for men. The likelihood ratio test for the gender-homeless interaction term was not significant (p=0.28). However, the adjusted OR for ART utilization, homeless vs. not homeless, was calculated to be 0.16 (95% CI 0.06-0.38) for women and 0.41 (95% CI 0.18-0.94) for men. Aim 3 The Xxxxxx-Xxxxxxxx p-value was 0.91, suggesting that the model fit was good. The area under the curve (AUC) for Atlanta was 0.814 and the AUC for Miami was 0.577 (Figure 3). In the second model, the variable for visits to HIV primary care was added into the prediction model. The Xxxxxx-Xxxxxxxx p-value was 0.54, suggesting a good fit for the model. The AUC for Atlanta was 0.840, and the AUC for Miami was 0.678 (Figure 4). Aim 4 The frequency of covariates comparing ART users with undetectable vs. detectable HIV-1 viral load can be found in Table 12, Table 13, Table 14, Table 15, and Table 16. Among the covariates, there appeared to be less frequent crack and alcohol use among individuals with undetectable viral load (p<0.05). The variables loaded into the multivariable model included: age, gender, race, city, education, alcohol use, marijuana use, crack use frequency, depression, HIV knowledge, patient-provider relationship, trust, and homelessness. None of the examined covariates was independently associated with viral load suppression.
