Statistical analysis
Sample size and power calculations are based on comparing expected change in medication adherence rate of patients in each group at six months – an expected adherence mean difference of 10 % based on our pilot study findings and the literature. We use an alpha of .05 and provide for 90 % power to detect indicated effect sizes in this two-arm randomized study. An effect size difference of 70 % is based on a conservative estimate of our pilot work. A sample size of 46 older KT recipients per group (final total sample 92) will meet these assumptions and provide sufficient power. Recruitment and retention rates are calculated from our pilot study, other adherence studies at the same sites, and are documented in one similar RCT adherence study in older KT recipients [63, 64]. We selected an adherence rate of 85 % to divide the adherers from the non-adherers based upon our preliminary work describing 4 clusters of KT medication adherers: those who (1) take medications on time (1.0-.85 MA rate), (2) take medications on time with late/missed doses (.84-.70 MA rate), (3) rarely take medications on time and who were late with morning and/or evening doses (.69-.20 MA rate), and (4) missed many doses (.20). Even minor deviations in dosing adherence lead to poor outcomes, though no studies have determined the criterion adherence “dose” that distinguishes good and poor outcomes.
Appropriate descriptive analyses will be performed to examine distributional characteristics for collected measures, as well as to summarize changes over time as a function of group assignment. During this initial phase, we will explore bivariate relationships among primary and secondary outcome measures and variables thought to affect medication adherence. In addition, we will conduct analyses to determine whether the randomly assigned groups are equivalent at the start of the study on the demographic and other measures collected at baseline. Before hypothesis-testing analyses are conducted, exploratory analyses will be performed to examine the effect of various mediators and moderators on the relationship between intervention, adherence, and clinical outcome. The results of these analyses will determine what additional variables will be incorporated in the subsequent hypothesis testing (e.g., analysis of covariance).
Our primary analysis assesses whether the SystemCHANGE™ intervention is more effective than the attention-control intervention in increasing MA in adult kidney transplant recipients at the completion of the 6-month intervention and 6-month maintenance phases. We hypothesize that adult kidney transplant recipients receiving the SystemCHANGE™ intervention will have higher immunosuppressive MA rates than the attention-control group at the completion of intervention and maintenance phases. Since rate responses will most likely violate the normality assumption, the non-parametric method, Mann- Whitney test, will be used for comparing the two groups. However if the normal assumption is satisfied through transformation or as raw data measures, t-test will be applied for group comparison. Possible covariates resulting from demographic data and screening phase MA will be included in the analysis to adjust for possible bias.
Our secondary analysis assesses the MA patterns in both the SystemCHANGE™ and attention-control groups. Specifically we are interested in determining when the intervention becomes effective (e.g., what “dose” is needed) and the pattern of decay in MA over time in both groups. The dependent variable for these research questions are the repeated measurements of immunosuppressive MA rates at 12 time points [i.e. 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11 and 12 months] and the independent variable is group assignment and time effect. Poisson regression analysis will be used for these questions. Proc Nlmixed procedure in SAS will be used for Poisson regression modeling. In order to answer the hypothesis we will test for group-by-time interaction to test if the two groups have different time profiles for MA or not. Possible covariates resulting from demographic data and screening phase MA will be included in the model to adjust for possible bias. Repeated measures from the same Pp will be accounted for using a random effect in the model.
Our exploratory analyses focuses on three aims: 1) to determine whether the SystemCHANGE™ intervention is more effective than the attention-control intervention in decreasing poor health outcomes (e.g. increasing creatinine/BUN, infection, acute/chronic rejection, graft loss, death), 2) to evaluate the role of potential mediators (social support, and systems-thinking) and moderators (ethnicity perceived health and level of medication nonadherence) of MA and health outcomes in adult kidney transplant recipients receiving the SystemCHANGE™ intervention, and 3) to determine if the SystemCHANGE™ intervention is cost-effective.
We expect to observe lower levels of poor health outcomes in the SystemCHANGE™ group as compared to the control group. The dependent variables are the dichotomous outcomes such as, infection, acute and chronic rejection, graft loss, and death and numeric outcomes such as creatinine, and BUN.
The independent variable is group assignment. Chi-square tests will be applied to estimate and test the relationships between dichotomous outcome variables and the independent variable. For continuous outcomes variables t test or Mann-Whitney test will be conducted to test for group effect depending on the satisfaction of normality assumption.
Our secondary exploratory aim is to evaluate the role of potential mediators and moderators of MA and health outcomes in adult kidney transplant recipients receiving the SystemCHANGE™ intervention and recipients receiving the attention-control intervention. We hypothesize that exploring potential mediators and moderators in the analyses will enhance the interpretation of treatment effect on MA. To test for mediator effect, the dependent variable is MA and the independent variable is treatment group. Potential mediators are social support and systems thinking. Poisson regression will be applied to estimate the mediator effect [65] and the Sobel test [66] will be used to test if the mediator effect is significant. Changes in the variance with mediator in the models will be estimated and reported as part of the analysis results. To test for moderator effect, the dependent variable is MA and the independent variable is treatment group. Potential mediators are ethnicity group, perceived health and MA level (different strata). Poisson regression will be applied to estimate and test for possible moderator effect through interaction terms between the independent variable and moderator [65].
Our third exploratory aim is to determine if the SystemCHANGE™ intervention is cost-effective. Our hypothesis is the cost for the SystemCHANGE™ intervention will be less than the cost for the attention-control intervention. To determine the cost-effectiveness of the SystemCHANGE™ intervention compared to the attention-control intervention, both intervention and resource use costs will be evaluated and compared to MA change. A cost-effectiveness analysis will be performed at the end of the intervention period and again at the end of the maintenance period. If there is no treatment effect, a cost-analysis will not be performed. The analysis performed at the end of the maintenance period will be cumulative, incorporating costs and benefits incurred throughout the project. A cost-effectiveness analysis, performed at the end of the maintenance period (calculated for both the intervention and maintenance periods), will evaluate both intervention and control, and resource use costs which will be compared to adherence change. The sum of the total intervention and control costs and resource use costs will be the numerators for testing this hypothesis. The change in adherence (from baseline to end of intervention [or end of maintenance] period) will be the denominator. We will identify all direct intervention costs related to the intervention and the control (planning, designing, and implementation of each intervention, personnel, supplies, travel, and equipment). We will identify resource use costs (hospitalizations, clinic, observation, and ER visits) for both groups. Resource use costs will be obtained from publically available data, e.g. Hospital Compare, Hospital Stats, H-CUP. The DRG will be obtained with a conversion rate and then adjusted by hospital specific information, e.g. academic, location.