Background

Fluid therapy is commonly administered in patients with septic shock. Crystalloid solutions are recommended for the initial resuscitation of sepsis [1]. In the past decades, normal saline (0.9% NaCl), was the most frequently used fluid in adult ICU patients [2]. Normal saline, which contains 154 mmol/L of both sodium and chloride, can induce hyperchloremic acidosis with recently suggested adverse events [3]. Acute kidney injury (AKI) represents the most alarming one. Balanced crystalloid solutions, in which chloride anions are replaced by lactate or acetate, are increasingly used alternatives [2].

Three recent pragmatic RCTs have been conducted to compare the effectiveness of balanced crystalloids against normal saline for fluid therapy in heterogeneous populations of ICU patients [4,5,6]. None of them found a significant difference in AKI and mortality. However, in the “SMART” trial, normal saline significantly increased the incidence of major adverse kidney events (MAKE) within 30 days, a composite end point of persistent renal injury, renal replacement therapy and death [4]. Moreover, in the pre-specified subgroup of patients with sepsis, mortality was significantly higher in the saline group [4]. The significant reduction in mortality among patients with sepsis receiving balanced crystalloids has been confirmed in a recent meta-analysis [7]. However, no randomized study is available specifically designed to compare outcomes of septic patients resuscitated with balanced crystalloids or saline. In the absence of chloremia monitoring in the “SPLIT” trial [5] or a significant, but limited difference in chloremia in the SALT [6] and SMART trials [4], a cause-effect relationship between hyperchloremia and outcome remains uncertain. Moreover, patients included in these three RCTs were of low severity and only received a limited amount of fluids. The effects of chloride content and of hyperchloremia are, therefore, still controversial [3, 8]. Patients with septic shock requiring larger volumes of fluids and being at higher risk for AKI could be more exposed to the risk of chloride load. In this population, serum lactate represents an additional strong anions involved in acid–base balance. Compared to lactic acidosis, hyperchloremic acidosis has been experimentally shown to increase inflammation and hypotension [9]. Therefore, this post hoc analysis of the “HYPER2S” study comparing normal saline to 3% NaCl for septic shock resuscitation, we assessed the impact of hyperchloremia on AKI and mortality, stratified by lactatemia [10].

Methods

For all participating centres, the “HYPER2S” trial was approved by the ethics committee of the Angers University Hospital. Written informed consent was obtained from all patients, their next of kin, or another surrogate decision-maker, as appropriate. The “HYPER2S” trial was registered with Clinicaltrial.gov (NCT 01722422).

“HYPER2S” cohort

The present study is a post hoc analysis of the “HYPER2S” trial database, of which primary results were previously published [10]. Adult patients with septic shock admitted in the 22 participating ICUs were considered eligible for enrolment. Inclusion criteria comprised the need for epinephrine and/or norepinephrine infusion for less than 6 h at a minimum infusion rate of 0.1 µg/Kg/min, mechanical ventilation and plasma sodium concentration ≥ 130 mmol/L or ≤ 145 mmol/L. Patients were randomized to be treated with 0.9% (isotonic) saline or 3% (hypertonic) saline. Fluids were infused in a double-blind fashion during the first 72 h after inclusion. Study fluids were stopped if [Na+] > 155 mmol/L or > 12 mmol/L increase over 24 h together with a switch to open label isotonic saline. After the 72-h period of study fluid resuscitation, open label 0.9% saline was used for the remainder of the ICU stay in all patients.

Metabolic parameters and arterial blood gazes were concomitantly recorded at inclusion (H0), H12, H24 and H72. Serum creatinine and urine output (UO) were recorded daily up to D7 and the need for RRT up to day-28.

The trial was stopped prematurely for safety reasons after enrolment of 442 patients (434 analyzable), due to a toxic effect of hyperoxia. At study stop, day-28 mortality was 37% in the isotonic group vs 42% in the hypertonic saline group (p = 0.25)

Definitions The definitions of metabolic and acid-base disorders are provided in Additional file 1: Table S1. Hyperchloremia was defined by [Cl] ≥ 110 mmol/L. This cut off value has been shown to be associated with more AKI and mortality rate in previous studies [11,12,13]. Definitions of acidemia (pH < 7.35) and of metabolic acidosis (bicarbonate < 22 mmol/L) were based on previous studies as well [14]. Hyperlactatemia was defined by a serum concentration > 2 mmol/L according to the sepsis-3 definition [15].

Because cumulative urine output was recorded daily only, AKI definition was based on serum creatinine criteria [16]. Serum creatinine at inclusion (i.e. just before exposure to chloride rich crystalloids) was used for baseline value. The proportion of patients with Major Adverse Kidney Events (MAKE), a composite of death, new receipt of RRT or RRT dependence at day 28, was computed and censored at hospital discharge or 28 days after H0, whichever came first [4, 17].

Data analysis

In this post hoc analysis, only patients with available data on chloremia were analysed. Patients were classified according to the presence of at least one recorded episode of hyperchloremia or not. Patients’ characteristics were compared at baseline (H0), metabolic evolution was compared up to H72 and outcomes up to day-28.

In the context of septic shock, hyperlactatemia is a determinant of acid-base balance and represents a strong marker of severity with a significant higher risk of death. We have previously confirmed in the “HYPER2S” cohort that hyperlactatemia is a marker of higher severity and has a strong impact on outcome [18]. Analysis of the role of hyperchloremia on acid-base balance and outcomes was, therefore, adjusted to serum lactate concentration. Quantitative variables were expressed as median and 25th–75th interquartile range (IQR) and were compared using the Mann–Whitney test. Qualitative variables were compared using χ2 or Fisher’s exact test. General mixed models were developed to compare the evolution of metabolic parameters and renal function over time. The model was linear for pH and non-linear for chloride, lactate, bicarbonate, [Na]–[Cl] difference, serum creatinine and urine output.

To identify factors associated with AKI and mortality, uni- and multivariate analyses were computed using Cox regression using time-dependent covariates. Variables known to be significantly associated with AKI or mortality were a priori selected to be entered in the models. Serum creatinine, chloremia, vasopressor dose and volume of fluid resuscitation were used as time-dependent covariate. Patients with at least one missing baseline values were excluded from this analysis. Other missing values for time-dependent covariates were imputed by the last and next method [19].

Several chloride parameters were studied, i.e. chloremia within 72 h, hyperchloremia, maximal chloremia, delta chloremia, [Na–Cl] difference. These parameters were separately entered in each model.

To take into account the competing risk of early death, a sensitivity analysis was performed among patients alive at day 3 and different scenario of hyperchloremia imputation were tested in patients dead at day-3 without hyperchloremia. In these patients, hyperchloremia was imputed at H0 alone, H24 alone and at both.

All analyses were performed using R software (R Core Team. R Foundation for Statistical Computing, Vienna Austria, version 3.4.4).

Results

413 patients of the 434 included could be analysed totalizing 1736 metabolic points recorded. Hyperchloremia was observed in 257 patients (62%) and hyperlactatemia in 294 (71%), p = 0.006 (Additional file 1: Figure S1). The cumulative number of patients with hyperchloremia was 111 at H0, 215 at H12, 242 at H24 and 257 at H72. Episodes of hyperlactatemia were more frequent among patients with hyperchloremia, 75% vs. 65%, p = 0.04. Hyperchloremia and hyperlactatemia were recorded in the same patient in 192/413 (46%).

Comparison of patients with and without hyperchloremia

Patients with and without hyperchloremia had similar baseline characteristics (Table 1). Patients with hyperchloremia were more frequently included in the hypertonic saline group and received significantly higher volumes of fluids from H0 to H72: 2.2 (1.1–3.9) vs. 1.4 (0.6–2.3) litres for fluid resuscitation, p < 0.001 and 8.2 (5.4–11.4) vs. 5.4 (3.2–8.6) litres for all iv fluids, p < 0.001.

Table 1 Characteristics and outcomes of patients with and without hyperchloremia
Full size table

Figure 1 shows the evolution of metabolic parameters in patients with and without hyperchloremia. Serum chloride was significantly higher all over the time in the group of patients with hyperchloremia. Serum bicarbonate, pH and [Na]–[Cl] difference were significantly lower at H0 among patients with hyperchloremia. The evolution of all parameters (except for chloremia) were similar. Of note, the slopes of acidemia corrections were comparable between patients with and without hyperchloremia (Fig. 1).

Fig. 1
figure 1

Evolution of metabolic parameters between patient with and without hyperchloremia from H0 to H72. Means and 95% IC are provided. *p value for comparison at H0. #p value for comparison of the evolution over time (results of Generalized Linear Mixed Model for pH and Generalized Non-Linear Mixed Model for others)

Full size image

No matter the presence of hyperlactatemia, metabolic acidosis was significantly more frequent in patients with hyperchloremia (Additional file 1: Table S2). In patients without hyperlactatemia, acidosis occurred in 60% of patient with hyperchloremia vs. 36% without hyperchloremia (p < 0.01). In patient with hyperlactatemia, acidosis occurred in 86% of patients with hyperchloremia compared to 64% (p < 0.001) in patients without hyperchloremia.

Impact of hyperchloremia on outcomes

Patients with and without hyperchloremia had similar outcomes regarding AKI (45% vs 51%, p = 0.29), need for RRT (36% vs 36%; p = 0.98), MAKE-28 (51% vs 53%, p = 0.66) (Table 1). The evolution of serum creatinine and urine output were similar between patients with and without hyperchloremia (Fig. 2). The evolution of serum chloride concentration was similar between patients with and without AKI (Additional file 1: Fig S2).

Fig. 2
figure 2

Evolution of renal function between patient with and without hyperchloremia from H0 to day-7 among patient free of RRT at H0. Means and 95% IC are provided. *p value for comparison at H0. #p value for comparison of the evolution over time (results of Generalized Non-Linear Mixed Model)

Full size image

Mortality up to day-28 was 37% vs 38%, p = 0.76 in patients with and without hyperchloremia, respectively (Table 1). The evolution of serum chloride concentration was similar between survivors and non survivors (Additional file 1: Fig S2).

In multivariate analysis, the risk of AKI (Table 2) and death (Table 3) was not independently associated with chloride parameters.

Table 2 Uni and multivariate analysis for chloride parameters and AKI
Full size table
Table 3 Uni and multivariate analysis for chloride parameters and mortality
Full size table

Sensitivity analysis

Among the 73 patients who died early, only 29 were not exposed to hyperchloremia. Sensitivity analysis performed in patients still alive at days 3 or with the imputation of hyperchloremia is reported in Additional file 1: Table S3. Sensitivity analysis of AKI and death did not modify the main results. Nonetheless, we observed a trend toward an increased risk of AKI when hyperchloremia is imputed to the 29 patients at H24, but not at H0.

Discussion

We report the first analysis of the potential role of chloremia on the outcome of patients with septic shock. Our results show that hyperchloremia was frequent in patients with septic shock resuscitated with chloride rich crystalloids, however, significantly less frequent than hyperlactatemia. Independently of the presence of hyperlactatemia, hyperchloremia was associated with more frequent metabolic acidosis and more pronounced acidemia. Nevertheless, resuscitation using chloride rich fluids did not worsen shock-related metabolic acidosis nor impeded its correction, even in the presence of hyperchloremia. Hyperchloremia was not associated with an increased risk for AKI or death.

Strong associations between high chloride levels and worse outcome, AKI or death, have been reported; however, in various cohort studies discrepant results exist [12, 13, 20, 21]. Our findings are consistent with several negative retrospective cohorts assessing the role of hyperchloremia in ICU patients. In a single centre study enrolling 1045 patients with sepsis, 29% of them developing hyperchloremia, Yessayan et al. [22] did not show an association between hyperchloremia nor delta chloremia (as defined by increase > 5 mmol/L) and AKI. In a multicentre study including patients with severe acidemia, 54% of them with septic shock, the pH value and lactatemia were significantly associated with increased mortality, while chloremia was not [23]. Similarly, in a general population of ICU patients, Gunnerson et al. [24] found that lactate was a strong and independent predictor of mortality contrary to chloremia. A moderate increase in serum chloride of 5 mmol/L, even in patients with chloremia in normal range, has been shown to impair renal function and worsen survival in septic patients [12, 13]. We, therefore, tested different chloride parameters (i.e., hyperchloremia, chloremia as a time-dependent variable, maximal concentration, increase in concentration > 5 mmol/L); none of them was significantly associated with neither AKI or mortality. The negative results of our study deserve discussion of the underlying hypothesis. We assessed the role of chloremia in the most severe population reported so far. On the one hand, severe patients who receive large volumes of fluids are more exposed to the risk of hyperchloremia and its related adverse effects, on the other hand, stronger factors determining renal function and survival could hide the specific effect of chloremia. In patients with shock, the benefit of fluids on shock reversal could outweigh the risks of hyperchloremia. In our study, despite the occurrence of hyperchloremia, fluid resuscitation did not worsen acidemia, which is in contrast to data reported in surgical patients with normal pH at baseline [25]. Of note, in cohort studies reporting a negative impact of chloride load, the volume of fluid administered (i.e. around 5 litres within 24 or 48 h) was usually higher than in our study, despite a lower severity of patients studied [26,27,28].

We assessed the impact of hyperchloremia in the context of hypertonic fluid resuscitation. Occurrence of hyperchloremia was, therefore, driven by hypernatremia which per se might have a different impact on the outcome. However, the reduced difference between sodium and chloride concentrations indicating excessive chloride load was not associated with worse outcome. In trauma patients, hypertonic fluid resuscitation induced a huge, but transitory, increase in chloremia [29]. Compared to isotonic crystalloids, hypertonic solutions have never been identified as a significant risk for inducing AKI under these conditions [30].

The mechanism by which hyperchloremia would induce AKI remains highly speculative, particularly in the context of and the presence of markedly altered vasomotion. Comparing NaCl solutions of different chloride concentrations, animal studies have suggested that infusing a chloride-rich solution, directly into the renal artery, critically determined changes in renal blood flow due to intra-renal vasoconstriction [31, 32]. This finding was not confirmed in a more relevant animal model of sepsis resuscitated with large amounts of crystalloids [33]. Compared to a balanced solution, normal saline induced hyperchloremia and acidosis; however, no difference was found for renal hemodynamics and function [33]. Moreover, decreased renal blood flow velocity and cortical tissue perfusion have been reported in healthy volunteers after infusion of 2L of saline [34]. Finally, ICU patients in whom renal perfusion was assessed by renal Doppler, no relation was found between chloremia and resistive index [35].

Our study has several strengths: (i) the study design combines a number of features that reduce the risk of bias, (ii) metabolic parameters were prospectively recorded at precise time points with a concomitant record of serum lactate and chloride allowing precise adjustment, (iii) the choice of the fluids, the metabolic parameters to be recorded and the renal monitoring were imposed by the study design. Hence, in contrast to previous retrospective cohort studies based on administrative funding databases, biases related to fluid selection and metabolic dosage determined by patient characteristic and physician choices were ruled out. Such biases are present in the “SMART” trial in which the type of balanced solution, metabolic and renal function monitoring were left at the discretion of investigators [4].

Our study also has limitations that deserve attention. It is a post hoc analysis and the number of patients, hence, is quite limited. Nevertheless, it is noteworthy that no trend of worse evolution was observed among patient with hyperchloremia. Of note, we observed a trend toward higher risk of AKI in the sensitivity analysis when hyperchloremia was imputed to the early dying patient at H24 but not at H0 (Additional file 1: Table S3). Crystalloids used were only chloride rich with a high proportion of patient developing hyperchloremia. However, a sufficient number of patients remained free of hyperchloremia. The rules applied in the study for stopping hypertonic saline when hypernatremia occurred might have minimized the risk and the intensity of hyperchloremia. However, hyperchloremia over beyond 120 mmol/l was observed in 25% of the patients (Additional file 1: Table S2).

Our study did not assess the potential beneficial effect of using balanced crystalloid. The resolution of metabolic acidosis could have been more rapid using balanced solutions. The impact of the rapidity of acidemia correction on the patients’ evolution, however, remains unknown. Our results merely suggest that the potential beneficial effects of balanced solution in septic shock patients are might be unrelated to the serum chloride concentration. The definition of AKI was based on serum creatinine level at inclusion in the “HYPER2S” study. Using baseline serum creatinine before septic shock occurrence may have allowed detection and exclusion of patients with AKI occurrence before exposure and may have resulted in different finding. We used baseline creatinine at H0, i.e. just before exposure, because otherwise patients developing AKI before the start of sepsis start and exposure to chloride-rich solutions (H0) would have been wrongly classified as AKI-related to hyperchloremia. Considering the design of our study and the large confidence intervals of associations, our results do not rule out a negative association between chloride parameters and outcomes in patients with septic shock. Our results need to be confirmed by a larger study specifically designed to address this question in patients with septic shock of similar severity.

In conclusion, patients with septic shock resuscitated with chloride rich crystalloids are frequently exposed to hyperchloremia. However, the impact of hyperchloremia on outcomes might be negligible compared to the inherent risks from acute disease. Further studies are required to identify the specific role of chloride disturbances and use of balanced crystalloids on the outcome of septic shock patients.