

• To propose and validate a novel machine learning method for estimating patient HIV risk using historical clinical data.
• To develop an objective, data-driven approach for HIV risk assessment independent of expert opinions.
• To support clinical decision-making at the point-of-care for HIV testing and PrEP initiation.

Main Methods:

• A machine learning approach based on association rules derived from clinical data.
• Calculation of incidence rate ratios (IRR) for each derived rule.
• Mean IRR of applicable rules used to estimate individual patient HIV risk.
• Validation on the MIMIC-IV database (approx. 525,000 hospital stays).

Main Results:

• The proposed method achieved an Area Under the Curve (AUC) of 0.88 with 53 association rules.
• A threshold of 0.66 yielded a sensitivity of 98% and a specificity of 53%.
• Identified risk factors included drug abuse, psychological illnesses (e.g., PTSD), known conditions (e.g., pulmonary diseases), and diagnostic procedures.

Conclusions:

• A novel, data-driven machine learning method for HIV risk estimation has been developed.
• The method leverages extensive clinical data, providing objective, expert-independent risk assessments.
• This tool aids healthcare providers in making informed decisions for HIV testing and PrEP within point-of-care settings.