Early versus late initiation of renal replacement therapy impacts mortality in patients with acute kidney injury post cardiac surgery: a meta-analysis

We searched the Chinese Biomedical Database, the Cochrane Library, EMBASE, Global Health, MEDLINE and PubMed for articles from November 1971 to August 2016. The predefined key search terms included “cardiac surgery” or “coronary artery surgery” or “coronary artery bypass grafting” or “cardiopulmonary bypass”, and “acute kidney injury” or “acute kidney failure” or “acute renal injury”, and “renal replacement therapy” or “hemodialysis”, and “early” or “late” or “time”. We reviewed the related research references at the same time.

The inclusion criteria for studies were: (1) original research that related to early RRT initiation in adult patients with AKI after cardiac surgery, (2) articles that provided exact data on mortality in AKI and (3) articles that reported a clear comparison of early versus late RRT initiation with a direct effect on mortality. The exclusion criteria were: (1) duplication, (2) studies such as systemic reviews, meta-analyses, comments, case reports, animal experimental studies etc. and (3) studies of patients with pre-existing renal disease or who received RRT before undergoing cardiac surgery.

Data were extracted independently by two investigators (HZ and QH). All potentially eligible citations that we searched were studied in detail to identify studies that satisfied the criteria. To identify duplicate records pertaining to a single study, we considered the PubMed database to take precedence; the details of the selection process are shown in Fig. 1. There would be a debate to reach consensus when there were disagreements. Data extraction included the first author’s name, year of publication, study design, RRT modality, definition of early and late RRT, number of patients, number of deaths due to AKI in post cardiac surgery, the mean and standard deviation or median of the ICU LOS, hospital LOS, duration of RRT and mechanical ventilation time: details are presented in Tables 1 and 2.

The assessment of study quality for the randomized controlled trials (RCTs) was performed using Review Manager (version 5.3) risk-of-bias tool, including four sections: selection, performance/detection, attrition and reporting bias (Fig. 2). The Newcastle-Ottawa Scale (NOS) (range 0 − 9 stars) was used to evaluate the cohort study quality (Table 1). For cohort studies, stars are awarded after evaluation of the three main categories of selection, comparability and outcomes: a study can be awarded no more than one star for each numbered option within the selection and exposure categories and no more than two stars can be awarded for comparability [21].

The data were abstracted and analyzed using Review Manager (version 5.3) and STATA statistical software (version 12.0) to make the outcomes more comprehensive. Estimation of effect was performed using a random-effects model and was summarized by forest plot, with data expressed as odds ratio (OR) with 95% confidence interval (CI) for dichotomous outcomes and mean difference (MD) with 95% CI for continuous outcomes. A random-effects model was used to deal with data in light of the heterogeneity in the results and the clinical characteristics of the study, while a fixed-effects model was used when there was poor heterogeneity. Heterogeneity was evaluated by the Q statistic and I ² tests, and low, moderate, and high heterogeneity was represented by thresholds of <25%, 25–75%, and > 75%, respectively [22]. P ≤ 0.05 was considered significant in all statistical tests. Subgroup and sensitivity analyses were used to explore the potential sources of heterogeneity. Egger and Begg test funnel plots were used to assess publication bias and the differences in the studies. Meta-regression was used to test the influence of baseline characteristics as potential effect modifiers, and the variables, including the year of publication, study design, RRT modality, definitions of early and late RRT, late RRT mortality rate ≥50% and classification of AKI.

Fifteen trials [17‐20, 23‐33] with a total of 1479 patients (771 patients receiving early RRT versus 708 patients receiving late RRT) ultimately met our criteria (Table 1); these included five RCTs [17, 18, 23, 25, 33], one prospective cohort study [29] and nine retrospective cohort studies [19, 20, 24, 26‐28, 30‐32]. The characteristics and methodological quality of all the included studies are shown in Table 1 and the outcomes in patients with AKI after cardiac surgery are shown in Table 2. Ten of the included articles were evaluated for study quality based on the assessment of the NOS [19, 20, 24, 26‐32].

The effect of lower mortality was obvious in patients with AKI who underwent cardiac surgery and who received early RRT compared to patients who received late RRT (OR 0.36; 95% CI 0.23 to 0.57; I ² 60%) (Fig. 3). Eight of the included articles showed that the ICU LOS (MD -2.50 days; 95% CI -3.53 to -1.47; I ² , 88%) (Table 3) and hospital LOS (MD -0.69 days; 95% CI -1.13 to -0.25; I ² 88%) (Table 3) were also significantly decreased in early RRT, respectively, five included articles also showed that RRT duration was similarly decreased (MD -1.18 days; 95% CI -2.26 to -0.11; I ² 69%) (Table 3). It was necessary to perform subgroup analysis to increase the reliability of the results and clarify the source of the high heterogeneity that we identified.

A subgroup analysis based on different time periods to initiate “early and late” RRT included thirteen articles [17‐20, 23‐26, 28‐30, 32, 33] (Table 3), while one of the excluded article was not referred to time cutoffs [31], the other was the only one categorized to the group of 72 hours resulted in there was no comparability [27]. The outcomes of when to start early RRT were that in AKI after cardiac surgery, the incidence of mortality was significantly decreased by early initiation of RRT within 12 hours (OR 0.23; 95% CI 0.08 to 0.63; I ² 73%) and within 24 hours (OR 0.52; 95% CI 0.28 to 0.95; I ² 58%) compared to initiation within 48 hours (OR 0.43; 95% CI 0.17 to 1.09; I ² 15%). In other words, early RRT initiation within 24 hours significantly reduced mortality compared to initiation after 24 hours in patients with AKI post cardiac surgery.

We also analyzed a subgroup based on RRT modality (Table 3): the impact of RRT modality on mortality was that the modality, whether continuous RRT (CRRT) (OR 0.36; 95% CI 0.19 to 0.67; I ² 55%) or intermittent hemodialysis (IHD) (OR 0.27; 95% CI 0.11 to 0.66; I ² 50%), had no effect on mortality in patients with AKI post cardiac surgery.

Furthermore, in another subgroup analysis based on study design (Table 3), in cohort studies there was a statistically significant decrease in mortality among patients who received early RRT (OR 0.33; 95% CI 0.23 to 0.46; I ² 15%), while the decrease in mortality in the RCTs (OR 0.41; 95% CI 0.14 to 1.24; I ² 74%) was not statistically significant.

Finally, there was insignificant statistics in the subgroup of AKI classification (Table 3), the outcomes of different classifications on AKI dignosis impacted on the mortality as following KDIGO 1 (OR 0.40; 95% CI 0.23 to 0.69; I ² 46%), KDIGO 2 (OR 0.04; 95% CI 0.01 to 0.15; I ² 0%), KDIGO 3 (OR 0.31; 95% CI 0.14 to 0.66; I ² 0%) and unclassified (OR 0.66; 95% CI 0.41 to 1.07; I ² 23%).

Sensitivity analysis indicated that the meta-analysis has low sensitivity and high stability in analysis of patients with AKI who had undergone cardiac surgery, which is demonstrated in Fig. 4. The Egger and Begg test funnel plots were used to explore publication bias (Fig. 5), the Egger linear regression test (P > 0.062) and the Begg rank correlation test (Pr > |z| = 0.113), which provided no evidence of substantial publication bias in this meta-analysis.

We performed meta-regression to investigate the sources of heterogeneity. The association between early RRT initiation and mortality was not influenced by year of publication (P = 0.949), either in study design (P = 0.114), RRT modality (P = 0.366) or a late RRT mortality rate ≥50% (P = 0.087). Stratification of studies according to the definition of AKI showed no impact on mortality when early and late RRT criteria were defined on the basis of time cutoffs and categorized by urine output, creatinine and blood urea (P = 0.541).