Confirmatory factor analyses. As an initial step in testing our model, we used confirmatory factor analyses to test the viability of measurement models for each of our latent variableswithout causal paths between them. For all latent constructs described above, measured variables loaded onto their respective latent variables, with the exception of the adherence latentconstruct. For the conceptualized adherence construct (i.e., five items that make up the TxEQAdherence scale, immunosuppressant target, and MPR) the latter two variables did not loadonto the same factor as the five TxEQ items. As a result, target level and MPR were removedand placed on a second adherence variable; however, these two variables again did not loadonto the same factor and were thus removed from the subsequent SEM analyses. The CFI andRMSEA of the final adherence latent variable were optimal (CFI = 1.00; RMSEA = .005;Table 1 shows lack of significant correlations between the adherence variables). Fit indices forthe remaining latent variables were all acceptable (EPS: CFI = .99; RMSEA = .052; depression:CFI = 1.00; RMSEA < .001; self-efficacy: CFI = 1.00; RMSEA < .001; traditional neurocognition: CFI = .98; RMSEA = .071).