Change in Albuminuria and GFR as End Points for Clinical Trials in Early Stages of Chronic Kidney Disease

Selection of StudiesAs part of the prior individual patient meta-analyses, we had previously performed systematic reviews of the literature to identify RCT of kidney disease progression. The last search was performed in 2007. We thus restrict the literature search to studies since 2007. For consideration of inclusion of all studies (whether part of prior work or new studies identified in the new search), we plan to use different criteria for selection of studies of patients known to have CKD or who are at high risk for CKD conditions (diabetes, hypertension, CVD, heart failure). The rationale is that as a nephrology community, we need to develop the best answers for surrogacy for specific kidney diseases, despite the limitations of the data. In contrast, in high risk populations, there are multiple studies and we can focus the analyses on studies which will provide sufficient power. In particular this is relevant for the study of glomerular disease, for which we will apply different criteria for such studies than other causes of CKD (See table for specifics).We plan to include in the albuminuria analyses only those studies with interventions whose effects on kidney disease progression are thought to proceed primarily through pathways that lead to changes in albuminuria within several months. We will designate a committee to review each study to make this assessment a priori. Since GFR reduction is directly on the pathway to ESRD, all studies will be eligible for the GFR analyses.Criterion:1.   RCT2.   Articles published in English3.   Human subjects 4.   Adults5.   Follow up > 12 months after first follow up measurement of UP or GFR6.   Quantifiable ACR (ie not dipstick)7.   GFR > 15 (can apply to a subset of participants within a study, as long as the other criteria, including number of endpoints are met)8.   Frequency of measurements and number of events , different by population8a Studies of glomerular disease (IgA, membranous, lupus, possibly FSGS)a.   ACR or SCR at 6+ 3 monthsb.   >10 events(ESRD, 2X scr, death, 40% decline, 30% decline)8b Studies of kidney disease due to diabetes, hypertension, PKD, nonspecified or otherc.   ACR or SCR at 6+ 3 monthsd.   Person time - 500 person years (approximated for inclusion decisions as follow up time in years X number of participants)e.   >30 events (ESRD, 2X scr, death, 40% decline, 30% decline)8c Studies of high risk population (diabetes, HTN, CVD, heart failure not selected for having kidney disease)*f.   ACR or SCR at 6+ 3 monthsg.   1000 person-years of follow-up ( estimated by N * mean length of FU) h.   > 30 events (ESRD, 2X scr, death, 40% decline, 30% decline) *This could include a subset of participants from these studies that meet the criteria for high risk. This criteria to be determined by ACR > 30 or prediction equation for development of 30% or 40% decline in GFR (to be developed)Established endpoints In the 2012 NKF-FDA Scientific Workshop evaluating GFR decline as an alternative endpoint for kidney disease progression, we considered as the primary established endpoint time to the treated kidney failure [end-stage renal disease (ESRD), defined as initiation of treatment with dialysis or transplantation], untreated kidney failure, defined as (GFR < 15 ml/min/1.73 m2 in those with GFR > 25 ml/min per 1.73m2 at baseline, or doubling of serum creatinine. The primary analysis was the time to the first occurrence of ESRD or the composite of the three, censoring for death. In secondary analyses, we included time to death in the composite endpoint. For the proposed analyses here, we will include these two outcomes, but will also consider including 30 and 40% decline as possible components of the established endpoint. The advantage of including these endpoints is that it will allow studies of higher GFR to be included in the pooled analyses, the target population of interest. We recognize that these are not true clinical endpoints, i.e., these are themselves are alternative endpoints, and that 30% and possibly 40% decline are not appropriate for interventions with substantial acute effects. We will consider methods to incorporate these concerns. For example, we expect that we will need to compare the alternative endpoints to several of the composite clinical endpoints above. Our hypothesis is that the inclusion of 30 and 40% will provide more power to the analyses but would not change the overall result compared to using doubling of creatinine and ESRD alone. If the results change substantially, a possible conclusion is that the 30 or 40% decline are picking up acute effects on change in GFR, and likely should not be used. In our prior analyses, we found that use of confirmed declines in GFR resulted in stronger treatment effects for all magnitudes of eGFR decline. This is likely due to the fact that the “noise” in eGFR that could be related to short-term fluctuation in measured GFR (for example, due to acute kidney injury) or in non-GFR determinants of serum creatinine are not sustained across a subsequent visit. We therefore plan to use confirmed changes in 30 and 40% decline and in reaching GFR < 15 in the current analyses. Confirmed change is defined as whether the magnitude of GFR decline is confirmed by an eGFR determination at the next visit. If the endpoint occurs at the last visit, we will consider it as confirmed. For studies where serum creatinine measurements are very far apart, we might not be able to use confirmed endpoints; we will review this by study based on examination of the data. The list below lists possible composite established endpoints to be considered across our full set of analyses. In the detailed analytical sections below, we describe in detail which endpoints would be considered for individual analyses. In the specific analytical sections below, we describe the use of these endpoints.•   ESRD, doubling of serum creatinine, GFR < 15 confirmed, death •   ESRD, doubling of serum creatinine, GFR < 15 confirmed, 30%, 40% decline confirmed, death •   ESRD, doubling of serum creatinine, GFR < 15 confirmed (censoring for death)•   ESRD, doubling of serum creatinine, GFR < 15 confirmed, 30%, 40% decline confirmed (censoring for death)*ESRD indicates treated kidney failure defined as initiation of treatment with dialysis or transplantation*Increasingly recognized that death should not be included in primary endpoint for interventions directed at kidney disease progression and therefore endpoints that include death will be secondary endpointsSubgroups to be considered. In the specific analytical sections below, we describe the use of these subgroups•   Treatment: treatment, placebo or control treatment•   Intervention: RASB, BP, Diet, IS, Lipid lowering, ESA, others•   Diseases: Diabetes, Hypertension, IgA, membranous, mixed and unspecified, PKD•   eGFR: < 45, 45-59, 60-89, > 90 ml/min per 1.73m2 •   Urine protein <500; 500-1000, 1000-3500 and > 3500* •   Geographical region: NA and Europe, Asia, SE Asia** •   Race : Black, White, Asian, other**Definition of alternative endpoints based on short-term changes in albuminuria (Aims 1 and 3)Preliminary analyses: Description of change in ACR over the first year and associations with change in BP and change in GFR•   As mentioned above, there will be heterogeneity of methods used to measure urine protein amongst the studies. For assessment of change, we plan to describe the percent or multiplicative change within the individual using the method used in the individual study and not converts to a single method across all studies. We propose to log transform ACR for most analyses. Previously we have used log base 2 so that we can express per doubling or halving of ACR and will consider this approach. It is possible that the change in ACR on the absolute scale is more meaningful. If so, we will need to convert to a common measure and will use the same methods described above using the untransformed values.•   We will investigate longitudinal changes in ACR from baseline to follow-up assessments over the first year of follow-up in order to assess the stability of ACR changes during this early follow-up period. For example, we would like to answer the question that if a change is present at 6 + 3 months, is it then sustained at 12 months? We will also look specifically for acute declines in first few months where this is available and see if the decline is persistent or changes with time. We will compare and associate the changes in ACR to changes in BP and GFR.Aim 1: Examine associations of early changes in ACR with adverse outcomes and examine consistency of associations across different diseases, geographical regions, race, level of ACR and GFRNote, analyses also to be performed by observational analytical group who will be able to address other components of this overall aim Specific analytical goals for clinical trial analytical group a)   To estimate the individual-level association of early change in ACR with subsequent development of ESRD and death within clinical trialsb)   To determine the consistency of this association across different patient factors defined at baseline, as well as across treatment armsc)   To determine the consistency of this association with individual-level association in observational studiesd)   To compare the strength of individual level association with multiplicative and absolute changes in ACR •   Alternative endpoint (Baseline period): Endpoints based on change in ACR over baseline period of 6-12 months•   Established endpoint (Events after the baseline period): Composite of ESRD, GFR < 15 and death that occurs after the baseline period. Participants that have events during the baseline period will be excluded•   Adjustment: Analyses will be performed with and without adjustment for baseline patient characteristics which may influence the change in ACR or the outcome. To be discussed based on availability across all studies but probable factors to include treatment assignment, age, sex, race, baseline ACR, baseline SBP, baseline eGFR. Change in blood pressure is not available in all studies but we will investigate whether sufficient studies are available for sensitivity analysis to adjust for change in BP•   Analyses within each study, Cox regression analyses will be performed to relate the hazard rates for the outcomes to early change in ACR, adjusting for randomized group and the indicated covariates. Analyses will be performed within subgroups where appropriate at the study level (eg GFR, UP). Diagnostic analyses will be performed to assess a) proportional hazards, b) linearity of relationships on the log-hazard scale, c) dependence of hazards ratios on the randomized treatment group, d) interaction with baseline ACR level. If non-linearity is found, we will consider using a spline function to model the relationship between the outcomes and change in ACR •   Meta-analyses: After obtaining separate results for each study, we will use the results from the individual studies for a two stage random-effects meta-analysis. We will explore heterogeneity using the X2 test and I2 statistic, and explored sources of heterogeneity with random effects meta-regression (disease, intervention, other characteristics as appropriate). We will consider stratifying results by level of GFR and UP given difference in absolute risk of ESRD depending on baseline estimated GFR and UP.Aim 3: Examine evidence for using ACR changes as an outcome, with focus at higher baseline GFR Note, analyses also to be performed by simulations analytical group who will be able to address other components of this overall aim Specific analytical goals for clinical trial analytical group a)   To evaluate the agreement of treatment effects on endpoints based on early change in ACR with established endpoints across previous clinical trialsb)   To evaluate if estimates of the effects of randomized interventions on endpoints defined by early changes in ACR reliably predict the effects of randomized interventions on established endpoints•   Alternative endpoints: Percent (multiplicative) and absolute change in ACR over 6-12 months•   Established endpoint: Primary endpoint likely to be ESRD, doubling of serum creatinine, GFR < 15 confirmed, 30%, 40% decline confirmed. •   Adjustment: No covariate adjustment is planned since these are randomized trials. •   Analytic Methods•   Within study analyses: •   Cox regression analyses will be performed to estimate treatment effects on the established endpoints (as defined above). Results will be expresses as HR, SE, N events and the standardized Z score obtained by re-estimating the standard error corresponding to an N of 1000.•   Analyses of covariance will be performed to estimate treatment effects on endpoints based on change in ACR, expressed as geometric mean ratio:[exp {adjusted mean(log ACR follow-up) - adjusted mean(log ACR baseline)}] in treatment ÷ [exp {(adjusted mean(log ACR follow-up) - adjusted mean(log ACR baseline)}] in controlResults will be expressed as the geometric mean ratio (GMR), SE and the standardized Z score obtained by re-estimating the standard error corresponding to an N of 1000.•   Estimates of treatment effects on ACR and on the established endpoints will be described overall and within subgroups. •   Bootstrap resampling will be used to estimate the correlation between the estimated treatment effects on the two endpoints. •   For RCTs with adequate power, we will estimate the relative effect defined as the ratio between the estimated treatment effects the established endpoint (expressed has hazard ratios) vs. the treatment effect on endpoints based on change in ACR (defined as GMRs). Standard errors will be estimated by bootstrap resampling. •   Meta-analyses: •   Using the results from the within study analyses, separate random effect models will be used to obtain pooled estimates for 1) the treatment effects on the established endpoints, 2) treatment effects on ACR and 3) the relative effects (ratio). Pooled estimates will be computed for the overall group as well as by subgroups defined by baseline eGFR , baseline ACR and treatment arm. We will explore heterogeneity using the X2 test and I2 statistic, and explored sources of heterogeneity with random effects meta-regression (disease, intervention, other characteristics as appropriate). •   We will then apply random effects analyses to relate the treatment effects on the established to the treatment effects on ACR across trials. The result of these analyses will be expressed in terms of meta-regression equations that predict the treatment effect on the established endpoint from the estimated treatment effects on endpoints based on short term changes in ACR. Interaction terms will be considered with study factors (e.g., class of intervention, mean baseline eGFR, geometric mean ACR, disease type) to evaluate if the slope of this regression differs depending on study level factors. Evaluation of endpoints based on eGFR Slope (Related to Aims 2 and 4 above)Preliminary Analyses: Development of general analytic framework to apply to analyses of slope. Ed Vonesh and Tom Greene are in the process of developing a general analytic approach for the analysis of eGFR slope which can be broadly applied across CKD randomized trials. Longitudinal eGFR measurements from the clinical trials in the original CKD-EPI data base are being utilized for this purpose. The analytic framework will include estimates of the effect of the treatment on the following quantities: a) The acute change in eGFR, defined as the mean change in eGFR from baseline to 6 months follow-upb) The chronic slope, defined as the mean eGFR slope after 6 months, c) The total slope to a) 2 years, b) 3 years, and c) 4 years, defined by taking the appropriate linear combinations of (a) and (b) above. The period of the acute phase may be modified from 6 months for trials which do not include eGFR measurements prior to 9 months after baseline. By consensus of the broader analytic team, Vonesh and Greene will develop an analytic approach which can be applied to eGFR without applying a logarithmic transformation. However, models based on log transformed eGFR may also be considered. The preliminary model development work by Vonesh and Greene is addressing the following issues: a) How to address nonlinear patterns of mean decline to handle acute effects as well as long-term deviations from linearity during the chronic phase of the studiesb) How to incorporate higher variability of eGFR measurements at higher eGFR levelsc) How to address auto-correlation of eGFR residuals after accounting for linear components of trajectories [this can also be thought of as accounting for nonlinear trajectories at the individual patient level]d) How to account for informative censoring by death & ESRD using shared parameter models.e) How to address cases where the estimated covariance matrix of the random effects in the slope models is singular or nearly singularAim 2: Examine associations of slope of GFR with adverse outcomes and examine consistency of associations across different diseases, geographical regions, race, level of ACR and GFRNote, analyses also to be performed by observational analytical group who will be able to address other components of this overall aimSpecific analytical goals for the clinical trial group: a)   To compare the strength of the individual-level association of alternative parameters defined by eGFR slope with the established endpointsIn contrast to the ACR analyses, we are not able to look at the associations between change in ACR determined over a baseline period with subsequent development of endpoints for the GFR slope. GFR slope must be ascertained over a minimum of 2 years and in most trials, there will not be sufficient follow-up to examine association with slope in 2 years and events that follow. We will therefore attempt to evaluate the individual association between slope during a set time (eg 2 to 4 years) and events during this time period using a shared parameter model. This is not the ideal approach to examining this question. This therefore is not a key analysis of the clinical trial group and we anticipate that the strength of the analyses addressing the question will come from the observational study group•   Alternative endpoints: a) chronic eGFR slope, b) total eGFR slope to 2, 3 or 4 years. •   Established endpoints: a) eGFR = 15 ml/min/1.3m2, ESRD, or death and b) ESRD or Death, and c) eGFR = 15 ml/min/1.3m2 or ESRD.•   Subgroups: A limited set of subgroups will be defined from the detailed list of subgroups defined above.•   Covariate adjustment: Analyses will be performed with and without adjustment for baseline patient characteristics which may influence the eGFR slope-endpoints and/or established endpoints. To be discussed based on availability across all studies but probable factors to include treatment assignment, age, sex, race, baseline ACR, and baseline SBP.•   Analytic strategy. The first phase of these analyses will be performed by fitting shared parameter models for joint analyses of the longitudinal eGFR measurements and the more definitive time to event outcomes separately for each trial. The shared parameter models will be used to estimate hazard ratios relating the time-to-event outcomes to a) the chronic slope, and b) alternative estimates of the total slope to 2, 3 or 4 years. These hazard ratios will be obtained in models which condition on intercepts defining each patient's initial “true” eGFR level as well as the measured baseline covariates. In the second phase, meta-analytic and meta-regression methods will be used to characterize the variation in the hazard ratios across studies and to relate the hazard ratios to study level mean baseline levels of eGFR, UP and other characteristics. Aim 4: Examine evidence for using eGFR slope as an outcome, with a focus at higher baseline GFR (including consideration of acute effects and their reversal off treatment, “on treatment” slopes) Note, analyses also to be performed by simulations l analytical group who will be able to address other components of this overall aimSpecific analytical goals for the clinical trial group:    a) To evaluate the agreement of treatment effects on endpoints based on eGFR slope with established endpoints across previous clinical trialsb) To compare this agreement between alternative endpoints based on eGFR slope•   Alternative endpoints: a) chronic eGFR slope, b) total eGFR slope to 2, 3 or 4 years. •   Established endpoint: Primary endpoint likely to be ESRD, doubling of serum creatinine, GFR < 15 confirmed, 30%, 40% decline confirmed. •   Adjustment: No covariate adjustment is planned since these are randomized trials. •   Analytic Methods•   Within study•   Cox regression analyses will be performed to estimate treatment effects on the established endpoints. Results will be expresses as HR, SE, N events and the standardized Z score obtained by re-estimating the standard error corresponding to an N of 1000.•   Mixed effect analyses will be used to estimate the treatment effects on a) the mean chronic slope, and b) the mean total slopes to 2, 3, and 4 years (depending on the duration of each study). If the risk of bias from informative censoring is found to be substantial, these analyses will be performed using shared parameter models which account for time to ESRD and death. •   Estimates of treatment effects will be described overall and within subgroups. •   Bootstrap resampling will be used to estimate the correlation between the estimated treatment effects on the two endpoints. •   For RCTs with adequate power, we will estimate the relative effect defined as the ratio between the estimated treatment effects the established endpoint (expressed has hazard ratios) vs. the treatment effect on endpoints based on change in ACR (defined as GMRs). Standard errors will be estimated by bootstrap resampling. •   Meta-analyses: •   Using the results from the within study analyses, two level random effect models to obtain pooled estimates of the treatment effect on clinical endpoints, on treatment effects on GFR slope and on the relative effect (ratio). We will explore heterogeneity using the X2 test and I2 statistic, and explored sources of heterogeneity with random effects meta-regression (disease, intervention, and other characteristics as appropriate). •   We will then apply mixed effects analyses to relate the treatment effects on the established endpoints to the treatment effects on eGFR slope across trials. The result of these analyses will be expressed in terms of meta-regression equations that predict the treatment effect on the more definitive outcome from the estimated treatment effects on endpoints based on GFR slope. Interaction terms will be considered with study factors (e.g., class of intervention, mean baseline eGFR, geometric mean ACR, disease type) to evaluate if the slop of this regression differs depending on study level factors. Aim 5: Develop methods to combine change in ACR and GFR for combined endpoint (individual vs group)The analytic strategy for this aim is still under development. Analyses are to be done by the three analytical groups and conclusions drawn from all three. As a general approach, the clinical trial group will first summarize the results for GFR slope and ACR for the individual trial and describe concordance or discordance. Next we will consider a trial level analysis to evaluate if concurrent demonstration of beneficial treatment effects on endpoints defined by ACR AND on endpoints defined by eGFR slope are predictive of beneficial effects on the established endpoints. To do this, we will extend the mixed models for Aims 3 and 4 above to jointly relate the treatment effects on the established endpoints to the estimate treatment effects on endpoints based on change in ACR and to the estimated treatment effects on endpoints based on eGFR slope across studies. Other considerations that may be included in the analyses or discussed at the conference as additional review aims:1.   Concept of hierarchical endpoints - can our data help to develop formal recommendations?2.   Off target effects3.   Using eGFR slope and change in ACR within a trial to forecast future clinical events for patients who have not yet reached these events when the trial is completed. This strategy may provide increased statistical power, without incurring the risk of having to extrapolate from results from previous trials. 4.   Hybrid approach based on a joint analysis of a newly conducted RCT in conjunction with results of past RCTs using a multilevel model.