Impact of timing of renal replacement therapy initiation on outcome of septic acute kidney injury

This retrospective study was based on the NSARF database, which was established in the 64-bed surgical ICU of a tertiary hospital and its three branch hospitals in different cities [17‐20]. The database prospectively collected data from patients requiring RRT during their ICU stays, and continuously recorded data from all patients for outcome analyses. In this study, we enrolled patients who underwent acute RRT because of septic AKI between July 2002 and October 2009. Those enrolled subjects were treated by one multi-modality team, composed of physicians, surgeons, technicians, and nursing personel. Septic AKI was defined as AKI development after sepsis without other etiology. Sepsis was classified according to the American College of Chest Physicians and the Society of Critical Care Medicine consensus [21]. Sepsis was defined by the presence of both infection and systemic inflammatory response syndrome (SIRS). SIRS was considered to: be present when patients had more than one of the following clinical findings: body temperature above 38°C or below 36°C, heart rate of more than 90 beats/min, hyperventilation evidenced by a respiratory rate of more than 20 breaths/min or a partial pressure of arterial carbon dioxide of less than 32 mmHg, and a white blood cell count of more than 12 × 10³ cells/μl or less than 4 × 10³ cells/μl. Infection was defined as a pathologic process caused by the invasion of normally sterile tissue or fluid or body cavity by pathogenic or potentially pathogenic microorganisms. Exclusion criteria included patients aged less than 18 years, patients with an ICU stay of less than two days [22], and patients who only underwent acute RRT for less than two days. Approval for this study was obtained from the Institutional Review Board of National Taiwan University Hospital, Taipei, Taiwan (No. 31MD03). Informed consent was waived because there was no breach of privacy and the study did not interfere with clinical decisions related to patient care.

All data were prospectively collected. Data variables included demographic data, comorbid diseases, septic AKI developed post-surgery (or not), and the indications for RRT. Biochemistry data such as complete blood cell count, blood urea nitrogen (BUN), serum creatinine (sCr), serum glutamate oxaloacetate transaminase (GOT), serum total bilirubin, serum albumin, and serum potassium (sK⁺) were recorded upon ICU admission and RRT initiation [18, 20]. Moreover, the clinical parameters and severity score were also recorded at these two time points. The clinical parameters included heart rate, systolic and diastolic blood pressures, central venous pressure (CVP) level, partial pressure of arterial blood gas oxygen and fraction of inspired oxygen. Severity scores included Glasgow Coma Scale (GCS) score, Acute Physiology and Chronic Health Evaluation II (APACHE II) score [23], Sequential Organ Failure Assessment (SOFA) score [24], and Simplified Acute Physiology Score III (SAPS III) [25]. The usage of mechanical ventilation was recorded and the inotropic equivalent dose was calculated [26]. Definitions were made as follows: hypertension was blood pressure above 140/90 mmHg or usage of anti-hypertension agents; diabetes was previous usage of insulin or oral hypoglycemic agents; congestive heart failure was low cardiac output with a CVP above 12 mmHg and dopamine equivalent above 5 μg/kg/min [26]; and chronic kidney disease (CKD) was sCr of 1.5 mg/dl or greater documented prior to this admission.

The indications for RRT were: (1) azotemia (BUN > 80 mg/dL and sCr > 2 mg/dl) with uremic symptoms; (2) oliguria (urine amount <100 ml every eight hours) or anuria refractory to diuretics; (3) fluid overload refractory to diuretics with a CVP level above 12 mmHg or pulmonary edema with a partial pressure of arterial oxygen/fraction of inspired oxygen ratio below 300 mmHg; (4) hyperkalemia (sK⁺ > 5.5 mmol/L) refractory to medical treatment; and (5) metabolic acidosis (pH <7.2 in arterial blood gas) [27].

According to previous studies [2, 28, 29], simplified RIFLE classification was used only with the glomerular filtration rate (GFR) criterion for classification because the eight-hourly urine volumes in our database did not match the 6- or 12-hourly urine output criterion in the RIFLE classification. The baseline sCr was the data acquired at hospital discharge from the prior admission among the patients with more than one admission [2], or the data estimated using the Modification of Diet in Renal Disease (MDRD) equation [30] in those with only one admission (assuming an average estimated GFR of 75 ml/min/1.73 m²). The peak sCr was defined as the highest sCr before RRT initiation in ICU. Those who initiated RRT when in sRIFLE-R (risk) or sRIFLE-0 [31], that is not yet reaching the sRIFLE-R level, were defined as the early dialysis (ED) group, while those in the sRIFLE-I (injury) or sRIFLE-F (failure) groups were classified as the late dialysis (LD) group.

The modality of RRT was chosen according to the hemodynamics of the patients. Continuous venovenous hemofiltration was performed if more than 15 points of inotropic equivalent (IE) [26] were required to maintain systemic blood pressure up to 120 mmHg. The effluent flow and blood flow were 35 ml/kg/hour and 200 ml/min, respectively. Extended RRT such as sustained low efficiency RRT (SLED) with or without hemofiltration (SLED-f) was performed if IE was between 5 and 15 points. For SLED, blood flow and dialysate flow were 200 ml/min and 300 ml/min, respectively. When hemofiltration was added, the hemofiltration rate was 35 ml/kg/hour. The duration of hemofiltration was about 6 to 12 hours, according to the amount of ultrafiltration. Intermittent hemodialysis, which was chosen if IE was less than five points, was performed for four hours every session with a dialysate flow of 500 ml/min, and blood flow of 200 ml/min. As hemodynamics change, the patients may receive different RRT modalities [19].

The endpoint of this study was in-hospital mortality. The survival period was calculated from RRT initiation to mortality (in non-survivors) or to hospital discharge (in survivors).

To balance the selection bias in an observational trial such as the current study, we used propensity score selection and the matching method [32, 33]. Further models were adapted in our study. In model 2, we conducted Cox proportional hazard models using a propensity score, and included all patients based on the probability of late RRT. In model 3, we identified factors associated with late RRT in the entire cohort, using stepwise logistic regression. Based on the factors identified, we matched patients with 1:1, 2:2, 3:3, or 4:4 blocks manually [32]. We subsequently compared outcomes between patients undergoing early dialysis or late dialysis. In addition, a sensitivity analysis was also carried out among the subset of patients undergoing dialysis due to azotemia, which represented the largest proportion of our study population.

Finally, Kaplan-Meier curves obtained with the log-rank test were plotted to demonstrate the differences in patient survival between the two groups (ED versus LD).

Cox proportional hazard model were conducted with the whole cohort to identify factors associated with in-hospital mortality. We found that patients underwent operations before RRT (hazard ratio (HR) = 0.631, P = 0.0011), pre-RRT CVP (HR = 1.030, P = 0.0140), pre-RRT diastolic blood pressure (HR = 0.987, P = 0.0089), pre-RRT GCS scores (HR = 0.929, P < 0.001), pre-RRT plasma lactate (mM) (HR = 1.086, P < 0.001), SOFA score on ICU admission (HR = 0.941, P = 0.0015), and SOFA scores on RRT commencement (HR = 1.068, P = 0.0058) were independently associated with in-hospital mortality.

The Cox proportional hazard model was conducted using the whole cohort, including propensity score as a covariate, and identified pre-RRT diastolic blood pressure (HR = 0.987, P = 0.013), pre-RRT GCS scores (HR = 0.923, P < 0.001), pre-RRT lactate (mM) (HR = 1.073, P < 0.001), pre-RRT SOFA score (HR = 1.104, P < 0.001), and SOFA score on ICU admission (HR = 0.934, P < 0.001) predicted in-hospital mortality when propensity scores were conditioned (HR = 0.085, P < 0.001).

By logistic regression, we identified differences between early RRT and late RRT groups, and found that male patients (odds ratio (OR) = 1.588, 95% confidence interval (CI) = 1.012 to 2.492, P = 0.0444), patients with CHF (OR = 0.520, 95% CI = 0.297 to 0.909, P = 0.0217), patients receiving operations (OR = 1.618, 95% CI = 1.041 to 2.516, P = 0.0326), and patients with higher admission creatinine (OR = 1.184, 95% CI = 1.051 to 1.333, P = 0.0055) could predicted late dialysis. This model had a good discriminating power (c-index = 0.637), and validation (Hosmer-Lemshow's statistics, P = 0.07, with chi squared = 14.6, df = 8) was fair.

We matched patients by 1:1 fashion according to each patient's propensity to late RRT. After careful matching, there were 178 patients in each cohort. Table 3 showed the demographic data of the matched cohort. No differences about hospital mortality were detected in both groups according to head-to-head comparison of demographic data. Log Rank test of Kaplan-Meier curves (Figure 1) was insignificant between these two groups (HR = 1.13, P = 0.33).

Further sensitivity analyzes were undertaken using patients undergoing RRT because of uremic symptoms. Hospital mortality was associated with post-operative status (HR = 0.651, P = 0.002), pre-RRT CVP level (HR = 1.031, P = 0.002), pre-RRT diastolic blood pressure (HR = 0.9687, P = 0.0029), pre-RRT GCS scores (HR = 0.969, P < 0.0001), pre-RRT lactate level (HR = 1.091, P < 0.0001), SOFA score on ICU admission (HR = 0.921, P = 0.0033), and SOFA score on starting RRT (HR = 1.071, P = 0.0021).

As an observational study, this investigation has several limitations to be addressed. First, only the estimated GFR criterion of the RIFLE classification was used in this study. Urine output may also be an important indication for initiation of RRT. Thus we used the term 'sRIFLE' to distinguish this classification from the original RIFLE. Second, our patients received different modalities as dictated by their individual hemodynamics. Therefore, we were not able to manage the impact of different doses and modalities. Third, we only included patients who actually underwent RRT. There will be a subset of patients with AKI who are not dialyzed "early" who never undergo RRT because they either die or recover kidney function before meeting the "late" criteria for RRT. Fourth, we did not correct sCr by the degree of fluid [50].