«Persons using assistive technology may not be able to fully access information in this file. For assistance, e-mail niddk-cr Include the ...»
A ‘probable’ cardiovascular outcome will be defined by any of the following:
• Non-fatal MI and documentation by a clinical report of myocardial infarction from the investigator but lacking confirmation of elevated enzymes or EKG changes OR
• Permanent neurological deficit of at least 24 hours duration attributed to stroke requiring hospitalization but lacking confirmation by radiographic imaging OR
• A centrally read ECG that documents a new MI in comparison to the baseline ECG OR
• A centrally read wall motion defect on echocardiogram in comparison to the baseline echocardiogram.
The Cardiovascular Outcome Committee will review potential cardiovascular hospitalizations using discharge summary and laboratory reports from these hospitalizations. Clinical centers are required to query patients about hospitalizations. All hospitalizations are reported. If the clinical center determines that cardiovascular event may have occurred during a hospitalization, a
Each hospitalization that is potentially cardiovascular in nature are reviewed by two members of the Cardiovascular Outcome Committee. If the two members of the Cardiovascular Outcome Committee are in agreement as to whether a cardiovascular outcome has been met, the case will be classified as such. If the two members of the Cardiovascular Outcome Committee are in disagreement as to whether a secondary or tertiary outcome has been met, the case will then come before the full Outcome Committee for review and adjudication.
Statistical Analyses: The analysis plans for the AASK Cohort are based on three periods of
Period 1: The randomized trial phase, including data from February 1995 through September 30, 2001.
Period 2: The complete data collection period, including both the randomized trial and continued follow-up in the AASK Cohort. This period extends from February 1995 though June 30, 2007.
Period 3: The AASK Cohort period, from February 1, 2001 through June 30, 2007.
Even though many analyses for Period 1 will be carried out in conjunction with the randomized trial, other Period 1 analyses, particularly those based on new measurements obtained in afterthought serum and urine specimens, will be performed by the AASK Cohort investigators.
Certain analyses for Periods 2 and 3 will be carried out prior to the end of AASK Cohort data collection in 2007. These will be conducted using the same methods as analyses for the full follow-up periods, but with earlier administrative censoring dates. We first review the main outcomes and summarize the general analytic approach for relating the renal outcomes to baseline factors for the three periods. Subsequently, we outline the specific analyses to be used to relate renal and cardiovascular outcomes to baseline and follow-up factors as appropriate for each of the eight primary research questions specified in Section 2.
Basic Renal Analytic Approach Period 1 Renal Outcomes: Two primary renal clinical outcomes for Period 1 are defined as the time from
randomization to the following composite events:
G1 - First confirmed GFR event (defined by a 25 ml/min/1.73m2 or 50% reduction in GFR from baseline), renal failure, or death.
G2 - First confirmed GFR event (defined by a 25 ml/min/1.73m2 or 50% reduction in GFR from baseline), or renal failure.
October 1, 2002 40 Outcome (G1) was the main secondary clinical outcome for the randomized trial. This outcome includes death primarily to avoid bias that can result if deaths are censored. Because inclusion of deaths may obscure purely renal effects, the second renal composite outcome (G2) is defined by confirmed GFR events and renal failure alone, censoring deaths. Additional secondary renal clinical outcomes include a composite of a confirmed GFR event, renal failure and cardiovascular deaths (censoring other deaths), and a “hard-endpoint” composite including renal failure and deaths of all causes.
The primary mechanistic renal analysis in Period 1 is based on the mean rate of change in GFR (GFR slope). A key secondary renal outcome is the change in the urine protein/creatinine ratio.
This section describes methods for analyses designed to relate renal outcomes to baseline factors while controlling for the randomized study interventions.
Time-to-Event Analyses: The association of baseline factors with the clinical renal outcomes will be evaluated with Cox regression models (Klein 1997) including both the baseline factors of interest and indicator variables for the six cells of the 2x3 factorial trial design to control for the randomized treatment interventions. For patients randomized to the calcium channel blocker arm, a time-dependent indicator variable will be used to distinguish between the periods before and after September 22, 2000 when the calcium channel blocker intervention was terminated.
Cox regressions of outcome (G1) will be administratively censored on September 30, 2001 or the date of final loss of contact with the patient; Cox regressions of (G2) will be censored for death, final loss of patient contact, or September 30, 2001.
Analyses of GFR Slope: The association of baseline predictor variables with GFR slope will be analyzed with mixed effects models (Laird 1982) containing fixed effects terms for the predictor variables of interest and their interaction with follow-up time along with additional fixed effects terms to control for differences in the mean rates of GFR decline among the six cells of the 2x3 factorial design. For the four cells within the ACE inhibitor and Beta Blocker arms, the fixed effects terms will express a 2-slope spline model with separate slopes prior to and after 3 months follow-up. For the Calcium Channel Blocker arm, the fixed effects terms will express a 4-slope spline model, with one slope in months 0-3 after randomization, a second slope from month 3 to termination of the Calcium Channel Blocker arm on September 22, 2000, a third slope from September 22, 2000 to January 1, 2001, and a fourth slope thereafter. The third slope will control for a reversal of the acute effect observed in months 0-3. The random effects terms in the mixed
A potential complication of the slope-based analyses is the risk of informative censoring due to loss-to-follow-up due to death, dialysis, or dropout. If censoring is informative, the standard mixed effects models may give biased estimates of the effects of predictor variables that are associated with early loss-to-follow-up. Therefore, during the initial months of the Cohort Study, the results of the standard mixed effects models will be compared to extensions of these models that account for informative censoring. In particular, the selection model of Schluchter (1992) and DeGrutollo and Tu (1994) provides a natural extension of the standard mixed effects model used in the randomized trial. Alternative pattern mixture models will also be considered (Little 1995). This approach will be used to assess the bias due to informative censoring in the regression coefficients associated with a wide range of predictor variables. If substantial bias is identified for important predictor variables, informative censoring models will be routinely used in place of the standard mixed effects models.
Period 2 Renal Outcomes: The clinical and mechanistic outcomes for Period 2 (with up to 12 years of follow-up) will be analogous to those for Period 1 (with up to 6 years of follow-up), but will be based on serum creatinine rather than GFR. The primary renal clinical outcomes for Period 2 are
defined as the time from randomization to the following composite events:
S1 - First doubling of serum creatinine from baseline, renal failure, or death.
S2 - First doubling of serum creatinine from baseline or renal failure.
Other key secondary renal clinical outcomes include the first doubling of serum creatinine, renal failure, or cardiovascular death, and the combined hard endpoints of renal failure or death.
The primary mechanistic renal analysis in Period 2 is based on the rate of change in predicted GFR using the following regression model derived from AASK enrollees at baseline (Lewis, 2001):
Predicted GFR = 329 × (Pcr)-1.096 × (age)-0.294 × (0.736 for women).
Time-to-Event Analyses: The association of baseline factors with the clinical renal outcomes will be analyzed in Period 2 using Cox regression models similar to those used for Period 1, but with modified definitions of the terms used to control for the randomized treatment groups.
Specifically, the Cox models will include indicator variables defining the patient’s randomized group, as well as separate time-dependent indicator variables specifying the time periods that the patient was actually assigned to the randomized intervention (randomization to September 22, 2000 for the Calcium Channel Blocker group, and randomization to September 30, 2001 for the ACE inhibitor and Beta Blocker groups). Cox regressions of the first composite (G1) will be administratively censored at the end of The AASK Cohort on June 30, 2007 or at the date of loss of contact with the patient; analyses of (G2) will be censored at death, June 30, 2007, or loss of patient contact.
Analyses of Predicted GFR Slope: Similarly to Period 1, the association of baseline factors with predicted GFR slope in Period 2 will be investigated with mixed effects models including terms for the baseline factors of interest and their interaction with follow-up time plus additional terms to control for the randomized treatments. Multi-slope linear spline models will be used to account for different mean rates of change over different periods in the respective randomized treatment groups. The time-points for the changes in mean slope under these models will be determined after examination of the data, but can be expected to include changes in slope at the same time points as the Period 1 mixed effects models. Due to the long follow-up period several alternatives for the random effects component of the mixed models will be considered, including models which allow for higher correlations between measurements spaced closer together than for measurements further apart.
Period 3 Renal Outcomes: The outcomes of Period 3 will be defined in terms of creatinine similarly to Period 2, but time 0 will be the time of the initial AASK Cohort protocol visit. The baseline serum creatinine for Period 3 analyses will be taken as the serum creatinine measured at this visit. Period 3 analyses will be restricted to patients who attended the initial AASK Cohort
Analyses of Clinical Renal Events: Cox regressions similar to those used for Phase 1 and 2 will be used to relate the clinical renal event outcomes to baseline factors. Even though patients will no longer be on their randomized interventions, indicator variables defining the original randomized groups will be used to control for any carryover effects of the randomized treatments. Period 3 analyses of the composite (G1) will be administratively censored on June 30, 2007 or at final loss of contact with the patient; analyses of (G2) will be censored at death, June 30, 2007, or loss of patient contact.
Analyses of Predicted GFR Slope: Analyses of predicted GFR slope in Phase 3 will be conducted with mixed effects models similar to those of Phases 1 and 2, with terms for the baseline factors and their interaction with follow-up time plus additional terms to account for the patients previous randomized group during the trial.
Analysis Plans for Specific Research Questions Listed in Section 2 of this Protocol Research Question 1. What is the long-term course of kidney function in this population?
The long-term course of kidney function will be characterized in Period 3 by investigating a) the rates of the composite outcomes (S1) and (S2) and other secondary renal event endpoints as a function of follow-up time, b) the distributions of long-term average rates of predicted GFR decline among patients belonging to relevant subgroups, and c) the pattern of change in predicted GFR over time within individual patients.
For (a), cumulative incidence curves will be constructed for each renal clinical outcome variable to characterize the proportions of AASK patients reaching renal events at different follow-up times. In addition to considering the full study group as a whole, Cox-regression techniques will be used to estimate rates of renal events as a function of follow-up time for patients with specific combinations of baseline characteristics. For (b), “smooth” density function estimates will be obtained for the distribution of long-term mean GFR slopes (after controlling for the randomized October 1, 2002 44 groups as described in the Basic Renal Analytic Approach) for the full cohort and for relevant patient subgroups. This will allow evaluation of many different aspects of the distributions of long-term rates of decline. Of particular interest are the proportions of patients who are nonprogressors as defined by a long-term average GFR slope greater than –1 ml/min per
1.73m2/yr (corresponding to normal aging), or who are rapid progressors, defined by an average GFR slope less than –4 ml/min per 1.73m2/yr, say. To address (c), parametric and nonparametric extensions of the basic linear mixed effects model will be used to determine whether individual patient’s long-term GFR declines follow constant linear paths or more complex nonlinear trajectories with different slopes during different periods of time.
Research Question 2. What are the environmental, genetic, physiologic, and socioeconomic factors which predict the progression of kidney disease?