As shown in , younger (age <50 years) and older (age ≥50 years) participant groups were not significantly different in terms of gender, race/ethnicity, education, mean estimated verbal intelligence quotient, most recent CD4 count, months on HAART, number of total medications (HAART and non-HAART), mean BDI-II total score, or prevalence of alcohol dependence. However, more individuals in the younger group met criteria for AIDS and current drug dependence.
Proportions of medication classes tracked by MEMS (protease inhibitor, nucleoside reverse transcriptase inhibitor, nucleotide reverse transcriptase inhibitor, and nonnucleoside reverse transcriptase inhibitor) were not significantly different between the age groups (χ2(3) = 0.845, p = 0.84). Likewise, the proportion of younger versus older adults on once-daily (younger: 92, 26.14%; older: 11, 13.92%), twice-daily (younger: 240, 68.18%; older: 62, 78.48%), and three-times-daily (younger: 20, 5.68%; older: 6, 7.59%) dosing schedules for these medications was not significantly different (χ2(2) = 5.40, p = 0.07).
Greater variability for MEMS and 30-day self-report measures was found among younger than among older adults (Levene’s F
[1, 429] = 15.58, p <0.001, and Levene’s F
[1, 414] = 13.99, p <0.001, respectively). As shown in , older adults demonstrated significantly better adherence than younger adults on these measures when equal variances were not assumed. No significant differences were found between age groups for qualitative self-report of adherence. The proportion of individuals reporting a missed dose in 1-day self-report was also not significantly different between younger (12.68%) and older (9.09%) participants, χ2(1) = 0.77, p = 0.38.
Descriptive Statistics and t Tests of Adherence to Medication
Independent-samples t tests for each of the six demographically corrected cognitive domain t scores (presented in ) demonstrated no significant between-group mean differences. However, greater variability was found in the older group for learning/memory (Levene’s F
[1, 429] = 4.17, p = 0.04) and verbal fluency (Levene’s F
[1, 429] = 47.49, p <0.001).
Descriptive Statistics and t Tests of Cognitive Domains
CFA of Cognition and Adherence
A CFA was conducted for cognitive and adherence variables so that comparisons of these constructs could be made between the younger and older age groups. CFA models consisted of two correlated latent factors: “Cognition“ and “Adherence.” Indicators of Cognition included standardized t scores of processing speed, memory, attention, executive functioning, verbal fluency, and motor functioning (tests for each domain are identified in the ). Indicators of Adherence included MEMS adherence, qualitative self-report, 30-day self-report, and 1-day self-report. CFA models were first evaluated separately for the younger and older age groups (Models A and B). Subsequently, data from both groups were modeled simultaneously (Model C0).
CFA tests of invariance were then conducted, comparing absolute fit between nested versions of Model C by imposing successive constraints on the equality of measurement weights (C1), measurement intercepts (C2), and structural covariances (C3) between age groups. A significant change in fit between models C0 and C1 (Δχ2(8) = 29.99, p <0.001) demonstrated differences in factor structure between groups. Further examination demonstrated that the factor structure of cognition did not differ significantly between age groups, although 30-day and 1-day self-report were weaker indicators of medication adherence for older than for younger participants. In addition, as demonstrated by the change in fit between models C2 and C3(Δχ2(3) = 60.73, p <0.001), the covariance of cognition and adherence differed significantly between age groups. Fit statistics for selected latent models are shown in .
Goodness-of-Fit Statistics for Latent Models of Cognition and Adherence
Relationship between Cognition and Adherence by Age Group
Model C0 was examined further to obtain estimates of the relationship between cognition and adherence. Although Cognition and Adherence were not significantly related in the younger group, r = 0.02, z = 0.23, p = 0.82, these latent factors were strongly related in the older group, r = 0.49, z = 2.57, p = 0.01. This model is represented in .
Latent Model of Cognition and Medication Adherence Among Younger and Older HIV+ Adults
Additional models were tested to more closely examine cognition-adherence relationships. In Model D, the latent Adherence factor was permitted to covary freely with each of the six individual indicators of cognition. Adherence was not significantly related to any indicators of cognition in the younger group. However, in the older group, Adherence was related to executive functioning (r = 0.45, z = 2.50, p = 0.01), motor functioning (r = 0.46, z = 2.57, p = 0.01), and processing speed (r = 0.35, z = 2.09, p = 0.04). In Model E, the latent Cognition factor was permitted to covary freely with each of the three individual indicators of adherence. Cognition was not significantly related to any indicators of adherence in the younger group, but Cognition was significantly related to both MEMS adherence (r = 0.38, z = 2.85, p = 0.004) and qualitative self-report of adherence (r = 0.31, z = 2.40, p=0.02) in the older group.
Our findings of differential cognition-adherence relationships cannot be accounted for by the previously described between-group differences in variability, as adherence variability was lower in the group demonstrating stronger relationships (older adults). Likewise, the cognitive domains with greater variability among older adults (learning/memory and verbal fluency) were not individually related to adherence in either age group and, therefore, do not appear to be driving the overall cognition-adherence relationship.
Influence of Health Factors
Individual follow-up analyses were also conducted to evaluate and control for relevant health-related factors, including CD4 count, alcohol problems, drug problems, depression, hepatitis C, history of learning disabilities or neurological illness, and MEMS regimen complexity. Models were tested controlling for the potential influence of each of these variables on the relationship between Cognition and Adherence. In Model F, higher CD4 count was associated with better Adherence (β = 0.38, z = 2.53, p = 0.01), but only for individuals in the older group. In Model G, greater drug problems were also a strong predictor of poorer adherence in the older group (β = −0.59, z = −4.31, p <0.001). In Model I, increased depression (BDI-II score) was significantly related to poorer Adherence (β = −0.14, z = −2.29, p = 0.02) and Cognition (β = −0.14, z = −2.41, p = 0.02) in the younger group only. The magnitude of the depression-adherence relationship was similar among older adults (β = −0.25, z = −1.69, p = 0.09) but failed to reach statistical significance. In Model J, greater MEMS regimen complexity (dosing frequency) was significantly associated with poorer adherence in the younger group (β = −0.19, z = −3.22, p = 0.001). No other control variables were significantly associated with adherence or cognition in either age group. Notably, the cognition-adherence relationship was not meaningfully affected in any control analyses.