Acute kidney injury among critically ill patients with pandemic H1N1 influenza A in Canada: cohort study

This cohort study is a secondary analysis of prospectively collected data from a multi-centre inception cohort of all adults (age >18 years), admitted to any of 51 Canadian ICUs, with confirmed or probable influenza A (pH1N1)-related critical illness, from 16 April 2009 to 12 April 2010 [2]. Members of the Canadian Critical Care H1N1 Collaborative are listed in Additional file 1. The 51 ICUs that agreed to collect data for this study constituted a subset of the 286 Canadian ICUs that provide mechanical ventilation (R. Fowler, personal communication). Data on a subset of 50 patients in this cohort, all from Manitoba, Canada, were published previously [27].

Data from patients’ medical records were captured on standardized case-report forms, as described elsewhere [2]. Briefly, we collected data on dates and times of hospital and ICU admission and discharge, demographics, co-morbid diseases, acute physiology, laboratory parameters, illness severity (APACHE II [Acute Physiology and Chronic Health Evaluation II] [30] and SOFA [Sequential Organ Failure Assessment] [31] scores), treatment intensity (including mechanical ventilation, inotrope and vasopressor support, and RRT, with each determined on days 1, 3, 7, 14, and 28 of ICU admission), and vital status at hospital discharge. The comorbid condition of obesity was defined as body mass index ≥30 kg/m². Diagnostic tests, including scheduled serum creatinine levels, were not mandated by the study protocol.

Patients with pH1N1 were included if confirmed or probable according to case definitions of the World Health Organization and the Canadian National Microbiology Laboratory [32]. Critical illness was defined by: 1) admission to an ICU and either 2) receipt of mechanical ventilation (invasive or non-invasive) or 3) receipt of inotrope or vasopressor infusion [2]. Chronic kidney disease (CKD) was defined as pre-morbid baseline serum creatinine ≥1.5 times the upper limit of normal. End-stage kidney disease was defined as outpatient dialysis dependence preceding ICU admission.

The primary outcome was AKI, defined according to the RIFLE (Risk; Injury; Failure; Loss; End-stage kidney disease) classification scheme [33]. Patients were classified according to the worst RIFLE category achieved while in ICU, with RIFLE categories Injury and Failure used to define AKI. Categories were determined on days 1, 3, 7, 14, and 28 of ICU admission, using both creatinine and 24-hour urine output on those days, with baseline glomerular filtration rate estimated as 75 ml/min/1.73 m²[33]. Because this assumption may overestimate the incidence of AKI, we determined RIFLE categories using two additional approaches in sensitivity analyses: (1) assuming ICU day 1 creatinine as the baseline, and (2) assuming the lowest creatinine in the ICU as the baseline, except for patients who received RRT, who were all assumed to be in the Failure category. Both approaches would tend to underestimate the incidence of AKI because any creatinine measured in the ICU in a patient not on RRT is unlikely to be lower than baseline.

Secondary outcomes included utilization of RRT (any of continuous renal replacement therapy, slow low-efficiency hemodialysis, intermittent hemodialysis); hospital mortality; and lengths of ICU and hospital stay.

The study sample size was based on the number of patients with pH1N1-related critical illness admitted to a study ICU during the period of pH1N1 activity. Continuous variables were summarized as means (standard deviations) or medians (quartile 1 [Q1] - quartile 3 [Q3]) as appropriate and categorical variables as proportions. We used Student’s two-sample t-test or Wilcoxon rank sum test to compare continuous variables and chi-square or Fisher’s exact test for categorical variables. Univariable and multivariable logistic regression were used to evaluate associations between clinical characteristics and the outcomes of AKI, RRT, and hospital mortality. Age and sex were included in all models, AKI was forced into the mortality model, and other explanatory variables were included if p ≤0.2 by univariable analysis. We used univariable analyses to screen potential variables because we anticipated having too few events to include all variables in the initial model. All selected variables initially entered in the multivariable models were retained. We excluded variables if data were missing for more than 15% of the study sample. Interaction terms were not considered. There was no evidence of multi-collinearity in any of the multivariable models (all correlations <0.8). We examined model calibration using the Hosmer-Lemeshow test; no model demonstrated lack of goodness-of-fit. We considered p < 0.05 to be statistically significant. All analyses were performed using SAS version 9.2 (SAS Institute Inc, Cary, USA).

Of 562 study patients, 342 (60.9%) developed AKI (worst RIFLE category Injury [n = 129, 23.0%] or Failure [n = 213, 37.9%]; Table 4). Using the 2 alternate approaches to estimate baseline kidney function, the incidence of AKI (Table 4) was lower (49.9%) if ICU day 1 creatinine was assumed as baseline and similar (61.2%) if the lowest ICU creatinine was assumed as baseline, with those receiving RRT designated as RIFLE Failure.

In univariable analyses, AKI patients were more likely to be older (49.1 [15.1] vs. 46.4 [14.8] years, p = 0.03), obese (34.5% vs. 14.6%, p < 0.0001), diabetic (31.1% vs. 17.7%, p = 0.0003) and have CKD (12.9% vs. 1.8%, p < 0.0001) (Table 2). Similarly, AKI patients were also more severely ill on ICU day 1 (Table 3) compared to those without AKI, with higher APACHE II scores (23.0 [9.6] vs. 18.1 [7.6], p < 0.0001), SOFA scores (12.0 [3.8] vs. 9.7 [3.2], p < 0.0001), and worse PaO₂/FiO₂ ratios (143 [94] vs. 167 [98], p = 0.006). AKI patients had greater elevations in serum creatinine (155 [144] vs. 76 [31] μmol/L, p < 0.0001) and serum urea (14.4 [26.9] vs. 5.1 [2.8] mmol/L, p = 0.03) and lower 24-hour urine output (1070 [939] vs. 2058 [946] mL, p < 0.0001) on the first day of ICU admission.

Multivariable logistic regression (Table 5) showed that obesity (OR 2.94; 95% CI, 1.75-4.91), CKD (OR 4.50; 95% CI, 1.46-13.82), and increasing APACHE II score (OR per unit 1.06; 95% CI, 1.03-1.09) were independently associated with AKI. Higher P_aO₂/F_iO₂ ratio was protective (OR per 10-unit increase 0.98; 95% CI, 0.95-1.00).

Patients with AKI were more likely to receive invasive mechanical ventilation and inotropes or vasopressors in the ICU, and had longer ICU but similar hospital length of stay (Table 3). Eighty-five patients received RRT (24.9% of patients with AKI and 15.1% of entire cohort); 82 achieved maximum RIFLE category of Failure and 3 achieved maximum RIFLE category of Injury prior to RRT initiation. Multivariable logistic regression showed that independent predictors of receiving RRT (Table 6) were obesity (OR 2.25; 95% CI, 1.14-4.44), day 1 mechanical ventilation (OR 4.09; 95% CI, 1.21-13.84), and increasing APACHE II score (OR per unit, 1.07; 95% CI, 1.03-1.12) and day 1 creatinine (OR per 10 μmol/L, 1.06; 95% CI, 1.03-1.10).

Hospital survival status was available for 545 patients (97.0%). When stratified by AKI severity (Table 4), crude mortality differed (p < 0.0001) across RIFLE categories of None (21/178, 11.8%), Risk (7/38, 18.4%), Injury (20/125, 16.0%), and Failure (65/204, 31.9%). Mortality in the AKI and no AKI categories was similar among methods used to estimate baseline renal function (Table 4). Univariable analysis showed that AKI was associated with hospital mortality, but this association was no longer significant after multivariable adjustment (OR 1.35; 95% CI, 0.74-2.48; Table 7). Independent predictors of increased hospital mortality were age (OR 1.02 per year; 95% CI, 1.00-1.04) and APACHE II score (OR 1.05 per unit; 95% CI, 1.01-1.08), whereas higher P_aO₂/F_iO₂ ratio (OR 0.95 per 10 units; 95% CI, 0.92-0.99) and platelets (OR 0.92 per 10 × 10⁹/L; 95% CI, 0.89-0.96) on day 1 were associated with lower mortality.

We performed a prospective multi-centre cohort study in representative ICUs from across Canada of patients with confirmed or probable pH1N1 infection, the majority of whom were mechanically ventilated, to describe the incidence of AKI, rate of RRT utilization, and associated outcomes. AKI was common, complicating the course of 60.9% of patients. Most developed more severe forms of AKI, defined by RIFLE categories of Injury and Failure. Moreover, 24.9% of those with AKI, or 15.1% of the entire pH1N1 cohort, received RRT. In general, we found that AKI was more likely to occur in older patients with a higher burden of comorbid disease, specifically obesity, diabetes mellitus, and CKD. In addition, those developing AKI presented with higher acuity of illness, greater burden of organ dysfunction and showed worse kidney function early after ICU admission compared to those not developing AKI. Finally, pH1N1 patients whose course was complicated by AKI received far greater intensity (i.e. invasive mechanical ventilation, vasoactive support, and RRT) and duration of support in ICU, with comparable length of hospital stay and mortality.

Our study has several strengths. It provides prospective, multicentre, nationally representative data of critically ill patients with pH1N1 and describes the incremental burden associated with development of AKI. Our study also provides complementary data to other large observational studies from distinct geographic regions on the attributable burden of AKI in pH1N1-related critical illness [1, 5, 23, 25]. However, we also recognize our study’s limitations. First, despite a large cohort, data on kidney function were missing or unavailable for 16.2% of patients. These excluded patients were less severely ill and less intensively treated, and thus we may have overestimated the incidence of AKI. Second, we assumed normal baseline renal function among included patients and may thus have overestimated the incidence of AKI, but the incidence of AKI and mortality among these patients did not dramatically change in sensitivity analyses that would be expected to underestimate the incidence. Only 24-hr daily urine output, as opposed to hourly urine output, was available for determination of RIFLE category. Therefore, we modified our criteria as previously reported [34], possibly underestimating the incidence of AKI by missing patients with transient decrements in urine output. Third, we may not have accounted for additional confounding variables in the associations of clinical characteristics and AKI, RRT, and hospital mortality. In particular, we did not have information on the type or amount of fluid resuscitation, non–H1N1-related infections, or other nephrotoxins. We also did not collect data on the modality, mode, or dose of RRT, although none of these has been shown to consistently influence mortality [35]–[37].

In our study, the incidence of AKI complicating the course of pH1N1-related critical illness in Canada was higher than reported from Spain (17.7%) [23], South Korea (22.6%) [19], Australia and New Zealand (34.0%) [26], United States (42.0%) [16], and similar to that reported in Argentina (51.0%) [24, 25]. Many risk factors for AKI identified in our cohort, specifically older age [15, 19], diabetes [19, 23, 26], CKD [16, 19, 26], and illness severity [13, 19, 23, 25, 26] have been consistently shown to predict AKI in prior studies. Other studies have also identified male sex [23], pregnancy [26], immunosuppression [19], hypertension [19], and co-infection [23] as risk factors.

Not only was AKI common in our cohort, but most patients with AKI progressed to the RIFLE category of Failure (62.3%), similar to previous investigations [19, 23, 25]. Rates of RRT utilization among all pH1N1 patients (15.1%) and among those developing AKI (24.9%) were also relatively high and within the ranges (7.6-23.8% and 22.0-44.0%, respectively) described previously [19, 23, 25, 27]. These data contrast with large observational studies in non-pandemic populations that have generally shown that only 4-6% of critically ill patients receive RRT [38]. The higher rate of RRT utilization in pH1N1-related critical illness may be explained by near universal lung injury and higher acuity of illness in this population, which may increase clinicians’ desire to mitigate fluid accumulation and attempt extracorporeal removal of extravascular lung water. Other reasons for variation in RRT utilization among pH1N1 patients may include differences among regions and clinicians in the perceived indications, utilization, and timing of initiation of RRT.

Obesity has been found to be highly prevalent among critically ill patients with pH1N1-related respiratory failure [2, 5, 39]. Whether obesity identifies poorer general health characterized by cardiovascular disease, diabetes, or maladaptive immune function, or contributes independent risk due to alterations in pulmonary function and diminished reserve, remains uncertain. Obesity was associated with 3-fold higher odds of AKI, confirmed in other studies [16, 23, 25].

Elevated creatine kinase (CK) has been implicated as a contributing factor for AKI in pH1N1 [14, 17, 22, 28]. Pettila et al. reported that 15.8% of AKI patients had CK levels exceeding 5000 IU/L and found a trend of increasing CK levels with worsening AKI [26]. Demirjian et al. found higher median CK levels in AKI compared with non-AKI patients, with peak CK >2000 IU/L occurring 22% of those with AKI and in no patient without AKI [16]. In our study, CK levels were higher in AKI compared with non-AKI patients; however, this difference was not statistically significant. While rhabdomyolysis may have a supporting causal role in AKI among patients with pH1N1-related critical illness, the etiology is likely to be multi-factorial. Recent data have suggested that acute glomerular disease or active viral replication in renal epithelial tissue may also contribute [20, 24].

From a health policy and planning perspective, anticipating the need for additional resources for a pandemic such as pH1N1 [6], with a relatively high attack rate and propensity for organ failure, is vital. In the case of pH1N1, needs included additional ICU bed capacity, availability of ventilators, and access to ECLS [9, 10]. In addition, pH1N1 patients had greater relative utilization of RRT, highlighting the importance of expanded capacity for RRT services. Each of these technologies would be expected to increase costs related to staffing, disposables, and length of stay. This added resource burden may contribute to decreased ICU bed availability and hospital surge capacity during periods of heightened demand and complicate triage decision-making during future pandemics [40].