Abstract P275: Long-Term Mortality Prediction Model for HIV Patients With Heart Failure: Analysis From a Multi-Center Cohort Database in New York City Health + Hospitals

Abstract

Introduction: Given the intricate interplay of HIV infection and heart failure (HF), conventional mortality prediction models for general heart failure may prove insufficient for this specific population. Our objective is to assess the fitness of current HF death prediction models and develop a prediction model tailored to the unique challenges of long-term mortality risk in HIV patients with heart failure. Method: The statistical models were developed through New York City Health + Hospitals HIV Heart failure (NYC 4H) retrospective cohort. Demographic and medical factors were obtained through chart reviews, while social adversities were assessed via licensed social workers' psychosocial assessment notes. Patients with baseline cancer, no baseline ejection fraction (EF), or those below 18 years of age were excluded from the study. Multivariable logistic regression models were constructed to predict the risk of overall death. Model 1 was derived from the existing MAGGIC Risk prediction model for general heart failure patients. Model 2 employed backward step-wise selection from NYC 4H cohort for HIV HF biologic factors, functional status, medical and interventional treatment to predict mortality risk. Model 3 further adjusted for single significant social factors and insurance. Model comparison was performed using the area under the ROC curve. Result: The study comprised 597 HIV heart failure patients, averaging 61.1 years in age. Over the 4-year follow-up, 102 deaths (17.1%) were recorded. Model 1 (MAGGIC Risk prediction model) had an AUC of 0.72. Model 2, which included backward-selected variables such as age, race, HIV viral load at baseline and follow-up, CD4 count at baseline, EF at baseline, smoking status, chronic obstructive pulmonary disease, pulmonary hypertension, peripheral arterial disease, chronic kidney disease, activities of daily living, obesity, implantable cardioverter-defibrillator implantation, and the total number of guideline-directed medical therapy, significantly improved the AUC to 0.85. Incorporating social factors such as exposure to any lifetime adversities, specific adversities (lack of family support, polysubstance abuse, and mental illness), and insurance status, model 3 raised the AUC to 0.88. In heart failure with reduced EF (HFrEF) subset (n=502), Model 2 and Model 3 achieved AUCs of 0.88 and 0.90, respectively. Conclusion: HIV heart failure patients exhibit unique mortality prediction factors compared to general heart failure patients without HIV infection. Considering the high burden of social adversities, socioeconomic factors can enhance the predictive yield of models within this specific population.