Prevention of Chronic Kidney Disease and Subsequent Effect on Mortality: A Systematic Review and Meta-Analysis

Objectives To perform a systematic review of randomized controlled trials to determine whether prevention or slowing of progression of chronic kidney disease would translate into improved mortality, and if so, the attributable risk due to CKD itself on mortality. Background CKD is associated with increased mortality. This association is largely based on evidence from the observational studies and evidence from randomized controlled trials is lacking. Methods We searched Ovid, Medline and Embase for RCTs in which an intervention was given to prevent or slow the progression of CKD and mortality was reported as primary, secondary or adverse outcomes were eligible and selected. For the first phase, pooled relative risks for renal endpoints were assessed. For the second phase, we assessed the effect on mortality in trials of interventions that definitively reduced CKD endpoints. Results Among 52 studies selected in first phase, only renin-angiotensin-aldosterone-system blockade vs. placebo (n = 18 trials, 32,557 participants) met the efficacy criteria for further analysis in the second phase by reducing renal endpoints 15 to 27% compared to placebo. There was no difference in all-cause mortality (RR 0.99, 95% CI 0.92 to 1.08) or CV death (RR 0.97, 95% CI 0.78 to 1.21) between the treatment and control groups in these trials. There was sufficient statistical power to detect a 9% relative risk reduction in all-cause mortality and a 14% relative risk reduction in cardiovascular mortality. Conclusions Firm evidence is lacking that prevention of CKD translates into reductions in mortality. Larger trials with longer follow-up time are needed to determine the benefit of CKD prevention on survival.


Introduction
Chronic kidney disease (CKD) represents an increasing burden on health care systems worldwide. The prevalence of CKD has increased over the past several years. It is currently estimated that 17% of people in the United States have CKD, and worldwide the prevalence is 23-36% in people aged $64 [1,2]. Over the past several years, it has been generally accepted in the medical literature and community that CKD is independently associated with premature mortality [3][4][5].
In order to confirm that a nontraditional factor, such as CKD, is a causal risk factor for mortality, the following conditions should be met: (i) biological plausibility as to why the factor may promote premature mortality; (ii) demonstration that the mortality risk increases with severity of CKD; (iii) demonstration of an association between the CKD and mortality in observational studies; and (iv) demonstration in placebo-controlled clinical trials that treatment of CKD decreases mortality. There is an abundance of evidence for first three conditions, however, the veracity of the last condition is largely unproven. [6][7][8][9][10] Randomized controlled trials eliminate the possibility that other conditions such as diabetes and hypertension, which cause CKD, confound the observed association between CKD and mortality. This is with the recognition though that an intervention's effects may be complex, impacting an outcome such as mortality through many potential pathways one of which may be CKD prevention. A sufficient amount of time would also need to elapse since an intermediate endpoint such as CKD was prevented, before one would expect to see a reduction in subsequent attributable deaths.
Furthermore, even evidence from recent observational studies also questions the causal relationship between decreased glomerular filtration rate (GFR) and mortality. Garg et al. demonstrated the risk for cardiovascular events and death in people who donate a kidney were no higher in the first decade after transplantation than in matched non-donors [11]. Wald et al. used a large Ontario database to perform a propensity score matched cohort analysis and found that survivors of acute kidney injury that required dialysis had a significantly elevated risk for development of end stage renal disease (adjusted hazard ratio 3.2) but all-cause mortality rates were not elevated (adjusted hazard ratio 0.95) [12].
With this background, we performed a systematic review and meta-analysis of randomized controlled trials (RCTs) to determine whether interventions that are efficacious for reducing the incidence or progression of CKD result in a commensurate reduction in mortality (cardiovascular or all-cause).

Methods
We used a standardized protocol to search the published literature and identify trials for our analysis.

Literature Sources and Search Terms
We performed an exhaustive search and evaluation of peerreviewed research published between 1948 and July 2011, including Ovid, MEDLINE and Scopus (EMBASE). We used many search terms and filters that include ''exp renal insufficiency, chronic'', ''hypertension, renal'', ''proteinuria'', ''diabetic nephropathies'', ''disease progression'', ''survival analysis'', ''treatment outcomes'', ''mortality.mp.'' and ''randomized controlled trials''. The search was limited to randomized controlled trials that studied human subjects without language restriction. All efforts were made to obtain the English translation of the trials published in non-English languages.

Study Selection and Data Collection Process
We included trials with both CKD and non-CKD participants. We included trials that reported CKD outcomes along with mortality as primary, secondary or adverse outcomes. All included trials had a mean/median follow up time of at least 1 year and a sample size of at least 100 total participants. Two authors independently reviewed the references and resolved disagreements by discussion. The authors analyzed the quality of reporting by using the Cochrane Collaboration's tool for assessing risk of bias: randomization method, allocation concealment, blinding of participants, staff and assessors, selective reporting (for renal endpoints), and description of withdrawals. [13] After careful assessment, we included a total of 52 trials for data abstraction.

Data Abstraction
We used a standardized data abstraction form for description of the trial, such as the title of the study, the authors, year of publication, country, trial design, number of participants and their baseline characteristics. We also abstracted the type of intervention and the incidence of primary and secondary outcomes in the control and treatment groups.

Outcome Measures
The renal endpoints included the following: doubling of serum creatinine (sCr), End Stage Renal Disease (ESRD) defined as initiation of dialysis or renal transplant, composite of doubling of serum creatinine or ESRD, albuminuria/proteinuria (incidence, progression, regression). The non-renal endpoints were cardiovascular (CV) death and all-cause mortality.

Statistical Analysis
Trials were grouped by type of intervention and then were analyzed in two phases. In the first phase of the analysis, pooled relative risks (RR) for each of the renal endpoints for each treatment vs. control were computed with Mantel-Haenszel statistics. Interstudy heterogeneity was calculated using the Chi 2 method and the I 2 statistic. A priori, we decided that if the pooled upper bound of the 95% confidence interval (CI) was ,1 and at least 3 individual studies demonstrated efficacy for the renal endpoints as evidenced by an upper bound of the 95% CI ,1, then the intervention was deemed ''efficacious for prevention of incident or progressive CKD'' and advanced to the second phase. We combined the trials evaluating either angiotensin converting enzyme inhibitors (ACEi) or angiotensin receptor blockers (ARBs) as one intervention called ''RAAS blockade''. In the second phase, the association between the renoprotective intervention and allcause and cardiovascular mortality was assessed. The results were considered significant with 2-sided a error ,0.05. All the results are reported with 95% CIs. Statistical calculations and graphs were made using the Review Manager (RevMan), Version 5.1. Copenhagen: The Nordic Cochrane Center, The Cochrane Collaboration, 2011.
In addition, for the second phase, we performed sensitivity analyses that examined the pooled relative risk of CV death and all-cause mortality by pooling only the trials within the type of intervention that demonstrated efficacy for the treatment for the CKD outcome. We also explored the diversity in study results and possible association of certain covariates to hard outcomes by performing sub-group analyses and by using meta-regression. For each meta-regression and subgroup analysis, only studies for which the factor of interest was available were included in the analysis. The statistical significance was determined by the proportion of variability explained by each study level characteristic and from the size of residual variance [14]. Best-fit lines in meta-regression graphs were estimated by generalized estimating equations using estimates from meta-regression models as the input values and were weighted by the variance of each estimate [15]. Metaregression analyses were conducted in SAS 9.3 and R 2.15.0.

Results
Our search identified 2844 citations for the first phase. Based on title and abstract review, we excluded 2631 citations. After a detailed review of 213 citations we further excluded 161 for various reasons. Among those, 20 trials had follow up time less than a year, 80 trials had small sample size, 19 trials did not report CKD outcomes or mortality, 30 citations were review articles or secondary analyses and 10 were excluded based on miscellaneous reasons ( Figure 1). We also were unable to obtain full text in English and excluded those 2 trials. After a careful and thorough review, we included 52 trials in phase 1 . The characteristics of these trials are listed in Table 1.
We pooled studies by type of intervention and assessed the effects on renal endpoints ( Table 2). Of all the interventions, only RAAS blockade (ACEi or ARB) vs. placebo consistently reduced renal endpoints as evidenced by a pooled upper bound of the 95% confidence interval (CI) ,1 with at least 3 individual studies having an upper bound of the 95% CI ,1. Thus, RAAS blockade was deemed ''efficacious for prevention of incident or progressive CKD'' and advanced to the second phase of our analysis. There were 18 trials of RAAS blockade vs. placebo [17,18,21,24,29,31,40,[47][48][49][50][51]53,54,57,69,70]. Two of these trials had two treatment arms which were compared with the placebo separately. [54,57] We report these data as two separate comparisons vs. placebo within each of these two trials, thus the total comparisons would be equal to 20 in the subsequent analyses.
A total of 32,557 patients were enrolled in these 18 RAAS trials. Study populations include adults with a mean age of participants 50.8 years. Approximately 62% were male. The median follow up time in these trials was 3 years with a range of 1.3 to 4.6 years. About 83% trials were in the setting of diabetes mellitus and 6 trials (33%) included patients with CKD (stage III or more and/or macroalbuminuria) [18,22,23,25,30,31,35,39,41,43,46,53,55,60,66,67,70]. The overall quality of studies included was good. All but 4 trials had low risk of selection bias due to random sequence generation although majority did not mention any specific methods for allocation concealment. Performance and detection biases were also low. About 25% trials had high risk of reporting bias due to selective reporting ( Figures S1A and B).

Renal Endpoints
RAAS blockade vs. placebo was effective for the following renal endpoints: incident albuminuria was reduced by 27% (n = 11 trials, pooled RR 0. 73

Sensitivity Analyses
We examined mortality only in the positive trials that demonstrated a statistically significant benefit for the renal endpoints. 8 trials reported on progressive renal disease as doubling of sCr and ESRD. 6 out of these 8 trials were protective for progressive renal disease (RR 0.70 [95% CI 0.59 to 0.82], RD       In order to further explore the causal relationship between the CKD and mortality within the group of RAAS vs. placebo trials, we performed several sub-group analyses and meta-regression by study-level variables and their association with the RR for mortality. When study-level variables were examined categorically, we found a non-significant trend towards reduction in mortality via RAAS blockade by length of follow up time, such that there was absolutely no benefit seen for studies with follow-up time ,2 years and 2 to 4 years, but there was a suggestion of protection with follow-up time .4 years (RR 0.91 [95% CI 0.71 to 1.17], RD 21 [95% CI 23 to 1], I 2 = 61%). However, only 4 trials had median follow-up greater than 4 years, and the statistical heterogeneity for this subgroup was high. There was no substantial differences in the pooled estimates when stratified by sample size (,1500 vs. .1500 participants), by mean GFR of study participants (.60 vs. #60 ml/min/1.73 m 2 ), by studies that enrolled participants exclusively with diabetes vs. other trials, by the proportion of participants in each trial with albuminuria (Table 4). Meta-regression was used to explore the association between three continuous study level variables and the risk of death: A) ARR of ESRD and doubling of sCr between RAAS blockade vs. placebo; B) ARR of albuminuria by RAAS blockade vs. placebo; C) and overall study sample size. None of the three study level characteristics explained a significant proportion of variability in the risk of death. ARR of albuminuria explained 13% of the variability, sample size explained 11%, and ARR of ESRD and doubling of sCr explained 3% of the variability ( Figure  S2).

Discussion
In this systematic review and meta-analysis, we sought to examine if the risk reduction in renal endpoints in randomized clinical trials also leads to a decrease in mortality. Despite the fact that RAAS blockade was beneficial for reduction in several renal endpoints (albuminuria, doubling of sCr and ESRD), we could not obtain firm evidence that renal benefit also translated into improved mortality, either all-cause or CV death. These results are contrary to those observed in observational studies. A large systematic review and meta-analysis of 41 studies showed that point estimate for the unadjusted relative risk of mortality in patients with CKD (vs. those without CKD) exceeded 1.0 in 40 studies and was significant in 93% cohorts. The overall pooled relative risk for mortality for CKD vs. no CKD was 1.77 (95% CI 1.33 to 2.34). [10] Meta-regression analyses revealed an increasing risk of mortality with decreasing renal function. These findings, though observational, would imply that mortality can be reduced when there is a reduction in CKD outcomes. However, despite examining the outcomes in RCTs containing over 32,000 participants in trials of RAAS blockade vs. placebo, we could not confirm the implications from the observed associations between CKD and mortality.
We had enough statistical power to detect differences in mortality and CV death of 9 to 14%. However, the point estimates for all-cause mortality and CV death were very close to 1, making it unlikely that larger sample size resulting in more statistical power would have demonstrated a meaningful reduction in mortality. Thus, our findings leave a few possibilities that warrant careful consideration: i) CKD is causal for mortality and CV mortality but the duration of follow-up of the trials was not sufficiently long enough to witness the reduction in these outcome, ii) the effect size on CKD by RAAS blockade was too small to translate into a reduction in mortality, iii) the etiologic fraction of CKD for mortality and CV mortality is too small and requires larger sample size, iv) the event rate of mortality and CV mortality was too low; v) CKD is associated with mortality although a direct causal pathway between them is not existent.  Was the length of follow-up in the trials not sufficiently long enough to witness a beneficial effect on mortality? The median follow up time in the trials included in our meta-analysis was 3 years for the RAAS trials. When we examined mortality by doing the sub-analysis of studies stratified by the median duration of follow-up, we observed a non-significant trend towards reduced mortality in the 4 studies with median followup .4 years. Recently, the data from two large CKD intervention trials with prolonged follow-up ( [71,72]. The AASK collaborative group reported a 24% reduction in doubling of sCr and ESRD in patients with urine protein to creatinine ratio (PCR) of .0.22 at 12 years but there was no improvement for mortality (Hazard Ratio 0.98, 95% CI 0.65 to 1.46). Data from DCCT/EDIC study showed that intensive glycemic control resulted in a 46% reduction in the endpoint of impaired eGFR, but there was no significant difference in mortality (HR 0.88, 95% CI 0.54 to 1.42). These two studies, despite very long follow up time, could not demonstrate that improvement in CKD outcomes manifested in a reduction in mortality, although by themselves, they were underpowered to detect a difference in mortality.
Was the effect size afforded by RAAS blockade for the renal outcomes too small to translate into reduction of mortality? In the RCTs herein, the relative risk reduction for renal outcomes was 15 to 27%, and the absolute risk reduction only ranged from 1 to 4%. Even if the mortality benefit completely paralleled that for CKD, then the largest absolute difference in mortality between the 2 interventions can only be 4%. However, the benefits of renal protection are not transmitted 100% towards the benefit of the hard outcome. This relates to the concept of etiologic fraction. The ''etiologic fraction'' or ''attributable risk exposed'' of CKD for non-renal outcomes lies somewhere between 0 to 99%. Using the data from meta-analysis by Tonelli et al. [10] the attributable proportion in the total population (Ap t ) of CKD for death is 34.1%, (Table 5) which means a 4% reduction in the risk of CKD will translate into a 1.3% reduction in all-cause mortality.
The 4 th possibility relates to the event rate of the hard outcomes in the trials. In order to appreciate a sizeable effect on an outcome, the event rate needs to be of a certain magnitude in order to influence it. The mortality rate in the RCTs ranged from 0.0 [47] to 21.65 [19]. Observational studies have demonstrated a mortality rate as high 6.8/100 person-years in patients with CKD [73]. Due to insufficient data to perform meta-regression in order to explore the association of event rate in studies with the risk of mortality, we are unable to conclude if a higher event rate would have explained the present lack of mortality benefit.
The last possibility is that CKD may not be directly causal for CV death and overall mortality. This notion although seems provocative and contrary to common belief in the nephrology community, to disregard it completely as a heresy may not be the right approach. Another way to look at this association would be to see the increased mortality in patients in which CKD is induced or incidence is increased iatrogenically. On the contrary, Lau, et al. have shown the renal cell carcinoma patients undergoing radical nephrectomy were at a 3-fold higher risk of developing CKD when compared to those who underwent nephron sparing surgery. Yet there was no difference in mortality in the two groups at 10 years [74]. Moreover, Wald, et al. used propensity-based methods to match patients with and without severe AKI. After a median follow up time of 3 years, they demonstrated that those who experienced AKI had a 3-fold increased risk of ESRD. Nevertheless, despite the fact that there were 3 times as many patients who developed ESRD compared to the no AKI group, there was absolutely no difference in allcause mortality [12]. The discordant association between CKD risk and mortality, in both directions, provides some equipoise around the assumption of a direct causal pathway between CKD and mortality. We as clinicians and researchers have witnessed several times over the past few years that approaching the targets based on surrogate markers has not translated into clinical improvement. Recent, well designed trials to evaluate the effect of reduction of proteinuria did not show any significant improvement in hard outcomes [75,76]. The same is true for anemia showing no improvement in clinical outcomes with correction of hemoglobin as well as HDL in patients with CVD [77][78][79][80][81]. The ''independent associations'' from observational studies may be confounded by other patient factors such as hypertension, dyslipidemia and diabetes that coexist with CKD. These established cardiovascular risk factors may be responsible for increased mortality witnessed in patients with CKD.

Conclusions
We have definitive evidence to demonstrate a reduction in CKD endpoints by RAAS blockade. However, we still lack conclusive evidence that the reduction in CKD endpoints will translate into a meaningful reductions in CV and all-cause mortality. To date, evidence from RCTs is not sufficient to fulfill the most important condition to prove the causality between CKD and mortality. However, the results of our analysis cannot be generalized to other interventions that may reduce or prevent CKD. For example, at the present time, there is insufficient data published for some interventions, such as allopurinol and alkali therapy, in CKD patients and the effect on mortality [82][83][84]. We suggest that unless more potent and efficacious agents for prevention or treatment of CKD are discovered (to increase the effect size), then RCTs of RAAS blockade (or other similarly effective agents) will need to be performed with longer duration of follow-up time than typically performed in CKD trials (at least 4 years or more of follow-up) in order to prove that direct causal relationships between CKD and mortality exist. Until then, we are unclear whether CKD shares the same company as other surrogate and non-causal endpoints in nephrology, such as anemia and vitamin D deficiency. [78,79,81,85]. Author Contributions