Icates standard deviation; IVDA intravenous drug abuse; MSM, men who have sex with men. a Value closest to date of survey completion, 630 days; CD4 cell count available for 85 of participants. doi:10.1371/journal.pone.0054729.tWe first assessed the relationship between buy Emixustat (hydrochloride) retention in HIV care, adherence to HAART and HIV suppression, controlling for age, race, ethnicity, IQ1 depression and health status. This constituted the baseline model. Next, we included overall patient satisfaction as a predictor latent variable to determine its effect on the relationship between retention, adherence, and, ultimately, HIV suppression. We tested the hypothesized models using SPSS AMOS 19.0 statistical software (SPSS Inc, Chicago, IL). We performed hypothesis testing by examining parameter estimates. The retention in HIV care and HIV suppression constructs were measured with single indicators. Since HIV RNA copies is the accepted standard measure of HIV suppression, the measurement loading for HIV suppression was set to 1.00 (i.e. no measurement error). Since no studies of reliability have been reported for the retention in HIV care construct and the construct is measured objectively, its measurement error was assumed to be 0 and the measurement loading was set to 1.00. The adherence to HAART construct has an estimated reliability of 0.67 (personal communication, Y. Lee, 2012). This was incorporated into the model by setting the measurement loading to 0.82 (the square root of the reliability 0.67) and the measurement error to 0.33 (1 minus the reliability 0.67). Model goodness-of-fit was evaluated using 3 widely used indexes: chi-square test (x2), the Comparative Fit Index (CFI) and Root Mean Square Error of Approximation (RMSEA) [14]. We 15481974 used conventional cutoff criteria for fit indexes: 1) nonsignificant x2 values, 2) CFI values .0.90 [25] or .0.95 [26], and 3) RMSEA values ,0.06 [26] or ,0.08 [27]. The Institutional Review Board (IRB) for Baylor College of Medicine and Affiliated Institutions approved this study. The IRB waived the need for written informed consent because this research involves no more than minimal risk to the participants. We collected verbal informed consent and documented the procedure. All data were de-identified and analyzed anonymously.evaluation of direct, indirect and total effects of multiple variables, and 3) accounts for measurement error in the process of modeling relationships between latent variables (i.e. variables that are not directly observed, but estimated from directly measured ones). Spearman’s partial correlation coefficients were calculated for all measures in the structural modeling by controlling for age, race, ethnicity, depression and health status. These computations parcel out the shared variance between each control variable and pair of measures. The resulting partial correlation matrix was used as the input for the structural model 
estimation (Table 2). Missing data were treated by pairwise deletion. The correlations between clinic sites were comparable.Results Study populationThe study sample includes 489 patients (94 of eligible patients approached; 388 from TSHC and 101 from VAMC). As shown in Table 1, the mean age was 48 years, 71 were men, 61 
were non-Hispanic black, and 54 had a household income of # 10,000. Participants and eligible non-participants did not differ significantly in terms of age, race, sex, and ethnicity (data not shown).Patient Satisfaction to Improve HIV AdherenceOverall patien.Icates standard deviation; IVDA intravenous drug abuse; MSM, men who have sex with men. a Value closest to date of survey completion, 630 days; CD4 cell count available for 85 of participants. doi:10.1371/journal.pone.0054729.tWe first assessed the relationship between retention in HIV care, adherence to HAART and HIV suppression, controlling for age, race, ethnicity, depression and health status. This constituted the baseline model. Next, we included overall patient satisfaction as a predictor latent variable to determine its effect on the relationship between retention, adherence, and, ultimately, HIV suppression. We tested the hypothesized models using SPSS AMOS 19.0 statistical software (SPSS Inc, Chicago, IL). We performed hypothesis testing by examining parameter estimates. The retention in HIV care and HIV suppression constructs were measured with single indicators. Since HIV RNA copies is the accepted standard measure of HIV suppression, the measurement loading for HIV suppression was set to 1.00 (i.e. no measurement error). Since no studies of reliability have been reported for the retention in HIV care construct and the construct is measured objectively, its measurement error was assumed to be 0 and the measurement loading was set to 1.00. The adherence to HAART construct has an estimated reliability of 0.67 (personal communication, Y. Lee, 2012). This was incorporated into the model by setting the measurement loading to 0.82 (the square root of the reliability 0.67) and the measurement error to 0.33 (1 minus the reliability 0.67). Model goodness-of-fit was evaluated using 3 widely used indexes: chi-square test (x2), the Comparative Fit Index (CFI) and Root Mean Square Error of Approximation (RMSEA) [14]. We 15481974 used conventional cutoff criteria for fit indexes: 1) nonsignificant x2 values, 2) CFI values .0.90 [25] or .0.95 [26], and 3) RMSEA values ,0.06 [26] or ,0.08 [27]. The Institutional Review Board (IRB) for Baylor College of Medicine and Affiliated Institutions approved this study. The IRB waived the need for written informed consent because this research involves no more than minimal risk to the participants. We collected verbal informed consent and documented the procedure. All data were de-identified and analyzed anonymously.evaluation of direct, indirect and total effects of multiple variables, and 3) accounts for measurement error in the process of modeling relationships between latent variables (i.e. variables that are not directly observed, but estimated from directly measured ones). Spearman’s partial correlation coefficients were calculated for all measures in the structural modeling by controlling for age, race, ethnicity, depression and health status. These computations parcel out the shared variance between each control variable and pair of measures. The resulting partial correlation matrix was used as the input for the structural model estimation (Table 2). Missing data were treated by pairwise deletion. The correlations between clinic sites were comparable.Results Study populationThe study sample includes 489 patients (94 of eligible patients approached; 388 from TSHC and 101 from VAMC). As shown in Table 1, the mean age was 48 years, 71 were men, 61 were non-Hispanic black, and 54 had a household income of # 10,000. Participants and eligible non-participants did not differ significantly in terms of age, race, sex, and ethnicity (data not shown).Patient Satisfaction to Improve HIV AdherenceOverall patien.
