Morbidity and Mortality of Crystalloids Compared to Colloids in Critically Ill Surgical Patients: A Subgroup Analysis of a Randomized Trial

PERIOPERATIVE hemodynamic instability may lead to cardiovascular morbidity and requires prompt recognition and correction. Possible causes include blood loss, fluid deficit, or sepsis. Fluid therapy is therefore a key component of the perioperative management of surgical patients. Resuscitation fluids are divided into two categories: colloid and crystalloid solutions. The ideal fluid to be used in the surgical setting remains uncertain.^1,2  Colloids are composed of heavy molecular weight molecules, which are retained in the plasma compartment. Hemodynamic goals are reached by administrating smaller volumes of colloids than crystalloids.^3–5  Among colloids, starches are the most commonly administered fluid. The use of starches has been restricted by the European Medicines Agency in sepsis, burns, or critically ill patients⁶  because of the risk of acute kidney injury and of death.^5,7  The U.S. Food and Drug Administration also issued a warning about the increased risk of renal failure or death, as well as a risk of bleeding after cardiopulmonary bypass associated with starches.⁶  However, because these data do not derive exclusively from surgical patients, extrapolation of these findings to the perioperative period is questionable. Indeed, crystalloids are not devoid of side effects such as hyperchloremic metabolic acidosis, reduced renal blood flow, or impaired renal cortical perfusion.^8,9 

The Colloids versus Crystalloids for the Resuscitation of the Critically Ill (CRISTAL) trial was designed to test the hypothesis that colloids altered 28-day mortality compared with crystalloids for fluid resuscitation in a general population of critically ill patients.⁴  This a priori defined secondary analysis tested the same hypothesis in the subgroup of surgical critically ill patients. Patients were identified as surgical whenever they underwent emergency or scheduled surgery immediately before or within 24 h of intensive care unit admission. Our primary outcome was 28-day mortality. Secondary outcomes included the occurrence of organ dysfunction over a 28-day period, as well as the need for renal replacement therapy, secondary surgical intervention, blood product administration, intensive care unit and hospital length of stay, and 90-day mortality.

The CRISTAL trial (ClinicalTrials.gov NCT00318942) randomly assigned 2,857 acutely hypovolemic patients from 57 participating centers in Europe, North Africa, and North America to receive either crystalloids or colloids.⁴  The study protocol was approved by local institutional review boards. Deferred written informed consent was obtained from participants or legally authorized surrogates. Included participants had not previously received any fluid in the intensive care unit and required fluid resuscitation for acute hypovolemia. Acute hypovolemia was defined by the combination of (1) hypotension: systolic arterial pressure of less than 90 mmHg, mean arterial pressure of less than 60 mmHg, orthostatic hypotension (i.e., decrease in systolic arterial pressure of at least 20 mmHg, from the supine to the semirecumbent position), or a delta pulse pressure of 13% or higher; (2) evidence for low filling pressures and low cardiac index, assessed either invasively or noninvasively; and (3) signs of tissue hypoperfusion or hypoxia, including at least two of the following clinical symptoms: Glasgow Coma Scale score of less than 12, mottled skin, urinary output of less than 25 ml/h, or capillary refilling time of 3 s or longer; and arterial lactate levels higher than 2 mM, blood urea nitrogen higher than 56 mg/dl, or a fractional excretion of sodium of less than 1%.⁴ 

A computer-generated list with fixed-block permutation (n = 4) was used to randomize patients on a 1 to 1 ratio. Randomization was stratified by center and by three admission diagnoses: sepsis, multiple trauma, and other causes of hypovolemic shock. Allocation concealment used sealed envelopes at the bedside to allow randomization of eligible patients without any delay. Investigators were blinded to block size.

Eligible patients were randomly allocated to fluid resuscitation with crystalloids or with colloids. In the crystalloids group, allowed treatments included isotonic or hypertonic saline and any buffered solutions. In the colloids group, hypooncotic (e.g., gelatins, and 4 or 5% of albumin) and hyperoncotic (e.g., dextrans, hydroxyethyl starches, and 20 or 25% of albumin) solutions were permitted. Within each treatment group, investigators could use whichever fluids were available at their institution. The amount of fluid and duration of treatment was left at the discretion of the investigators with the following restrictions: (1) the daily total dose of hydroxyethyl starch could not exceed 30 ml/kg of body weight and (2) investigators were required to follow any local regulatory agency recommendations governing use. Patients had to be managed according to their randomization arm except for (1) maintenance fluids, which were isotonic crystalloids, regardless of treatment group, and (2) in instances in which physicians wished to administer albumin in response to demonstrated hypoalbuminemia (serum albumin concentration less than 20 g/dl).

The blinding of the clinicians to the fluid interventions was considered by the study advisors to be inappropriate or infeasible because study treatments had to be available immediately for resuscitation to ensure avoidance of nonstudy fluids in emergent situations. In addition, because the intervention would be continued until intensive care unit discharge and could thus be highly variable, there was no practical way to stock sites with adequate supplies of masked fluid solutions. However, the mortality endpoints were collected and assessed by study members blinded to treatment assignment. Similarly, the principal investigator, study sponsor, and the members of the data and safety monitoring board remained blinded to the study interventions until all patients were followed up and the final analysis was executed.

For this analysis, we included all surgical patients included in the original trial. Surgical patients were a priori defined as patients requiring elective or unscheduled surgery, either before or up to 24 h after intensive care unit admission.

At the time of randomization, age, sex, cause of intensive care unit admission, type of admission (medical, elective surgery, unscheduled surgery, trauma), McCabe class,¹⁰  disability scale score,¹¹  cause of hypovolemia (divided into three separate strata: sepsis, trauma, and other), Simplified Acute Physiology Score II,¹²  Sequential Organ Failure Assessment score,¹³  Injury Severity Score,¹⁴  signs of hypovolemia, and amount of fluids administered before randomization were collected. The global Sequential Organ Failure Assessment score was recorded daily over a 7-day period and thereafter on days 14 and 28. Any occurrence of renal replacement therapy was recorded. We assessed the need for secondary surgical intervention and for blood product administration (including platelets, fresh frozen plasma, and packed erythrocytes) over a 7-day period. Outcomes included intensive care unit and hospital length of stay, as well as death by days 28 and 90.

Quantitative variables were expressed as median (interquartile range) and categorical variables as number (percentage). When designing the CRISTAL trial, we anticipated that responses to colloids versus crystalloids may vary across different groups of patients, namely sepsis, trauma, and other categories of acute hypovolemia. Thus, randomization was stratified according to these three groups of patients. In addition, we anticipated potential qualitative interaction between treatment responses within each of these strata and the type of admission, namely surgical versus medical. Thus, we planned to report the estimation of treatment effects in the surgical and medical groups of patients separately. We undertook an intention-to-treat analysis for the primary outcome, death by day 28, where patients, once selected in one treatment group according to randomization, were analyzed in the group assigned by the randomization, insuring the absence of any selection or attrition bias. No imputation was used. Nevertheless, in response to peer review, per-protocol analyses were added as a secondary analysis including all participants who adhered adequately to the assigned treatment. Categorical variables were compared with the Fisher exact test, and continuous variables were compared with the Wilcoxon rank sum test. To assess differences over time of the Sequential Organ Failure Assessment score across both arms, we built a linear mixed-effects model. This allowed us to model observational heterogeneity incurred by repeat measurements of the score in the same patient (with fixed effects of Sequential Organ Failure Assessment and time) and accounted for the fact that some individuals may have higher values than others (by using a random intercept).

All patients were followed until day 90 unless death occurred before day 90, so that analysis of mortality data across randomized groups used the chi-square test; estimate of odds ratio of death according to fluid used a logistic model, adjusted to the nature of surgery. To display the cumulative incidence of death, we used nonparametric estimator and then compared between randomized groups by the Gray test.

All analyses were preplanned, except for those factors selected for adjustment and additional analyses requested by the reviewers or editors. Statistical analyses were performed with SAS 9.3 (SAS Inc., USA) and R 2.13.0 (http://www.R-project.org/; accessed August 10, 2017) software. Tests were two-sided. The results were adjusted for multiple comparisons.¹⁵ P levels less than 0.05 were considered statistically significant.

Of 2,857 patients in the initial trial, there were 741 critically ill surgical patients (fig. 1). Of those 741 surgical patients, 484 (65.3%) were male, 369 (49.8%) suffered from sepsis, and the median age was 66 (52 to 76) yr. In total, 356 patients (48%) were allocated to the crystalloids arm, and 385 (52%) were allocated to the colloids arm. Surgery was elective for 198 (26.7%) patients and unscheduled for 543 (73.3%) patients; 395 patients (53.3%) underwent general or abdominal surgery, 198 patients (26.7%) underwent orthopedic surgery, 127 patients (17.1%) underwent cardiac surgery, and 21 patients (2.9%) underwent neurosurgery. In both arms, some patients received crystalloids and/or colloids in the operating theater before intensive care unit admission and before randomization (table 1). Nevertheless, severe acute hypovolemia was present upon randomization as highlighted by tachycardia, low systolic blood pressure and high diastolic blood pressure, low cardiac index, and high arterial lactate levels (Supplemental Digital Content, supplemental table 1, https://links.lww.com/ALN/B772).

Patients in the crystalloids arm received a median 7-day cumulative dose of fluids of 4,275 (2,000 to 7,500) ml versus 2,750 (1,500 to 4,500) ml for patients in the colloids arm (table 2). By day 1, the amount of fluids was markedly (P < 0.0008) higher in the crystalloids than in the colloids-treated patients (fig. 2). Over time, 14.8% (n = 57) of patients in the colloids arm received crystalloids, whereas 5.6% (n = 20) of patients in the crystalloids arm received colloids.

There was no difference in the occurrence of death by day 28 between groups: 84 (23.6%) in the crystalloids arm compared to 100 (26%) patients in the colloids arm (adjusted odds ratio, 0.86; 95% CI, 0.61 to 1.21; P = 0.768; fig. 3). No interaction of the intervention with any of the randomization strata (sepsis, trauma, and other) was found (fig. 4). No interaction of the intervention with the type of surgery was found (fig. 5). There was no difference in the occurrence of death by day 90 between groups: 111 (31.2%) in the crystalloids arm compared to 122 (31.7%) patients in the colloids arm (adjusted odds ratio, 0.97; 95% CI, 0.70 to 1.33; P = 0.919; fig. 6). The time course of the global Sequential Organ Failure Assessment score was similar in both groups (P = 0.915; fig. 7). The median number of days alive within the first 7 days with a Sequential Organ Failure Assessment score less than 6 did not significantly differ between arms: 2 (0 to 4.25) days in the crystalloids arm versus 2 (0 to 4) days in the colloids arm (P = 0.786). Renal replacement therapy was required for 42 patients (11.8%) in the crystalloids arm versus 49 patients (12.7%) in the colloids arm, over a 7-day period (P = 0.897). The median length of stay in the intensive care unit did not significantly differ: 7 (3 to 17) days in the crystalloids arm compared to 7 (3 to 15) days in the colloids arm (P = 0.855). The median length of stay in the hospital was 22 (11 to 40) days in the crystalloids arm compared to 21 (10.5 to 38) days in the colloids arm (P = 0.815). We also performed a per-protocol comparison of the 340 patients who effectively received colloids and the 259 patients who effectively received crystalloids. Of note, we excluded from the per-protocol analysis all patients in the crystalloid arm who had been treated by albumin even though the protocol provided for albumin supplementation in case of hypoalbuminemia. The per-protocol analysis did not find any difference in the rate of death by day 28 (87 [25.6%] vs. 54 [20.8%]; adjusted odds ratio, 0.75; 95% CI, 0.51 to 1.12; P = 0.429). The results of the per-protocol analysis are in table 3.

Secondary surgery was required in 278 (37.5%) patients, of which 131 (36.8%) were in the crystalloids group and 147 (38.2%) in the colloids group (P = 0.875). A median number of 2.7 (2.2 to 3.4) units of packed erythrocytes were administered to patients in the crystalloids arm compared to 2.7 (2.3 to 3.5) units in the colloids arm (P = 0.890). A median number of 2 (1 to 5) units of platelets was administered to patients in the crystalloids arm compared to 2 (1 to 3) units in the colloids arm (P = 0.533). A median volume of 450 (175 to 800) ml of fresh frozen plasma was administered in the crystalloids arm compared to 600 (400 to 1,200) ml in the colloids arm (P = 0.108). Additional information regarding the requirement for blood products are found in Supplemental Digital Content, supplemental tables 2 to 7 (https://links.lww.com/ALN/B772). The results of the per-protocol analysis regarding surgical interventions and blood transfusions are found in table 3.

In this subgroup analysis of a large pragmatic trial comparing the administration of crystalloids to colloids in surgical patients, we found that colloids did not exhibit a significantly different safety profile from crystalloids. The current analysis encompasses both elective and unscheduled surgery associated with clinically significant hypovolemia. Our main finding was that 28-day mortality did not differ between treatment arms, both in the intention-to-treat and the per-protocol analyses. Additionally, mortality by day 28 did not differ in any of the three prespecified strata. Mortality by day 90 did not differ between treatment arms, neither in the intention-to-treat nor in the per-protocol analysis. In-hospital and in–intensive care unit length of stay did not differ between treatment arms. The need for secondary surgical interventions did not significantly differ between groups. The amounts of packed erythrocytes, of platelets, and of fresh frozen plasma did not differ between patients treated by crystalloids and those receiving colloids. Importantly, the global Sequential Organ Failure Assessment score and the need for renal replacement therapy did not differ between groups. Because patients were randomized in CRISTAL to receive products belonging to a broad family of fluids, extrapolating our findings to a particular type of solution, such as balanced solutions, is hazardous.¹⁶  We did not record enough data to provide information regarding the development of acute kidney injury in this cohort. However, in similar trials, the occurrence of acute kidney injury was found to be inconsistent with other markers of kidney failure, such as the requirements for renal replacement therapy.^5,7 

The whole population of the CRISTAL trial mainly encompassed patients admitted to the intensive care unit for medical reasons (71%).⁴  In the primary analysis of the CRISTAL trial, death by day 28 did not differ significantly between the colloids and crystalloids groups (relative risk, 0.96; 95% CI, 0.88 to 1.04; P = 0.26), and mortality by day 90 was significantly lower in the colloids arm (relative risk, 0.92; 95% CI, 0.86 to 0.99; P = 0.03). In this exploratory analysis of the surgical population, mortality rates by days 28 and 90 are broadly similar to that of the global CRISTAL population, as well as the risk for renal replacement therapy requirement (relative risk, 0.93; 95% CI, 0.83 to 1.03; P = 0.19). Data pertaining more specifically to the surgical subpopulation (e.g., bleeding risk and need for secondary surgery) were not analyzed in the whole population of the CRISTAL trial.

The most thoroughly studied subtype of colloid in the surgical context is starch. Retrospective studies hint at the possibility of acute kidney injury after the administration of starches in the surgical setting.¹⁷  However, several small randomized controlled trials in elective surgery reported no renal side effect related to hydroxyethyl starch administration.^18,19  In the Crystalloid versus Hydorxyethyl Starch Trial, the number of surgical patients undergoing renal replacement therapy did not differ between both groups: 61 of 1,425 assigned to hydroxyethyl starch (4.3%) versus 45 of 1,447 assigned to saline (3.0%) (relative risk, 1.38; 95% CI, 0.94 to 2.01).²⁰  The Fluids in Resuscitation of Severe Trauma trial, comparing the administration of starches to that of crystalloids in both blunt and penetrating trauma, found no difference in mortality, whereas acute kidney injury occurred more frequently in the saline group.²¹  A recent trial found that the use of colloids was associated with fewer complications than use of crystalloids in elective abdominal surgery.²²  Several meta-analyses of surgical patients concluded that starches do not induce additional renal injury.^23–25  Although these analyses have been criticized,^26,27  our findings, namely no increased need for renal replacement therapy associated with the administration of colloids, are in keeping with previous reports. Such a difference in the occurrence of acute kidney injury between subjects affected by sepsis and those affected by surgery or trauma may be related to different pathophysiology of acute kidney injury. If blood loss from trauma or during surgery leads to hypotension and renal ischemia, prompt restoration of renal hemodynamic may reduce the incidence of acute kidney injury.²⁸  The pathophysiology of sepsis-associated acute kidney injury is wholly different, because acute kidney injury may occur despite normal renal blood flow.²⁹  Pathophysiologic mechanisms of sepsis-associated acute kidney injury include inflammation, microcirculatory dysfunction, and endothelial cell injury.³⁰  Starches may also affect bleeding in surgical patients. Several trials comparing starches to crystalloids found that starches reduced the clot strength and increased bleeding during both major surgery and cardiac surgery.^31–33  In the Fluids in Resuscitation of Severe Trauma trial, patients suffering from blunt trauma randomized to the hydroxyethyl starch group required significantly more blood products than those randomized to receive saline.²¹  We did not observe any difference in the required amount of packed erythrocytes, platelets, or fresh frozen plasma. Little is known of the effect of colloid versus crystalloid solutions on mortality in the surgical setting, because surgical-related mortality in most circumstances is extremely low. In the subgroup of patients undergoing surgery before randomization into the 6S trial, death by day 90 occurred in 61 of 131 patients in the hydroxyethyl starch subgroup versus 53 of 146 patients in the Ringer’s acetate subgroup (relative risk, 1.28; 95% CI, 0.97 to 1.70; P = 0.42).³⁴  A last point to be mentioned when comparing crystalloids to colloids is the cost of each product. CRISTAL was not designed to analyze the cost of both interventions.

No power computation was performed when this secondary analysis of the trial was undertaken. Nevertheless, given the sample size of 741 patients broadly divided into two equal-sized groups, the statistical power to detect an effect size of at least 0.2 was above 80%. Strengths of the current study include the fact that our data set stems from a pragmatic, randomized clinical trial, depicting the use of resuscitation fluids in real-world conditions. Additionally, our analyses were preplanned, and we report on a small number of outcomes, reducing the risk of false-positive results.³⁵  Our study has several limitations. First, we did not have access to perioperative blood loss; we therefore had to use a surrogate marker of blood loss, the number of packed erythrocytes administered. Second, because our population consists of patients who were transferred from the operating theater to the intensive care unit, extrapolation of our findings to all patients managed in the operating theater requires careful consideration.

In surgical patients included in the CRISTAL trial, we found no difference between colloids and crystalloids regarding safety, namely the risk of death or of organ failure, including acute kidney injury. The safety of colloids was comparable to that of crystalloids in our population of surgical patients treated for hypovolemic shock.

The authors thank Julie Lejeune, M.S. (Biostatistical Unit, Saint Louis Hospital, Paris, France), for her technical assistance, as well as all the investigators and patients of the CRISTAL trial.

Supported in 2001 and 2010 by grant No. AOM 01 020 from the French Ministry of Health.

The authors declare no competing interests.