Using propensity scores with multiple categories to assess the effects of HIV-serostatus and post-traumatic stress on cognition
Kathryn Anastos, Department of Medicine/Epidemiology, Montefiore Medical Center & Albert Einstein College of Medicine
Keywords: HIV, propensity scores, PTSD, cognition
Background: Despite the availability of effective therapies for HIV, cognitive complications of the disease remain high. Acute and chronic stressors are common among HIV-infected (HIV+) individuals. These stressors may contribute to cognitive dysfunction that interferes with treatment adherence and daily functioning.
Objective: Using data from the Women’s Interagency HIV Study (WIHS), the nation’s largest prospective multicenter cohort study of HIV+ and at-risk women (HIV-) women, we evaluated the effects of post-traumatic stress (PTSD) symptoms and HIV-serostatus on cognition. Specifically, we were interested in examining whether propensity scores improve adjustment for confounding.
Methods: October 2009 through March 2011, 1004 HIV-infected and 496 at-risk, HIV-uninfected WIHS participants (64% African-American, median age 46) completed a comprehensive cognitive test battery and self-reported stress measures. Cognitive outcomes were measures of verbal learning and memory. Women were considered to have probable PTSD if DSM-IV symptom criteria were met and total severity score exceeded 44 on the PTSD Checklist-Civilian. In the absence of published cognitive test norms for low income minority women, we used a regression-based approach to estimate premorbid levels of cognitive functioning for the total sample based on scores of the HIV uninfected women by regressing age, years of education, Wide Range Achievement Test, and race/ethnicity on each cognitive outcome. For comparison, we also examined raw test scores for each outcome. As a second approach, we generated a propensity score to estimate the probability of group membership (HIV-/PTSD-; HIV-/PTSD+; HIV+/PTSD-; HIV+/PTSD+) using multivariable multinomial logistic regression. Multivariable linear models incorporating propensity scores were used to weight models for both the normalized and raw scores. Results from the propensity score and multivariable models are presented for comparison.
Results: Using multivariable linear regressions on both raw and normative outcomes, HIV+ women performed worse than HIV- women. PTSD was not associated with cognitive outcomes and there were no significant interactions between HIV-serostatus and PTSD on cognitive outcomes. Using inverse propensity score weights on both raw and normative outcomes, HIV+ women performed worse than HIV- women. Notably, women with PTSD performed significantly worse on all cognitive outcomes compared to women without PTSD. There were no significant interactions between HIV-serostatus and PTSD on cognitive outcomes.
Conclusions: Our study illustrates how propensity score methods can reduce bias by minimizing confounding and creating more exchangeable exposure groups. Results using propensity score methods suggest that HIV and PTSD are both associated with verbal learning and memory difficulties among low income minority women. These findings underscore the importance of screening and treating PTSD in both HIV+ and at-risk HIV- women to decrease vulnerability to cognitive dysfunction and related treatment non-adherence.
Important Dates & Deadlines
- October 9 - 11, 2013