Medical management of cerebral edema and increased intracranial pressure (ICP) are critical components of perioperative care in neurosurgical practice. During a craniotomy, tumor bulk and vasogenic edema can generate clinical situations in which adequate intracranial volume management becomes a key factor to facilitate surgical removal of the tumor.
Twenty percent mannitol, usually given IV in a bolus dose of 0.5 to 1 g·kg−1, has been widely used to reduce brain bulk and facilitate the surgical approach.1 However, to our knowledge, no prospective study has examined the dose-response relationship of mannitol on brain relaxation in the surgical setting. We therefore designed this prospective, randomized, double-blind study to assess the difference in terms of brain relaxation between 2 doses of mannitol (0.7 and 1.4 g·kg−1) during elective supratentorial brain tumor surgery. We hypothesized that a high dose of mannitol (1.4 g·kg−1) would result in a better relaxation score than a lower dose of mannitol (0.7 g·kg−1).
METHODS
After Centre Hospitalier de l’Universite de Montreal (CHUM) ethical and scientific committee approval and written informed consent, 80 consecutive adult patients scheduled for an elective supratentorial brain tumor craniotomy were enrolled by our research assistant (MR) in this study. Exclusion criteria were ASA physical status IV and V, Glasgow Coma Scale score <13, preoperative hyponatremia or hypernatremia (serum Na <130 or >150 mEq·L−1), treatment with a hyperosmotic agent (mannitol or hypertonic saline) in the previous week, presence of congestive heart failure (ejection fraction <20%) or renal failure (creatinine clearance <30 mL·kg−1). After randomization (computer-generated random list by blocks of 4 patients in sealed envelopes created by our research assistant [MR]), patients were assigned to receive 0.7 g·kg−1 (group L) or 1.4 g·kg−1 up to a maximum of 100 g (group H) of 20% mannitol administered at skin incision over 30 minutes using an infusion pump. These doses of mannitol were chosen according to our standard clinical practice to obtain 50 and 100 g of mannitol for a 70-kg individual for the low- and high-dose group, respectively. Both the attending anesthesiologist and surgeon were blinded to the amount of mannitol given to the patient.
General anesthesia induction consisted of propofol (1–3 mg·kg−¹) and sufentanil (0.2 µg·kg−¹). Tracheal intubation was facilitated with rocuronium (0.9–1.2 mg·kg−¹). Anesthesia was maintained with desflurane (0.7–1 minimum alveolar concentration) in oxygen and air (fraction of inspired oxygen 0.50) along with a sufentanil infusion. Muscle relaxants were used as needed to maintain a single twitch on train-of-four stimulation. A radial artery catheter was inserted for arterial blood pressure monitoring and blood sampling. All patients received IV dexamethasone 4 mg before skin incision. Lumbar or ventricular drains were not inserted during the surgery.
Bladder temperature was maintained between 36°C and 37.4°C and the PaCO2 between 30 and 35 mm Hg. IV fluids were managed according to our standard practice: urine output was replaced with NaCl 0.9% and blood loss with a 1:1 ratio of hydroxyethyl starch 6% (Voluven®). The blood transfusion threshold was set at a hemoglobin value of 80 mg·dL−1.
Brain relaxation was assessed by the attending neurosurgeon immediately after opening of the dura on a scale ranging from 1 to 4 (1 = perfectly relaxed, 2 = satisfactorily relaxed, 3 = firm brain, 4 = bulging brain).2 If needed to facilitate surgical exposure, in case of significant cerebral edema (brain relaxation score 3 or 4), a rescue dose of 20% mannitol (0.25 g·kg−1) was administered.
Hemodynamic variables (mean arterial blood pressure, heart rate), temperature, urine output, perioperative fluid balance, blood loss, and laboratory values (blood gases, electrolytes, osmolarity, hematocrit, glycemia, lactates) were collected immediately before the infusion of mannitol and 30, 60, and 180 minutes after the administration of the study drug. Head position was also measured: angulation relative to the ground, rotation, flexion, or extension.
The type of cerebral lesion, its location and size on magnetic resonance imaging, as well as the presence and magnitude of a midline shift (defined as a deviation of >1 mm) were evaluated by the attending neurosurgeon.
Sample size calculation was based on the results obtained by Rozet et al.3 These authors obtained the following scores for their 2 study groups: 6/8/6/0 for hypertonic saline and 5/9/4/2 for mannitol using the same 4-point scale we used in the current study. We based our calculation on an average score computed using the 2 study groups of Rozet et al.3 Using a difference between the means of 1 with the same variance for the 2 groups, 20 patients in each group were deemed necessary to detect a difference of 1 point on the cerebral relaxation scale, a difference we defined as clinically significant, with a power of 90% and an α of 0.05 (2-tailed). A sample size of 40 patients per group was chosen to allow for a certain margin of safety given the suspected wide interpersonal variation in this group of patients.
Comparisons between the 2 arms of the trial were performed using the χ2 test for the categorical variables and t test for the continuous variables. Variables for which the P value was statistically significant, using a threshold of 0.05, were considered unbalanced between the 2 treatment arms.
The effect of the different doses of mannitol on brain relaxation was estimated using the proportional odds (PO) model, which is an extension of the logistic model for ordinal response variable of >2 categories. When estimated on an ordinal variable with 4 categories, the PO model estimates a single common odds ratio (OR) for the comparison of (a) category 1 versus 2, 3, 4; (b) categories 1 and 2 versus 3 and 4; and (c) categories 1, 2, 3 versus category 4. Hence, an OR >1 for the comparison of the treatment arm (higher dose of mannitol in this case) versus the control arm (lower dose of mannitol) is indicative of more favorable outcome in the higher-dose arm compared with the lower-dose group. For the purpose of the PO model, the response variable (brain relaxation score) was coded in reversed order: 1 = bulging brain, 2 = firm brain, 3 = satisfactorily relaxed, 4 = perfectly relaxed.
Hence, an OR superior to 1 for the treatment group illustrates the odds of a 1-point increase on the brain relaxation scale for the high-dose (H) group in comparison to the low-dose (L) group.
Similarly, an OR inferior to 1 for the treatment group illustrates the odds of a 1-point decrease on the brain relaxation scale for patients with a midline shift in comparison to those without.
An additional PO model was estimated to adjust the treatment effect for any variable whose distribution was deemed unbalanced between the 2 treatment arms. This is the second model presented in Table 1.
;)
Table 1: Proportional Odds Models
The validity of the PO model results depends on whether the hypothesis of PO is respected. For each PO model, we tested the proportionality of odds assumption using a difference of deviance test with the corresponding model that does not assume the proportionality of odds.4 For both models, the P value was respectively 0.384 and 0.457 indicating that the assumption is not rejected and that the results are valid.
The following procedure was undertaken to study the effect of the treatment on each of the 5 outcomes found in Table 2 (sodium, potassium, osmolarity, lactate, and urine output), at each point in time, using a linear mixed model for repeated measures with an interaction between time and treatment. Step 1: Evaluate the normality assumption of the outcome variable conditional on the time and group variables using QQplots and Shapiro tests. In case of violations of normality, transformations of the outcome, such as the log, were considered. Step 2: Evaluate the influence of potential outliers using the model selected in step 1. Exclude outliers (at the individual level) using visual inspections of dfbetas and Cook’s distance measures and reestimate the model, the QQplots, and the Shapiro tests. If the interaction between group and time is not statistically significant, remove it from the model. Step 3: Investigate the potential effect of heteroscedasticity (unequal variance over time) by fitting an alternative mixed model in which the variance of the outcome is allowed to differ over time. Compare the fit of that model with the model obtained in step 2 and select the model with the best fit using a likelihood ratio test. Step 4: Using the model selected in step 3, compute simultaneous (family-wise) confidence intervals (CIs) and P values for the difference between the 2 treatment groups at each point in time, thus factoring in the effect of multiple comparisons.5
Table 2: Electrolytes, Osmolarity, Lactates, and Urine Output Variation During the Study Period
All of the 5 outcomes were log-transformed to fit the normality requirement of the mixed model, conditional on the factors included in the model. For all models except that of lactate (Table 2), the interaction between treatment group and time was statistically significant at the 0.05 level, indicating that the effect of the treatment differed over time. Selected influential outliers (generally extreme values) were detected and thus removed from the outcome-related analysis: Na (2 in group H, 1 in group L), potassium (1 in group H), osmolarity (1 in group H, 1 in group L), lactate (1 in group L), and urine output (2 in group L). For osmolarity, lactate, and urine output, the adjustment for heteroscedasticity was necessary to account for the variances in outcomes that tended to differ over time.
Statistical analyses were conducted using R 2.15.0 (R Development Core Team, 2012. R: a language and environment for statistical computing; R Foundation for Statistical Computing, Vienna, Austria) and the following packages: multcomp,5 nlme,6 influence, ME,7 and contrast.8
RESULTS
From May 2010 to April 2011, 92 consecutive patients were considered for the study. Of this number, 12 patients did not meet inclusion criteria or refused to participate in the study. The remaining 80 patients were enrolled, 40 in each group. No patient was excluded from the study after randomization. There was no significant difference between the 2 groups regarding age, sex, body mass index, and brain tumor localization or size (Table 3). In group L 52.5% of patients and in group H 77.5% of patients presented a midline shift (P = 0.03). At the time of mannitol administration, patients in both groups had similar hemodynamics, PaCO2, temperature, head positioning, and laboratory values (Table 4).
Table 3: Patient Characteristics
Table 4: Baseline Intraoperative Data
Brain relaxation scores are presented in Table 5. The median scores (interquantile range) were 2.0 (1.75–3.0) and 2.0 (1–3) for patients in groups L and H, respectively (P = 0.41).
Table 5: Brain Relaxation Scores
In Table 1, we present both PO models. In the first model (not adjusted for midline shift) we can see that a higher dose of mannitol is not statistically associated with better relaxation scores (OR 1.8, 95% CI = 0.82–4.07, P = 0.14). However, in the second model, when adjusted for the presence of midline shift, the use of a higher dose of mannitol resulted in an OR of 2.5 (95% CI = 1.08–5.94, P = 0.03). This indicates that, taking into consideration the effect of a midline shift, the odds of having a 1-level improvement in relaxation score in patients who received a higher dose of mannitol (group H) was 2.5 times as large than the odds for the low-dose group. The OR of 0.29 (95% CI = 0.11–0.70, P = 0.007) for the midline shift indicates that it is associated with a higher probability of less-favorable relaxation scores on the 4-point scale. Figures 1 to 3 show the proportion probability plot by treatment groups, for the total study population and also stratified by midline shift.
Figure 1: Brain relaxation score, proportion plot, total study population. Brain relaxation score on the x-axis. Proportion (not cumulative) of patients for each relaxation score on the y-axis. Low-dose group in red and high-dose group in blue.
Table 2 presents the treatment group specific means and SD for each of the 5 outcomes (electrolytes, osmolarity, lactates, and urine output variation) at the 4 time points during the study. The family-wise P values, testing the treatment group difference in means at each point in time, were derived from the mixed models described in the Methods section, thus adjusted for the multiple pairwise comparisons.5
DISCUSSION
In this study, we show that 1.4 g·kg−1 of 20% mannitol results in equivalent brain relaxation scores as 0.7 g·kg−1 in patients undergoing craniotomy for supratentorial brain tumor. This is contrary to our a priori hypothesis. Unfortunately, patients presenting to surgery with a shift of the midline structure were unevenly distributed between our 2 study groups: 52.5% in group L and 77.5% in group H (P = 0.03). We therefore used a PO model to adjust for this unbalanced distribution. When corrected for the presence of midline shift, this analysis reveals that patients in the high-dose group (1.4 g·kg−1 of 20% mannitol in this study) had more chances of obtaining a better relaxation score compared with the lower-dose group. The fact that we could not demonstrate this with basic statistical comparisons in the overall study population is linked to the uneven distribution of patients with midline shift.
Mannitol is often recommended as a first choice hyperosmotic drug for the treatment of increased ICP and to decrease brain bulk during intracranial surgery.1 However, dosage regimens and thresholds for treatment still vary widely, and it is our experience that they are not necessarily tailored to the patient’s clinical situation. For example, a standard dosage is frequently chosen for a given surgical procedure notwithstanding the size of the lesion and the resulting mass effect. In fact, we could not find another study in the literature that compared different dosage regimens of mannitol in intracranial surgery.
In the intensive care setting, in patients with traumatic brain injury, Sorani et al.9 performed a retrospective study to characterize the dose-response relationship between mannitol and ICP. They included 28 patients who received a total of 135 doses of mannitol. The authors found a dose-dependent relationship between the administration of 50 and 100 g of mannitol and the ICP response. However, their statistical model did not consider the fact that the doses of mannitol were not independent variables in this study because there were several administrations of mannitol for each patient. This interaction should have been addressed in a more sophisticated statistical model. This and the fact that the study population and the clinical setting (intensive care unit) are different preclude comparisons with our study.
In recent years, 2 prospective studies have compared the effects of hypertonic saline and mannitol on brain relaxation.10,11 In a prospective, double-blind, randomized study including 40 patients undergoing elective craniotomy, Rozet et al.3 showed that mannitol and hypertonic saline resulted in similar brain relaxation scores, a median of 2 on the same 4-point scale used in our study. It is difficult to compare these data with those of our study, because a wide variety of pathologies were included in the study by Rozet et al. Only 10 patients presented with a supratentorial brain tumor, 6 of whom were in the mannitol group. No data regarding the severity of intracranial hypertension were provided for these patients, and a single dose of mannitol was administered (1 g·kg−1).
In another study comparing the effect of mannitol versus hypertonic saline on brain relaxation, Wu et al.10 showed that the use of mannitol was associated with poorer surgical conditions. In 116 patients undergoing supratentorial brain tumor craniotomy, 70% of patients had adequate brain relaxation in the mannitol group compared with 83% in the hypertonic saline group. Again, comparisons with our study are difficult because of the use of a 3-point scale and the absence of details on tumor size and midline shift. In our study, 60% of patients had adequate brain relaxation scores (grades 1 and 2) overall (55% in group L and 65% in group H).
As previously shown in other studies, increasing the dose of mannitol results in an increase in osmolarity, a decrease in serum sodium concentration, and an increase in urine output.11–15 The development of hyponatremia can be explained by the changes in osmolarity and the initial volume shift toward the intravascular compartment along the osmolar gradient and the resulting hemodilution. Higher doses of mannitol resulted in a dose-related increase in osmolarity, presumably a similar dose-related decrease in brain water content, and thus the better relaxation scores obtained in the group of patients with a midline shift in group H.8,9 The small clinically nonsignificant increase in potassium and lactate concentrations found in our study are consistent with previously published data. Manninen et al.11 described a significant increase in serum potassium level, reaching a maximum mean increase of 1.5 mmol·L−1 after high-dose mannitol (2 g·kg−1) in a group of 7 patients undergoing cerebral aneurysm clipping. The exact mechanism of this increase is still unknown. Rozet et al.3 suggested that because mannitol has a prominent diuretic effect, the resulting relative hypovolemia associated with its use could explain the increased blood lactate levels.
This study has some limitations. Even if the 4-point scale used to evaluate brain relaxation in this study was used by other authors, it is still a subjective scale. As such, it is subject to interpretation. Nonetheless, it reflects a significant end point because surgical conditions are always evaluated by the attending neurosurgeon, and clinical decisions are usually based on such evaluation. Clearly, our study was not designed to assess the weight of our intervention on patients’ outcomes. We can only speculate that poor brain condition, as assessed by the operating surgeon, will trigger an important clinical decision. Some of these decisions such as increasing the dimension of the craniotomy, initiating hyperventilation, opting for a partial resection, or increasing the pressure on brain retractors to optimize surgical exposure might negatively affect a patient’s outcome. Even if the difference between the 2 groups found in this study was small and demonstrated only in a subgroup of patients presenting with a midline shift, these are the patients for whom a tailored intervention is the most important. The difference between a satisfactorily relaxed brain (grade 2 on the 4-point scale) and a firm brain (grade 3) is critical. As such, the results obtained in this study certainly support the practice of anesthesiologists already using a tailored approach and they suggest that patients with a midline shift should be given a higher dose of mannitol.
ICP was not measured in this study for 2 main reasons. One is the issue of safety. We did not want to subject the patient to the additional risk associated with the insertion of a Camino monitor or a ventricular/lumbar drain. In addition, ICP is usually not measured during intracranial tumor surgery and thus seldom used to make intraoperative clinical decisions regarding brain relaxation. The higher dosage of mannitol used in this study (1.4 g·kg−1) is not per se within the usual range used in neurosurgery. Our conclusions might therefore not apply to lower doses of mannitol. Finally, the type and size of the craniotomy was not considered. This is a factor that affects surgical exposure, but given the design of this study, it is unlikely that a systematic bias was introduced.
To our knowledge, this is the first prospective study designed to demonstrate a differential effect of 2 doses of mannitol on brain relaxation during supratentorial tumor surgery. This study suggests that the dosage of mannitol should be tailored to clinical markers of increased ICP such as radiologic midline shift. In this subgroup of patients, a higher dosage of mannitol resulted in better relaxation scores than lower doses. These results suggest a dose-effect relationship in this population of patients.
Figure 2: Brain relaxation score, proportion plot, patients with midline shift. Brain relaxation score on the x-axis. Proportion (not cumulative) of patients for each relaxation score on the y-axis. Low-dose group in red and high-dose group in blue.
Figure 3: Brain relaxation score, proportion plot, patients with no midline shift. Brain relaxation score on the x-axis. Proportion (not cumulative) of patients for each relaxation score on the y-axis. Low-dose group in red and high-dose group in blue.
DISCLOSURES
Name: Charlotte Quentin, MD.
Contribution: This author helped design the study, conduct the study, analyze the data, and write the manuscript.
Attestation: Charlotte Quentin has seen the original study data, reviewed the analysis of the data, and approved the final manuscript.
Name: Sonia Charbonneau, MD.
Contribution: This author helped design the study, conduct the study, analyze the data, and write the manuscript.
Attestation: Sonia Charbonneau has seen the original study data, reviewed the analysis of the data, and approved the final manuscript.
Name: Robert Moumdjian, MD.
Contribution: This author helped design the study, conduct the study, analyze the data, and write the manuscript.
Attestation: Robert Moumdjian has seen the original study data, reviewed the analysis of the data, and approved the final manuscript.
Name: Alexandre Lallo, MD.
Contribution: This author helped design the study, conduct the study, analyze the data, and write the manuscript.
Attestation: Alexandre Lallo has seen the original study data, reviewed the analysis of the data, and approved the final manuscript.
Name: Alain Bouthilier, MD.
Contribution: This author helped design the study, conduct the study, analyze the data, and write the manuscript.
Attestation: Alain Bouthilier has seen the original study data, reviewed the analysis of the data, and approved the final manuscript.
Name: Marie-Pierre Fournier-Gosselin, MD.
Contribution: This author helped design the study, conduct the study, analyze the data, and write the manuscript.
Attestation: Marie-Pierre Fournier-Gosselin has seen the original study data, reviewed the analysis of the data, and approved the final manuscript.
Name: Michel Bojanowski, MD.
Contribution: This author helped design the study, conduct the study, analyze the data, and write the manuscript.
Attestation: Michel Bojanowski has seen the original study data, reviewed the analysis of the data, and approved the final manuscript.
Name: Monique Ruel, RN.
Contribution: This author helped design the study, conduct the study, analyze the data, and write the manuscript.
Attestation: Monique Ruel has seen the original study data, reviewed the analysis of the data, and approved the final manuscript.
Name: Marie-Pierre Sylvestre, PhD.
Contribution: This author helped analyze the data and write the manuscript.
Attestation: Marie-Pierre Sylvestre has seen the original study data, reviewed the analysis of the data, and approved the final manuscript.
Name: Francois Girard, MD.
Contribution: This author helped design the study, conduct the study, analyze the data, and write the manuscript.
Attestation: Francois Girard has seen the original study data, reviewed the analysis of the data, approved the final manuscript, and is the author responsible for archiving the study files.
This manuscript was handled by: Gregory J. Crosby, MD.
