Using propensity scores to assess the relationship between HIV status and acute myocardial infarction in the Veterans Aging Cohort Study
*Donna Almario Doebler, University of Pittsburgh, Graduate School of Public Health 
Adeel A. Butt, University of Pittsburgh, School of Medicine; VA Pittsburgh Health Care System 
Chung-Chou H Chang, University of Pittsburgh, School of Medicine 
Heidi M. Crane, University of Washington, Medicine 
Matthew S. Freiberg, University of Pittsburgh, School of Medicine 
Matthew B. Goetz, UCLA School of Medicine; VA Greater LA Health Care System 
Amy C. Justice, Yale School of Medicine, VA CT Health Care System 
Kevin H. Kim, University of Pittsburgh, School of Medicine 
Kathleen A. McGinnis, VA Pittsburgh Healthcare System 
Jasoon J. Sico, Yale School of Medicine, VA CT Health Care System 
Roslyn A. Stone, University of Pittsburgh, Graduate School of Public Health 
Hilary Tindle, University of Pittsburgh, School of Medicine 

Keywords: propensity scores, HIV, incident acute myocardial infarction, Veterans

Background: The objective of this study was to examine the association between HIV status and clinically confirmed incident acute myocardial infarction (AMI) in the Veterans Aging Cohort Study-Virtual Cohort-Ischemic Heart Disease Quality Enhancement Research Initiative (VACS-VC-IH-QUERI) using propensity score (PS) adjustments to account for selection bias.

Methods: The VACS-VC-IH-QUERI includes a large cohort of HIV+ Veterans matched to HIV- Veterans on age, gender, race/ethnicity, and site. A PS for HIV+ status was estimated using logistic regression; the PS included 14 HIV and/or AMI-related risk factors (age, race, smoking status, BMI, kidney function, cholesterol, antihypertensive medication use, systolic and diastolic blood pressures, diabetes status, cocaine use diagnosis, alcohol abuse or dependence diagnosis, hepatitis B and C infection). The hazard of incident AMI was predicted from HIV status using separate Cox regression models: (1) unadjusted, (2) covariate-adjusted for the 14 risk factors, (3) stratified on PS quintiles, (4) weighted by the PS, and (5) restricted to individually PS-matched pairs of HIV- and HIV+ Veterans.

Preliminary Results: This cohort included 34% HIV+ (n=21,320) and 66% HIV- (n=41,399) Veterans. The restricted matched sample retained almost all HIV+ (n=21,312) and 31% of HIV- (n=12,792) Veterans. Standardized covariate differences between HIV+ and HIV- Veterans ranged from 3%-70% in the full sample and decreased to 0%-2% in the PS matched sample. The hazard ratio (HR) for AMI associated with HIV+ status was 1.78 (95% CI: 1.45-2.18) in the unadjusted model, 1.90 (95% CI: 1.52-2.38) after covariate adjustment, 2.00 (95% CI: 1.58-2.53) in the stratified model, 1.73 (95% CI: 1.39-2.15) after PS weighting, and 2.19 (95% CI: 1.59-3.02) in the PS-matched sample.

Conclusions: PS adjustment appears to have eliminated observed baseline differences between HIV+ and HIV- Veterans. However, PS adjustment may not account for unobserved confounders. In this cohort, HIV+ status is associated with significantly increased risk of incident AMI, with estimated HRs ranging from 1.73 to 2.19 across the alternative versions of PS adjustment considered. The largest HR was estimated in the matched sample, which excluded a majority of HIV- Veterans who did not resemble HIV+ Veterans. HIV+ status itself, rather than characteristics of HIV+ patients, may act on the biological mechanisms leading to AMI or otherwise increase the risk of AMI.