Validation de l’algorithme de perfusion intraveineuse d’insuline DeLiT Trial pour le contrôle glycémique peropératoire en chirurgie non cardiaque: une étude randomisée contrôlée
- First Online:
- 589 Downloads
Nous ne disposons pas encore d’un algorithme de perfusion d’insuline formellement caractérisé et évalué, qui soit à la fois sécuritaire et efficace, et qui permette de maintenir un contrôle glycémique peropératoire rigoureux en chirurgie non cardiaque. C’est pourquoi nous rapportons la validation de notre algorithme de perfusion de l’insuline.
Des patients devant subir une chirurgie non cardiaque majeure ont été randomisés à une concentration glycémique peropératoire cible de 4,4-6,1 mmoL·L−1 (80-110 mg·dL−1) dans le groupe intensif ou à une concentration de 10-11,1 mmoL·L−1 (180-200 mg·dL−1) dans le groupe conventionnel. La glycémie a été contrôlée à l’aide d’un algorithme dynamique de perfusion intraveineuse d’insuline. Nous avons comparé les groupes randomisés en matière de moyenne pondérée dans le temps, de la proportion de temps passé à la concentration cible, du nombre d’épisodes d’hypoglycémie grave (< 2,2 mmoL·L−1 ou < 40 mg·dL−1) ou modérée (< 2,8 mmoL·L−1 ou < 50 mg·dL−1), ainsi que de la variabilité chez un même patient des concentrations glycémiques, exprimée en tant qu’écart type de la moyenne du patient.
Cent quatre-vingt-sept patients ont été attribués au groupe contrôle glycémique intensif, et 177 au groupe contrôle glycémique conventionnel. La médiane (valeur du quartile inférieur [Q1], valeur du quartile supérieur [Q3]) de la moyenne pondérée dans le temps pour les groupes intensif vs conventionnel était de 6 (5,6, 6,7) mmoL·L−1vs 7,7 (6,9, 9,2), respectivement; P < 0,001. Le groupe intensif a passé 49 % (29, 71) du temps dans la zone cible, soit considérablement plus de temps que le groupe conventionnel n’a passé dans la zone cible intensive ou dans sa propre zone cible (tous deux P < 0,001). Le groupe intensif a présenté une variabilité chez un même patient légèrement plus basse que le groupe conventionnel (0,9 [0,7, 1,3] mmoL·L−1vs 1,3 [0,8, 1,8] mmoL·L−1, respectivement; P < 0,001). Trois patients ont présenté une hypoglycémie modérée (groupe intensif), mais aucun patient n’a manifesté d’épisode grave.
Un contrôle glycémique peropératoire rigoureux peut être maintenu en chirurgie non cardiaque sans épisode hypoglycémique grave. (Numéro de ClinicalTrials.gov, NCT00433251).
Validation of the DeLiT Trial intravenous insulin infusion algorithm for intraoperative glucose control in noncardiac surgery: a randomized controlled trial
A safe and effective insulin infusion algorithm that achieves rigorous intraoperative glycemic control in noncardiac surgery has yet to be formally characterized and evaluated. We therefore report the validation of the DeLit Trial insulin infusion algorithm.
Patients scheduled for major noncardiac surgery were randomized to a target intraoperative blood glucose concentration of 4.4-6.1 mmoL·L−1 (80-110 mg·dL−1) intensive group or 10-11.1 mmoL·L−1 (180-200 mg·dL−1) conventional group. Glucose was managed with a dynamic intravenous insulin infusion algorithm. We compared the randomized groups on glucose time-weighted average (TWA), proportion of time spent within target, number of severe (< 2.2 mmoL·L−1 or < 40 mg·dL−1) or moderate (< 2.8 mmoL·L−1or < 50 mg·dL−1) hypoglycemic episodes, and within-patient variability in glucose concentrations expressed as standard deviation from the patient mean.
One hundred eighty-seven patients were assigned to intensive glucose control, and 177 patients were assigned to conventional glucose control. Median (lower quartile value [Q1], upper quartile value [Q3]) of intraoperative TWA for the intensive vs conventional groups was 6 [5.6, 6.7] mmoL·L−1vs 7.7 [6.9, 9.2] mmoL·L−1, respectively; P < 0.001. The intensive group spent 49% (29, 71) of the time within target, substantially more time than the conventional group spent either within the intensive target or within its own target (both P < 0.001). The intensive group had slightly lower within-patient glucose variability than the conventional group (0.9 [0.7, 1.3] mmoL·L−1vs 1.3 [0.8, 1.8] mmoL·L−1, respectively; P < 0.001). Three patients had moderate hypoglycemia (intensive group), but none experienced severe episodes.
Tight intraoperative glucose control in noncardiac surgery can be maintained successfully without serious hypoglycemic episodes. (ClinicalTrials.gov number, NCT00433251).
Glucose control is challenging even in outpatients, and even more so in critical care patients.1,2 Several studies have found that it is difficult to maintain glucose within target ranges.3 A special challenge in intraoperative management is the need to reach the target glucose range quickly and then to maintain control through a relatively short but highly dynamic period.
A simple, safe, and effective insulin infusion algorithm that achieves rigorous intraoperative glycemic control has yet to be formally evaluated, much less fully characterized using various statistical approaches.4
One arm of the Dexamethasone, Light Anesthesia and Tight Glucose Control (DeLiT) Trial evaluates the effects of intensive vs conventional glucose control on serious complications in patients having major noncardiac surgery.5
How best to manage glucose in patients undergoing noncardiac surgery remains poorly characterized.6 Earlier studies suggested that intensive (tight) glycemic control, specifically a target of 4.4-6.1 mmoL.L−1 (80-110 mg·dL−1), improved outcomes in surgical and medical intensive care unit (ICU) patients.2,7 Subsequent trials could not confirm a similar benefit,8-10 and others reported increased mortality with the same intensive control targets.10 Consequently, tight glucose control became less popular.11 Moreover, concerns have arisen regarding the high incidence of hypoglycemia in association with intensive glucose control8-10,12 and the possibility that hypoglycemia may offset its potential benefits.11 Despite these reservations, however, tight glucose control may be beneficial for some populations11,13,14 and clearly requires additional study.11,15,16
In this sub-study, we reported on the validation of our insulin infusion algorithm in terms of time to achieve target, percent of time spent within desired range, incidence of severe and moderate hypoglycemia, and glucose variability.
We report sub-study results from the DeLiT Trial,5 a factorial randomized single-centre study designed to test the primary hypotheses that major perioperative morbidity is reduced by 1) low-dose dexamethasone; 2) intensive perioperative glucose control; and 3) lighter anesthesia.
Selection and description of participants
Patients ≥ 40 yr of age, American Society of Anesthesiologists (ASA) physical status II-IV, and scheduled for elective major open vascular, abdominal, or urologic surgery at Cleveland Clinic, Cleveland, Ohio, USA were enrolled after written informed consent was granted. Initially, the trial started with inclusion criteria limited to patients who were ≥ 50 yr of age and scheduled only for major open vascular surgery. However, because of an inadequate number of potential study candidates, we broadened the inclusion criteria as above. The study was approved by the Cleveland Clinic Institutional Review Board. The enrolment period extended from March 2007 through May 2010.
All patients received general anesthesia and endotracheal intubation with sevoflurane in air and oxygen mixture and intravenous fentanyl infusion following a standardized anesthetic protocol. As part of the underlying factorial design, patients were randomized to blood glucose concentrations of 4.4-6.1 mmoL·L−1 (80-110 mg·dL−1) intensive control or 10-11.1 mmoL·L−1 (180-200 mg·dL−1) conventional control. Patients were also randomized to receive either dexamethasone 8 mg or placebo immediately preoperatively and either light or deep anesthesia. Randomization codes were generated by the PLAN procedure in SAS® statistical software and implemented using a web-based system that was accessed by research physicians shortly before anticipated induction of anesthesia. Randomization was stratified according to the presence or absence of history of diabetes to ensure balance for each intervention comparison within diabetes status. Clinicians were blinded to the dexamethasone.
In the intensive control group, intravenous insulin therapy was initiated when blood glucose concentration exceeded 6.1 mmoL·L−1 (110 mg·dL−1) and then adjusted to maintain the concentration within the desired target of 4.4-6.1 mmoL·L−1 (80-110 mg·dL−1). In the conventional control group, intravenous insulin therapy was initiated when blood glucose concentration exceeded 11.9 mmoL·L−1 (215 mg·dL−1) and then adjusted to maintain concentration within the desired target of 10-11.1 mmoL·L−1 (180-200 mg·dL−1).2
Glucose control began shortly after induction of anesthesia using the protocols described below (Appendices 1 and 2) and continued through the first two hours in the postanesthesia care unit. These algorithms were based on those used in the Cleveland Clinic’s critical care units, but they were modified by the investigators to different target glucose concentrations and more aggressive insulin infusion rates, to include intravenous insulin boluses as well as frequent glucose measurements to accommodate the relatively short duration and dynamic nature of surgery. Intravenous insulin infusion and boluses were administered by experienced research physicians under supervision of the attending anesthesiologist. The investigators evaluated glucose concentrations at 30-60-min intervals in both groups depending on previous glucose values and clinical judgement. The research physicians’ intraoperative responsibilities were only glucose control and other aspects of the underlying DeLiT Trial.
Glucose concentrations were usually evaluated from arterial blood using the Accu-Chek® Inform system (Roche Diagnostics, Indianapolis, IN, USA). However, concentrations provided by the Clinical Laboratory were also considered when available. All blood perioperative glucose concentrations were recorded, as were baseline patient characteristics and the type of surgery.
We evaluated the effectiveness of our glucose-control protocol using outcomes previously described for assessing efficacy and safety of an insulin infusion protocol. These included intraoperative glucose time-weighted average (TWA), time taken to achieve desired level of glucose control, proportion of time spent within target range, the number of severe hypoglycemic episodes (< 2.2 mmoL·L−1 or < 40 mg·dL−1),9,10,17 the number of moderate hypoglycemic episodes (< 2.8 mmoL·L−1 or < 50 mg·dL−1),−1,18,19 the hyperglycemic index,17,20 and within-patient variability in glucose values expressed as standard deviation (SD) from the patient mean and the glucose lability index (GLI).21
Time-weighted average glucose concentration was calculated for each patient as area under the curve for the duration of the glucose measurements divided by the duration. The hyperglycemic index (HGI) was calculated as the area under the curve above the upper limit of the relevant range (i.e., 6.1 or 11.1 mmoL·L−1 [110 or 200 mg·dL−1]) divided by the duration of measurements. The HGI is a measure of the magnitude of hyperglycemia; it is independent of the number of measurements and duration of measurement and is less affected by hypoglycemic episodes than the overall TWA.
Within-patient variability in glucose measurements was calculated as a traditional SD of a patient’s intraoperative glucose values. However, this method ignores the sequence of the glucose measurements and the time distance between measurements. We therefore also calculated the GLI, commonly used to monitor glucose in patients with chronic diabetes, as the sum of squared consecutive differences per time unit. However, this method is limited in that it ignores the number of measurements and is on a quadratic scale which inflates differences. We therefore devised a modified GLI which divides the sum of squared consecutive differences by the number of readings and then takes the square root.
The randomized glucose control groups were compared on TWA, percent within range, and variability measures using either Student’s t tests or Wilcoxon rank-sum tests, as appropriate. Additionally, because half of the patients were given dexamethasone 8 mg according to our factorial randomization strategy, we assessed the interaction between the dexamethasone (vs placebo) and intensive glucose control (vs conventional) interventions on glucose variability in a multivariable regression model.
For patients who had plasma concentrations that exceeded the intervention target values for their assigned group, the time was calculated from the start of insulin to back-in-range. Kaplan-Meier time-to-event analysis was used to estimate the time to back-in-range curve, censoring patients at end of surgery who remained out-of-range. For the intensive group, Cox proportional hazards regression was used to assess the relationship between diabetic status (yes vs no) and time to back-in-range adjusting for age, sex, body mass index, ASA status, and TWA glucose.
With the available patients (n = 364 at last interim analysis for the DeLiT Trial), we had 90% power at the 0.05 level to detect differences in means as small as 0.34 SD between the randomized groups; for TWA intraoperative glucose, the detectable difference was approximately 0.2 mmoL·L−1 (9 mg·dL−1). Data are reported as median (quartiles). The significance level for each hypothesis was 0.05.
All statistical analyses were undertaken using SAS software version 9.2 (SAS Institute, Cary, NC, USA) or R software version 2.8.1 (The R Foundation for Statistical Computing, Vienna, Austria).
Comparing treatment groups on demographics and surgical variables
Intensive (n = 187)
Conventional (n = 177)
Body mass index (kg·m−2), mean (SD)
Age (yr), mean (SD)
Female n (%)
Diabetes n (%)
Non-insulin requiring diabetes
Insulin requiring diabetes
ASA physical status n (%)
Surgery Type n (%)
Abdominal Aortic Aneurysm
Duration of Surgery hr mean (SD)
Comparing treatment groups on glucose and variability outcomes
Intensive (n = 187)
Conventional (n = 177)
Preoperative glucose (mmoL·L−1) (mg·dL−1)
# of glucose measurements
8 [6, 10]
6 [5, 9]
Received Insulin n (%)
Within-patient glucose TWA (mmoL·L−1) (mg·dL−1)
6 [5.6, 6.7]
7.7 [6.9, 9.2]
108 [100, 121]
139 [124, 165]
% time glucose within 4.4-6.1 (mmoL·L−1) 80-110 (mg·dL−1)
49 [29, 71]
3 [0, 25]
% time glucose > 6.1 (mmoL·L−1) > 110 (mg·dL−1)
43 [23, 68]
96 [74, 100]
% time glucose within 10-11.1 (mmoL·L−1) 180-200 (mg·dL−1)
0 [0, 0]
0 [0, 13]
% time glucose > 11.1 (mmoL.L−1) > 200 (mg.dL−1)
0 [0, 0]
0 [0, 0.5]
% time in relative target range**
49 [29, 71]
0 [0, 13]
5.2 [1.4, 12.8]
0 [0, 0]
Within-patient glucose SD (mmoL·L−1) (mg·dL−1)
0.94 [0.72, 1.3]
1.3 [0.8, 1.8]
17 [13, 24]
23 [15, 32]
Glucose lability index (GLI)
64 [30, 119]
50 [22, 112]
Adjusted glucose lability index
3.1 [2.3, 4.0]
3.1 [2.1, 4.4]
On the basis of the traditional SD method, the intensive group had lower mean within-patient variability than the standard therapy group (P < 0.001) (Table 2). However, the mean GLI was greater for the intensive group (P = 0.049). Our modified version of the GLI showed no difference between groups (P = 0.87).
No patient in either group experienced a severe hypoglycemic event (glucose < 2.2 mmoL·L−1 (40 mg·dL−1). Thus, the estimated incidence (95% confidence interval) of severe hypoglycemia is (0.2)% for each group. However, each of three patients (intensive group) experienced a single episode of glucose between 2.2 and 2.8 mmoL·L−1 (40 and 50 mg·dL−1), with values 2.56, 2.61, and 2.72 mmoL·L−1 (46, 47, and 49 mg·dL−1) and were treated immediately with administration of intravenous 50% dextrose 12.5 mL per protocol. The incidence of moderate or severe hypoglycemia (glucose < 2.8 mmoL·L−1 [50 mg·dL−1]) (3/187 vs 0/177, respectively) did not differ between the groups (Fisher’s exact test, P = 0.25).
The quality of an insulin infusion protocol is determined by at least three factors: 1) the incidence of hypoglycemia; 2) blood glucose variability; and 3) the fraction of time spent within the designated target range.17 We were successful on all three counts. There were no episodes of serious hypoglycemia, defined as < 2.2 mmoL·L−1 (40 mg·dL−1). Our low incidence is consistent with the findings of Gandhi et al. who reported a rate of only 0.5% during cardiac surgery,4 and it is also consistent with the findings of Vogelzang et al. who reported an incidence of 0.9% in critical care patients.20 In contrast, hypoglycemia rates as high as 29% have been reported in other circumstances.8-10,12
Patients assigned to intensive control had less glucose variability than those receiving conventional control. Furthermore, the median glucose concentration in the intensive group was 6 mmoL·L−1 (108 mg·dL−1), which is within our intended target and distinctly less than the 7.7 mmoL.L−1 (139 mg·dL−1) median in the conventional control patients. Patients assigned to intensive control were within the target range of 4.4-6.1 mmoL.L−1 (80-110 mg·dL−1) about half the time vs only 3% of the time in the conventional group. The median time to reach target within the intensive control group was 49 min (although some patients required two or even three hours to reach the target).
Thus, our glucose control approach was generally successful and fared well compared with results reported under other circumstances. For example, studies in cardiac surgery and critical care patients report times-to-target ranging from one to 15 hr and times-within-target-range of 27-67%.3,20,22-27 Furthermore, a comparison of paper-based vs computer-guided insulin infusion algorithms in cardiac surgery patients with a target glucose concentration of 5-8.3 mmoL·L−1 (90-150 mg·dL−1) reported mean times-to-range of 1(2) hr and 1(1) hr, respectively, and percent times-in-range of 27% and 49%, respectively.22In contrast, one study reported taking 9 to 12 hr to control glucose concentrations; furthermore, the percentage of time within target ranges from 48% to 63%.23 Others targeting glucose concentrations 4-7.5 mmoL·L−1 or 4.4-8.3 mmoL·L−1 (72-135 mg·dL−1 or 80-150 mg·dL−1, respectively) took two to 15 hr to achieve target, and the time spent within the target range varied from 52% to 67%.3,20,26,27
Relatively good control in our intensive patients might be attributed to our aggressive algorithm, which incorporated both insulin boluses and a continuous infusion, and to the fact that bolus size and infusion rate varied with the initial glucose concentration. However, it is probably at least equally important that insulin administration was managed by dedicated investigators whose major intraoperative responsibility was glucose control.
The relatively frequent glucose concentration determinations (every 30-60 min), the dynamic nature of the insulin infusion algorithm, and the vigilance of the investigators and clinicians presumably contributed to our low incidence of hypoglycemia. Historically, aggressive insulin protocols in cardiac surgery and critical care patients have been associated with considerably higher rates of hypoglycemia.25,26 Indeed, tight glucose control goals in critical care units have resulted in unacceptably high rates of hypoglycemia9,12 and increased mortality.8 Anesthesiologists are especially concerned about hypoglycemia because symptoms are masked by general anesthesia.28
The effective glycemic control achieved by our protocol was associated with low variability as evidenced by three distinct methods of characterizing glucose concentration variability: 1) SD29; 2) GLI; and 3) our novel adjusted GLI which adjusts for the number of measurements. Low glucose concentration variability might be attributed to the dynamic properties of our insulin infusion protocol, which considers not only absolute glucose concentration but also changes from previous readings. Identical glucose concentrations can thus trigger various responses depending on previous glucose concentrations. In fact, our intraoperative insulin infusion protocol fulfilled the two requirements suggested by Preiser et al.,17 namely, the insulin infusion algorithm should be dynamic, and it should specify the timing of subsequent glucose measurements.
Glycemic variability is an independent risk factor for mortality in the ICU.29,30 Glycemic variability, as characterized by GLI, is also independently associated with hospital mortality in septic patients.31 Interestingly, this relationship is even stronger in the euglycemic range. As a result, investigators have suggested that a glycemic variability metric should be considered in developing and evaluating protocols for glycemic control,30,31 because glucose variability and mean glucose concentrations contribute to outcome independently.31 Vogelzang et al. have proposed using the HGI to assess hyperglycemia. They observed that HGI predicts mortality better than other indices of blood glucose control that do not take into account the duration of hyperglycemia.20 We therefore characterize our results in terms of both HGI and GLI, along with our new adjusted GLI.
Our intensive control patients demonstrated less variability than the conventional control group, which is consistent with the findings of Egi et al. in critical care patients.29 Fluctuating glucose concentrations are associated with elevated concentrations of 8-iso-prostaglandin F2 alpha, a marker of oxidative stress that may mediate organ dysfunction32 and contribute to neurologic pathology.29
More patients in the intensive group than in the conventional group were given insulin, which was unsurprising considering the low threshold we targeted. Nonetheless, TWA concentrations differed by only about 1.7 mmoL·L−1 (30 mg·dL−1) even though the target ranges were 4.4-6.1 mmoL·L−1 and 10-11.1 mmoL·L−1 (80-110 mg·dL−1 and 180-200 mg·dL−1), respectively. The reason for a smaller difference between the two glucose control groups compared with the originally intended targets is that many of our patients in the conventional group, whether diabetic or not, never reached glucose concentrations sufficient to trigger treatment. These findings are consistent with Gandhi et al. who used almost the same targets in their study of cardiac surgery patients.4
Unlike other aspects of our study, diabetic status could not be assigned randomly. Therefore, comparisons between diabetic and nondiabetic patients were purely observational. Nonetheless, diabetic patients did not differ from nondiabetic patients with respect to how soon their glucose levels returned to within-range after initiation of insulin infusion; this is despite the fact that the same infusion protocol was used for both groups. This finding contradicts the belief of many clinicians that prior exposure to insulin reduces patients’ sensitivity to additional insulin treatment.33 Moreover, the intensive and conventional groups were well-balanced on diabetic status as well as type of diabetes; therefore, diabetic status did not confound comparison of the two randomized groups on outcome.
Half of our patients were given dexamethasone 8 mg. This dose is similar to the 4-8 mg of the drug that is commonly given prophylactically to prevent postoperative nausea and vomiting and for a variety of other reasons.34-36 Normally, dexamethasone administration increases plasma glucose concentrations. However, dexamethasone was given as part of a factorial randomization of the underlying DeLiT Trial and was equally balanced in the conventional and intensive control groups. Consequently, comparisons between the two treatment groups remain valid. Furthermore, glucose control was similar in patients who received and did not receive steroids, indicating that our insulin protocol was able to compensate dynamically for the normal hyperglycemic effect of steroids.
The blood glucose concentrations we targeted were those used most commonly in the ICU2 and cardiac surgery patients4 when the study was started in March 2007. Other target concentrations have been used since then,10 and the ideal perioperative target remains unknown. Nevertheless, there is every reason to believe that our general approach to glucose control will work comparably well at different target concentrations.
In summary, it remains to be determined whether tight intraoperative glucose control improves patient outcomes in noncardiac surgery. In the meantime, however, our results indicate that tight intraoperative glucose control can be maintained successfully without excessive hypoglycemic episodes during major noncardiac surgery.
The DeLiT Trial is supported by Aspect Medical (Newton, MA) and the Cleveland Clinic Research Project Committee. Aspect Medical was purchased recently by Covidien (Dublin, Ireland). None of the authors has a personal financial interest in this research.