«Persons using assistive technology may not be able to fully access information in this file. For assistance, e-mail niddk-cr Include the ...»
Research Question 2 will be addressed with analyses for each of the Periods 1, 2 and 3. For the renal clinical event outcomes, unadjusted risk ratios for baseline factors will first be obtained using Cox models with the baseline factors and indicators for the randomized groups (see Basic Renal Analytic Approach) as predictor variables. Subsequently, adjusted risk ratios will be obtained by adding other baseline factors (such as clinical center, demographic factors, or previously identified predictors of renal progression) to the models. Nonlinear relationships between a factor and risk of the clinical outcomes will be evaluated using splines or step functions (Therneau 2000). The possibility that a factor may have different risk ratios at different times (nonproportional hazards) will be investigated with residual plots and by interaction terms with follow-up time (Therneau 2000B). Interactions of baseline predictor variables with each other and with the treatment groups will be investigated by adding interaction terms to the Cox models. In particular, it will be of interest to test whether the association of renal events with baseline factors is modified by the initial level of GFR and/or proteinuria.
The association of follow-up factors with renal events will be investigated with Cox models that include time-dependent terms for the follow-up factors plus additional terms to adjust for the treatment groups and relevant baseline factors (see Basic Renal Analytic Approach). Depending on the application, the hazard rate at a particular time may be modeled as a function of
• The cumulative mean of the factor being investigated (averaging all measurements from baseline to the current time),
As for the renal events, the association of GFR decline (or predicted GFR decline) with specific baseline factors will be evaluated first with adjustment only for the randomized treatment groups, and then with adjustment also for relevant baseline factors. The adjustment for baseline factors will usually be accomplished by adding these factors and their interactions with follow-up time to the mixed effects models.
Several strategies will be used to relate the decline in GFR (or predicted GFR) to factors measured during follow-up. For simplicity, in many cases all follow-up measurements of a factor will be averaged to obtain a single mean value for each patient that can be entered in mixed effects models along with its interaction with follow-up time. Weighted averages may be used to account for the spacing between successive measurements. This type of model evaluates the cross-sectional association between the follow-up factor and GFR decline, and does not account for the temporal order of the measurements. Alternative models will also be considered in which the mean rate of GFR (or predicted GFR) decline between months t and t + 6 is related to the cumulative mean of the predictor variable up to month t or to the most recent measurement of the predictor variable prior to time t. Further terms will be added to distinguish between short-term acute effects and long term effects for factors whose changes are expected to cause hemodynamic changes in GFR.
Comparisons of Change in GFR to Change in Serum Creatinine
The analysis plans for Research Questions 2 and 3 depend on the validity of using serum creatinine in place of GFR in the definition of the renal outcomes. Therefore, during the initial months of the AASK Cohort Study, the data from the clinical trial phase of the study will be used to clarify the relationship between GFR and creatinine-based outcomes. The timing of GFR-based outcomes (G1 and G2) will be compared to creatinine-based outcomes (S1) and S2).
Joint mixed effect models including both the change in GFR and the change in the AASKpredicted GFR from serum creatinine will be used to evaluate the association of changes in GFR with changes in predicted GFR. The model of Schluchter will be used to compare the correlation of GFR slope with time-to-renal failure or death to the correlation of predicted GFR slope with time-to-renal failure or death. This will provide insight into the validity of GFR slope and
As described in the General Renal Analytic Approach, long-term effects of the AASK interventions on the renal clinical outcomes will be tested by time-dependent indicator variables for the randomized treatment groups in the Cox models which distinguish between the effects of the interventions during the randomized trial and the long-term effects following the end of the trial. The long-term effects of the interventions on predicted GFR slope will be evaluated by adding linear spline terms to the mixed effects models to allow comparison of mean predicted GFR slopes between the original randomized groups after the termination of the trial.
Research Question 4. Does the development of proteinuria predict the progression of kidney disease?
As an outcome variable, change in proteinuria will be analyzed using methods similar to those used for change in GFR. Proteinuria will be expressed as a urine protein/creatinine ratio as in the randomized trial, and log transformed due to positive skewness. Since the mean changes in the log urine protein/creatinine ratio are nonlinear, the mixed effects models will include spline terms allowing different slopes over each 6- or 12-month interval between scheduled urine protein/creatinine measurements.
The effect of proteinuria on subsequent renal clinical events will be evaluated first by Cox models relating baseline proteinuria to each of the renal outcomes during the follow-up period of the analysis (either Period 1, Period 2, or Period 3). Subsequently, models examining the association of the change in proteinuria during the first 6 months (or 12 months for Period 3 analyses) of the period of analysis will be related to rates of the clinical events after the first 6 (or
12) months. As for other risk factors, these models will control for the randomized treatment groups, and will be carried out both with and without adjustment for other potential risk factors.
Due to interactions between proteinuria and the randomized treatment comparisons identified in October 1, 2002 47 the randomized trial, relevant interaction terms between proteinuria and the treatment groups will be considered. A similar approach will be used to relate baseline proteinuria and initial changes in proteinuria to subsequent GFR slope (Period 1) and to predicted GFR slope (Periods 2 and 3).
Further longitudinal analyses, with proteinuria modeled as a time-dependent covariate, will be used to investigate whether GFR slope changes over time for individual patients following increases in proteinuria.
Rates of renal clinical events and predicted GFR slope will be compared during Period 3 between the AASK cohort participants and African Americans with hypertensive kidney disease enrolled in the Chronic Renal Insufficiency Cohort Study. Relevant baseline factors will be included as covariates to control for initial differences between the two groups. We expect that these comparisons will be carried out using Cox models and mixed effects models similar to those described above, but the specific analyses will developed jointly by the AASK Cohort and the Chronic Renal Insufficiency Cohort investigators.
Research Question 6: What comorbidities, particularly cardiovascular disease, occur in the setting of hypertension-related kidney disease?
Incidence of cardiovascular events will be recorded, and expressed per-patient year of follow-up.
Research Question 7: What risk factors predict the occurrence of cardiovascular disease?
Cox regression models analogous to those employed for the renal outcomes will be used in Periods 1, 2, and 3 to relate baseline and follow-up predictor variables to the incidence of various classes cardiovascular events defined by the Outcome Committee. Analyses will typically be structured to determine the additional predictive values of non-traditional cardiovascular risk factors (e.g., homocysteine and CRP) beyond those of traditional cardiovascular risk factors (smoking, diabetes, blood pressure, LDL and HDL). Left ventricular mass will be evaluated both as a risk factor and as an outcome variable for Period 3 analyses. When treated as an outcome variable, mixed effects models will be used to relate the slope of left ventricular mass
Standard techniques will first be used to evaluate changes in quantitative and categorical factors between their final pre-ESRD measurement and post-ESRD assessments for those patients who initiate dialysis during the cohort study. Assessment of these changes between the last preESRD measurement and post-ESRD measurements is complicated by the possibility that the parameters of interest may change after the last pre-ESRD measurement but prior to the transition to ESRD. Thus, we will also consider longitudinal analyses relating the metabolic and cardiovascular-renal risk factors to follow-up time, with initiation of ESRD entered as a time dependent covariate. In this way, the effect of initiation of dialysis can be evaluated after controlling for the rate of change in the factors being analyzed prior to ESRD.
Projected Follow-up and Power Methods of Projecting Numbers of Future Events: The statistical power for the main time-toevent analyses in Periods 1, 2, and 3 depends on the numbers of events for the respective composite outcomes during these periods.
For Period 1 (the AASK trial through September 30, 2001), the number of events were calculated based on the actual reported numbers of events during the trial.
For Period 3 (the AASK Cohort period from February 1, 2001 though June 30, 2007), the numbers of events were estimated using a 2-stop process. In the first step, Weibul models were used to estimate event probabilities for each composite outcome as a function of the initial prerandomization baseline urine protein/creatinine and baseline GFR for AASK patients in the ACE and Beta Blocker Groups. These models were then applied to the most recent urine protein/creatinine and GFR values to project future event rates for likely AASK Cohort participants in all three treatment groups. Likely AASK Cohort participants were identified as those patients who were alive and had not reached renal failure as of September 14, 2001 and who Clinical Centers indicated were not lost to follow-up and were likely to agree to participate in the AASK Cohort. An additional 3% loss-to-follow-up was assumed for each outcome.
October 1, 2002 49 Projection of future events for the Period 2 analyses had to account the definition of the creatinine-based events from a doubling from the pre-randomization baseline creatinine rather than from a new creatinine at the beginning of The AASK Cohort. To deal with this complication, the 2-step process used for the Period 3 power analyses was modified as follows for the composite outcomes which including a creatinine doubling component. In the first step, separate Weibul models were developed for the composite outcomes with creatinine events defined by increases from baseline by factors of either 1.333, 1.667, 2.0, 2.5, 3.0, or 4.0. In the second step, the likely AASK Cohort participants who had not previously had a doubling of serum creatinine were stratified into six groups depending on their most recent serum creatinine value. Then, the appropriate Weibul model from the first step was used to project the number of patients in each strata whose latest creatinine would increase during the AASK Cohort period to a value equal to at least two time the patients pre-randomization baseline creatinine or who would reach one of the other events defining the composite outcomes. An additional 3% loss-tofollow-up was assumed for each outcome during the AASK Cohort period.
Based on the responses of AASK centers to a questionnaire on enrollment of patients for AASK Cohort, it is projected that 677 of the initial 1094 randomized patients will be alive and not on dialysis and will agree to participate in the AASK Cohort Study. Table 7 gives the projected numbers of events for each composite outcome, as the associated projected minimum detectable treatment effects (with 80% or 90% power based on level 0.05 2-sided tests) for increases in risk associated with 1) a dichotomous risk factor with 50% prevalence in the AASK patients, 2) a dichotomous risk factor with 20% prevalence in the AASK patients, and 3) a 1-standard deviation change in a quantitative risk factor which is linearly related to the log-transformed relative risk. The power calculations correspond to unadjusted risk ratios, do not account for potential use of covariates which might be correlated with the risk factors in multivariate models.
To illustrate the power calculations, consider analyses done in Period 3 comparing non-dippers to dippers based on ambulatory blood pressure monitoring. Assuming (conservatively) that 20% of AASK patients are non-dippers, the AASK Cohort will have 80% power to detect a 62% increase in the rate of composite endpoint of doubling of serum creatinine, ESRD, or death for non-dippers compared to dippers. The study will have 90% power to detect a 75% increase in October 1, 2002 50 the event rate. If the proportion of non-dippers turns out to be 50%, the AASK cohort will have 80% and 90% power to detect 47% and 56% increases in event rates for this composite outcome.
As a second example, consider an analysis relating serum total cholesterol to the composite endpoint of doubling of serum creatinine, ESRD, or death over the full Period 3. The standard deviation of total cholesterol at baseline in the randomized trial was 45 mg/dL. The Period 3 analysis would have 80% and 90% power to detect 21% and 25% increases, respectively, in the event rate corresponding to a difference of 45 mg/dL in total cholesterol.