Association of Early versus Late Initiation of Dialysis with Mortality: Systematic Review and Meta-AnalysisPan Y.a · Xu X.D.b · Guo L.L.c · Cai L.L.a · Jin H.M.a
aDivision of Nephrology, No. 3 People’s Hospital, Shanghai Jiao Tong University School of Medicine, bDivision of Nephrology, Central Hospital of Minhang District, and cHemodialysis Center, Bao Shan Branch of No. 1 People’s Hospital, Shanghai Jiao Tong University, Shanghai, China Corresponding Author
Background/Aims: The association of the timing of dialysis initiation with mortality is controversial. We conducted a meta-analysis to determine the relationship between the risk of death and early initiation of dialysis, when the patient has a greater estimated glomerular filtration rate (eGFR). Methods: Prospective and retrospective cohort studies that independently measured the effect of early vs. late initiation of dialysis on risk of death were identified by review of several databases. Odds ratios (ORs) were estimated by comparison of the highest and lowest quartiles and combined by a random-effects model. Results: 15 studies (1,285,747 patients) met the inclusion criteria. Summary estimates indicated that early start of dialysis was associated with increased risk of mortality (OR = 1.33, 95% confidence interval (CI): 1.18–1.49, p < 0.00001). Subgroup analysis indicated that early starters were 6.61 years older (p < 0.00001) and more likely to have diabetes (OR = 2.23, 95% CI: 1.83–2.71, p < 0.00001) than late starters. Analysis of pooled results of early and late starters indicated that older age (OR = 1.18, 95% CI: 1.05–1.33, p = 0.006), diabetes (OR = 1.61, 95% CI: 1.38–1.87, p < 0.00001), and high comorbidity index score (OR = 2.38, 95% CI: 1.75–3.25, p < 0.00001) were strongly associated with increased risk of death. Conclusion: Our meta-analysis indicates that early initiation of dialysis (at higher eGFR) was associated with an increased risk of death. Older age, greater likelihood of diabetes, and the presence of severe comorbid disease(s) partly explain this effect.
Copyright © 2012 S. Karger AG, Basel
The timing for initiation of dialysis in patients with chronic kidney disease (CKD) is mainly based on clinical experience and international guidelines, but has been controversial. US Renal Data System (USRDS) records indicate that from 1996 to 2008, the percentage of patients initiating hemodialysis (HD) increased from 20 to 52% for those with estimated glomerular filtration rates (eGFRs) >10 ml/min/1.73 m2 and increased from 4 to 17% for those with eGFRs >15 ml/min/1.73 m2 . However, it remains uncertain whether the earlier initiation of HD was associated with better outcomes. Several observational cohorts and case-control studies have suggested that early initiation of dialysis was associated with improved survival and quality of life and fewer complications [2,3,4]. However, more recent observational studies and a randomized controlled trial (RCT) have indicated that early initiation of dialysis provided no benefit or may even be detrimental to clinical outcomes [5,6,7,8]. In 2006, the National Kidney Foundation (NKF) work group updated their guidelines for initiation of HD and stated that ‘at CKD stage 5, when the eGFR is <15 ml/min/1.73 m2, nephrologists should evaluate the benefits, risk and disadvantages of beginning renal replacement therapy’ . They also suggested that initiation of dialysis before CKD stage 5 may be appropriate for patients who have symptoms related to CKD.
There is a trend for administration of dialysis in patients with higher eGFRs in the USA, but the benefits and harms associated with early initiation of dialysis have not been definitively established. We performed a systematic review and meta-analysis of the effect of early initiation of dialysis on patient risk in order to resolve this issue.
All publications were identified by searches of MEDLINE (PubMed, 1966–2011), EMBASE (1974–2011), www.clinicaltrials.gov, and the Cochrane Controlled Clinical Trials Register Database with the following search terms: ‘initiation of dialysis’ OR ‘dialysis initiation’ AND ‘eGFR’ AND ‘all-cause death’ AND ‘mortality’. We also performed a manual search of references cited by the identified studies and relevant review articles. The search was restricted to human studies published in the English language. We identified 178 potentially relevant articles based on title and abstract. A study was eligible for inclusion if it was a prospective cohort study or an observational study that compared early and late initiation of dialysis and if the final outcome was all-cause death and mortality (table 1). Thus, 162 of these 178 studies were excluded. Figure 1 shows the detailed steps used for study selection. The bibliographies of all retrieved articles were also reviewed. Ultimately, 15 studies met our inclusion criteria for meta-analysis [5,6,7,8,10,11,12,13,14,15,16,17,18,19,20].
|Table 1. The baseline characteristics of patients with early vs. late dialysis in pooled studies|
|Fig. 1. Flow diagram of the identification process for eligible studies.|
Two reviewers (H.M. Jin and Y. Pan) performed the search and reviewed the results. All authors collected the data and two authors (H.M. Jin and Y. Pan) independently extracted the following study characteristics: surname of first author, year of publication, country or zone, study design, sample characteristics, follow-up period, age, sex, rate of primary diabetes, and eGFR (calculated by Modification of Diet in Renal Disease (MDRD) or the Cockcroft-Gault equation) at the initiation of dialysis. Disagreements were resolved by discussion between two authors (H.M. Jin and Y. Pan) in consultation with the other authors (X.D. Xu, L.L. Guo, and L.L. Cai).
Odds ratios (ORs) were used to assess the association of mortality with early vs. late initiation of dialysis and the association of mortality with an increase in the eGFR of 1 and 5 ml/min/1.73 m2 at initiation of dialysis. Four studies reported Charlson Comorbidity Index (CCI) scores (W.F. Clark, M. Evans, J.C. Korevaar, S. Wright), and predicted poor outcome for patients with advanced age, history of myocardial infarction, congestive heart failure, ischemic heart disease, peripheral vascular disease, cerebrovascular disease, chronic obstructive pulmonary disease, cancer, and diabetes. Thus, we considered the baseline levels of age, primary diabetes, and CCI for the entire study population.
We also performed subgroup analysis to identify confounding factors that may have affected outcomes, including age, diabetes, and CCI. We constructed a funnel plot for each outcome, with log OR versus standard error (SE), to assess potential publication bias. We also conducted a sensitivity analysis, in which each of the 15 studies was sequentially removed from the analysis, to evaluate the influence of each study on the final estimate.
We took the natural logarithm of ORs and calculated SEs and corresponding confidence intervals (CIs) . The generic inverse variance method in RevMan was used for analysis and the summary statistic and SE of each study were converted to the log scale before this analysis. RevMan converted the results back to ordinary ratios. An OR of the highest and lowest quartiles was used as measure of the association of mortality with early vs. late initiation of dialysis. ORs and their SEs were pooled with a random-effects model to assess statistical heterogeneity of the studies. All analyses were conducted in parallel by two investigators (Y. Pan and H.M. Jin) using meta-analysis packages (Review Manager (RevMan) Version 5.1, Copenhagen: The Nordic Cochrane Centre, the Cochrane Collaboration, 2011). Heterogeneity was estimated by the I2 statistic. Subgroup analysis based on age (mean ± SD, years), primary diabetes (%), and CCI was also used to identify potential confounding factors in the comparison of outcomes of early vs. late starters of dialysis. Statistical significance was set at a p value of 0.05 and 95% CIs were calculated.
Figure 1 shows the decision process used to identify studies for our meta-analysis. We ultimately included 15 studies (1,285,747 patients), of which 4 were prospective cohorts, 10 were retrospective studies, and 1 was a RCT [5,6,7,8,10,11,12,13,14,15,16,17,18,19,20]. Although the dialysis study of Rosansky et al.  included outcomes, we did not include it in our meta-analysis because the authors did not provide 95% CIs for their reported hazard ratios .
A total of 44% of the patients were female, but there was no significant difference in sex composition at baseline (table 1). The eGFR was calculated at several times during follow-up using the MDRD equation (eGFR = 186 × (SCr)–1.154 × (age)–0.203 × (0.742 if female) × (1.212 if black)) or the Cockcroft-Gault equation (eGFR = (140 – age) × mass × (0.85 if female) × (72 × SCr)–1), with correction for body surface area. Average eGFR was >10.97 ml/min/1.73 m2 in patients who started dialysis early and was <5.98 ml/min/1.73 m2 in patients who started dialysis late.
The pooled results indicated no benefit from early initiation of dialysis. Crude all-cause mortality increased with eGFR when there was no adjustment for age, sex, eGFR at initiation, CCI, primary renal disease, BMI, smoking, alcohol use, and level of education (fig. 2). Patients who initiated dialysis with eGFRs >10 ml/min/1.73 m2 had a 33.04% greater risk of death than those who initiated dialysis with eGFRs <3 ml/min/1.73 m2.
|Fig. 2. Crude incidence rate of mortality by category of eGFR.|
Figure 3 shows the ORs for all-cause death of the included studies, with adjustment for age, sex, ethnicity, serum albumin, and renal diagnosis. The results indicate that early initiation of dialysis was associated with greater risk for death (OR = 1.33; 95% CI: 1.18–1.49, p < 0.00001). Figure 4 shows that for each 1-ml/min/1.73 m2 increase in eGFR, the OR for death increased by 1.03 (95% CI: 1.02–1.04, p < 0.00001) and for each 5-ml/min/ 1.73 m2 increase in eGFR, the OR increased by 1.13 (95% CI: 1.07–1.19, p < 0.00001).
|Fig. 3. ORs of all-cause death among the included studies. The results had been adjusted for age, sex, ethnicity, serum albumin and renal diagnosis.|
|Fig. 4. ORs of mortality with each 1- and 5-ml/min/1.73 m2 increase in MDRD GFR among the included studies.|
Age, presence of diabetes, and CCI scores at the start of dialysis were potential confounders for the outcome and the primary variable of interest. Although Cox models and all multivariate models were used to adjust for numerous variables (age, height, weight, race, gender, diabetic status, comorbidity index, duration of predialysis nephrology care, type of dialysis, type of vascular access, cause of ESRD) at dialysis onset, the effect of confounders cannot be completely eliminated . Compared with patients who started dialysis early, late starters were 6.61 ± 1.30 years younger at the initiation of dialysis (95% CI: 5.31–7.91, p < 0.00001), 2.23-fold less likely to have diabetes (95% CI: 1.83–2.71, p < 0.00001), and 2.87-fold less likely to have a medium or high CCI (95% CI: 2.13–3.88, p < 0.00001) (fig. 5).
|Fig. 5. Baseline characteristics of age (mean ± SD, years), primary disease of diabetes (%) and CCI at start of dialysis in the observational cohort. a Age (mean ± SD, years). b Primary disease of diabetes (%). c CCI. Comorbidity index at dialysis start included: age, diabetes, history of myocardial infarction, congestive heart failure, ischemic heart disease, peripheral vascular disease (PVD), cerebrovascular disease, chronic obstructive pulmonary disease, cancer. The three-point score of low, medium or high comorbidity was used.|
Next we examined the pooled data of early and late starters of dialysis to identify significant predictors of outcome (fig. 6). The results indicate that older age at initiation of dialysis (OR = 1.18; 95% CI: 1.05–1.33, p = 0.006), diabetes at initiation of dialysis (OR = 1.61; 95% CI: 1.38–1.87, p < 0.00001), and a medium or high CCI at initiation of dialysis (OR = 1.59, 95% CI: 1.38–1.82, p < 0.00001 and OR = 2.38; 95% CI: 1.75–3.25, p < 0.00001, respectively) were all associated with increased risk of death.
|Fig. 6. Age (mean ± SD, years), primary disease of diabetes (%) and CCI in relation to mortality among the included studies. a Age (mean ± SD, years). b Primary disease of diabetes (%). c CCI.|
Most of the included papers made some adjustments for potential confounding factors, including age, primary diabetes, and CCI. Our sensitivity analysis indicated that exclusion of any individual study from the meta-analysis did not alter the overall conclusions of a positive association between early initiation of dialysis and increased risk of mortality (data not shown). We also considered variance in study design as a possible source of heterogeneity of the results. Sensitivity analysis indicated that this did not explain the heterogeneity of results, however we considered those heterogeneities were probably related to great different results among studies.
Finally, we assessed publication bias by visual examination of a funnel plot of precision versus hazard ratio and then formally calculated asymmetry with the Egger test (data not shown). The results indicated no significant publication bias (p = 0.10).
The main results of our meta-analysis indicate that (i) a higher eGFR at the initiation of dialysis was associated with increased risk of death, (ii) differences in baseline characteristics (age, diabetes, cause of CKD, CCI) of patients who started dialysis early partly accounts for their increased mortality relative to late starters, and (iii) advanced age, presence of diabetes, and elevated CCI were associated with increased mortality in a pooled analysis of early and late starters of dialysis.
It is important to account for the different baseline characteristics of early and late starters of dialysis in order to correctly assess the association of the timing of dialysis initiation with mortality. A RCT is clearly the best study design to assess this effect, but only one RCT was eligible for inclusion in our meta-analysis. Most of included studies were retrospective cohort studies, and are therefore subject to lead-time bias. In particular, lead-time bias in the timing of dialysis initiation would result in improved survival for patients who initiated dialysis earlier due to the patient’s greater residual renal function (RRF), but unrelated to the benefit provided by dialysis. This may explain the results of previous observational studies which reported that early initiation of dialysis (at a higher eGFR) was associated with better survival [2,3,4]. Despite the presence of lead-time bias in most of the studies that we examined, our meta-analysis indicated that early initiation of dialysis was associated with increased risk of death. However, potential confounding factors must be considered in the interpretation of this conclusion.
First, patient age was significantly different in the early starters and late starters. In particular, the mean age at baseline was 6.61 years greater for early starters and age was strongly associated with subsequent survival (HR = 1.18, 95% CI: 1.05–1.33, p = 0.006). This result is consistent with risk factors for mortality in HD patients reported in the Dialysis Outcomes and Practice Patterns Study (DOPPS) and other studies [23,24,25]. Previous studies indicated that the mortality risk of HD patients may be highest soon after initiation of HD. In the DOPPS study , 4,802 incident HD patients had a higher risk for death and cause-specific mortality rates during the first 120 days than during the subsequent 121–365 days. Factors associated with high mortality were older age, catheter vascular access, albumin <3.5 g/dl, and the presence of other comorbid diseases. Another study of the 6-month mortality rate of patients on maintenance HD reported that older age was independently associated with early mortality (HR = 1.36, 95% CI: 1.17–1.57) . A prognostic assessment of patients on HD for more than 1 year showed similar results, in that older patients had increased risk of death .
Second, the baseline difference in the number of patients with diabetes may be another confounding factor that affects the outcome of early and late starters of dialysis. In our study, early starters were 2.23-fold more likely to have diabetes, and diabetes per se increased the risk of death in a pooled analysis of early and late starters (RR = 1.61, 95% CI: 1.38–1.87, p < 0.00001). A previous study of 17,185 HD patients also reported that diabetes was significantly associated with increased risk of death . The increased death risk in HD patients with diabetes may be related to their decreased levels of serum albumin, peripheral artery disease, inflammation, and reduced visceral and somatic protein mass, all of which are associated with mortality in dialysis patients .
Third, we also noted there were more comorbid diseases (higher CCI scores) in our early starters. A pooled analysis of our data indicated that CCI score was strongly associated with increased mortality in early and late starters. This result is consistent with previous studies [27,28], and indicates that the CCI score may be a useful predictor of outcome in incident HD and PD patients.
It has been argued that patients without diabetes and lower comorbidity scores at early initiation of dialysis would have a survival advantage. Thus, Rosansky et al.  selected a cohort without diabetes and with low comorbidity scores. Compared with the reference group, the mortality HRs of the healthier cohort was 1.27 for patients with eGFRs of 5.0–9.9 ml/min/1.73 m2, 1.53 for patients with eGFRs of 10.0–14.9 ml/min/1.73 m2, and 2.18 for patients with eGFRs >15 ml/min/1.73 m2 in the first year of dialysis. This indicates that early initiation of dialysis might be harmful. Similarly, Kazmi et al.  also found a higher mortality rate in their low-risk cohort. It has been proposed that patients with high comorbidity scores and low levels of serum albumin may have a greater risk of death independent of HD, whereas patients with low comorbidity scores, normal nutritional status, and higher eGFR at initiation of dialysis might be more susceptible to potential harm from the HD procedure .
Survivor bias is another important concern in the comparison of early vs. late starters of dialysis and may potentially explain the improved survival of more healthy patients who start dialysis late . In particular, patients with lower eGFRs are ‘selected’ as survivors before the initiation of dialysis, and those who died before the initiation of dialysis were excluded. In an effort to reduce survivor bias, a Swedish study  enrolled patients prospectively when their eGFRs were <16 ml/min/1.73 m2. At the 5-year follow-up, the results were consistent with other observational studies that mortality was higher in patients who started dialysis with higher eGFRs. In another study that attempted to reduce survivor bias, Rosansky et al.  chose a ‘healthy’ cohort (younger nondiabetic patients with no reported comorbidities), used survival modeling to adjust for this confounding factor. Their results also demonstrated that early initiation of dialysis was strongly associated with increased mortality.
Random assignment of patients from a common eGFR starting point to early and late initiation of dialysis can reduce or eliminate survivor bias. The Initiating Dialysis Early and Late (IDEAL) study is the only RCT designed to determine whether early vs. late initiation of dialysis reduced the risk of death . During a median follow-up of 3.59 years, 37.6% of early starters and 36.6% of late starters died (HR for early dialysis: 1.04, 95% CI: 0.83–1.30, p = 0.75). Notably, 307 of the 828 patients died, indicating that these patients were in very poor health and probably not representative of stage 5 CKD patients. Also, the average eGFR at the time of dialysis initiation was 12 and 9.8 ml/min/1.73 m2 (Cockcroft-Gault equation) and 9.0 and 7.2 ml/min/1.73 m2 (MDRD equation) in the early and late starters, respectively. The mean differences (2.2 and 1.8 ml/min/1.72 m2) of the groups were statistically different, but the clinical significance of this difference remains uncertain. In addition, planned early initiation of dialysis was associated with greater cost, but not with improved quality of life .
The increased risk of death in patients who initiate dialysis early may be attributable to three important factors. First, the increased risk of death may be related to a rapid decrease in RRF after initiation of HD, because RRF is an important predictor of survival . Second, the increased risk of death may be related to recurrent episodes of myocardial ischemia and stunning due to excessive ultrafiltration relative to the interdialytic weight gain and frequent declines in blood pressure during HD . HD treatment per se is an independent risk factor for development of congestive heart failure and sudden cardiac death [33,34] and analysis of early outcomes of patients who initiated maintenance dialysis in the USA showed increased mortality and risk of hospitalization in the first 2 weeks of dialysis, and these remained elevated over the ensuing 90 days . Third, the increased risk of death may be related to chronic inflammation and oxidative stress, which can be induced by repeated exposure of blood to bioincompatible dialysis membranes and impure dialysis fluids. Previous research reported that conventional dialysis fluid can induce peripheral blood monocyte apoptosis and release of inflammatory factors, such as interleukin-6 and tumor necrosis factor-α [36,37], which are strongly associated with an increased mortality in HD patients.
There are several potential limitations in our meta-analysis. First, ten of the cohort studies had a retrospective observational design, which inevitably leads to lead-time bias for early initiation of dialysis. However, lead-time bias, if present, would have led to an underestimation of the survival benefit for late starters. In fact, lead-time bias may not be a valid advantage for early initiation of dialysis. Chan et al.  demonstrated a very high initial mortality when dialysis was initiated, and this initial mortality appeared to be even more marked in the cohort that started dialysis early. Second, RRF is an important predictor for survival in HD patients, but most of the included studies did not report these data after initiation of HD, so we cannot analyze the relationship of RRF and mortality. Third, all of the included observational studies may have been confounded by the strong effect of decreased muscle mass on serum creatinine-based measurements of eGFR. In other words, calculation of eGFR from the MDRD or Cockcroft-Gault equations may overestimate renal function in patients with advanced stage 5 CKD due to muscle wasting that is often related to underlying comorbidities. Previous studies have reported that in diabetic early starters and in new dialysis starters with body mass index levels <19.2 kg/m2 and serum albumin levels <25 g/dl, the MDRD equation is likely to overestimate GFR [38,39]. Other research indicated that low predialysis level of SCr was strongly associated with increased incidence of sudden cardiac arrest, and this may explain the harm associated with early initiation of dialysis . Fourth, there are no universal definitions of early and late start of dialysis, and these terms are simply defined by serum creatinine-based eGFR. The 2011 European dialysis guidelines  suggest that renal function should not be estimated from measurements of blood urea or creatinine alone. In particular, the Cockcroft-Gault equation should not be used when the GFR is <30 ml/min/1.73 m2 in order to determine the need for dialysis. They also concluded that the MDRD equation is useful in estimating the rate of disease progression, but should not be used to determine the need for dialysis or to estimate renal function in patients with stage 5 CKD (1A strong recommendation).
In conclusion, our meta-analysis indicates that older age, diabetes, and higher CCI score are strongly associated with increased risk of death in patients given early dialysis. Other studies have also indicated that the early initiation of dialysis can be costly and dangerous [22,30]. Thus, we suggest that the initiation of dialysis should be based on a patient’s condition, rather than the eGFR alone. More RCTs are needed to confirm the results of this meta-analysis and these RCTs should include more reliable measurements of renal function than simple estimations based on serum creatinine.
The authors have no conflicts of interest to disclose.
Hui Min Jin, MD
Division of Nephrology, No. 3 People’s Hospital
Shanghai Jiao Tong University School of Medicine
280 Mo He Road, Shanghai 201900 (China)
Tel. +86 21 5669 1101, E-Mail firstname.lastname@example.org
L.L. Guo, X.D. Xu, and Y. Pan contributed equally to this paper.
Received: November 24, 2011
Accepted: February 20, 2012
Published online: May 11, 2012
Number of Print Pages : 11
Number of Figures : 6, Number of Tables : 1, Number of References : 41
Nephron Clinical Practice
Vol. 120, No. 3, Year 2012 (Cover Date: August 2012)
Journal Editor: El Nahas M. (Sheffield)
ISSN: 1660-2110 (Print), eISSN: 1660-2110 (Online)
For additional information: http://www.karger.com/NEC