Introduction
Time in targeted blood glucose range (TIR) may be a suitable descriptor of the efficacy and safety of glycemic control and could be considered as a marker of the severity of dysglycemia and an index of the quality of care. Neither the study that ushered in the era of “tight glycemic control” nearly 14 years ago [
1] nor the trial that dampened enthusiasm for intensive control of blood glucose (BG) values 8 years later [
2] reported TIR.
The Glucontrol study was the only adult randomized controlled trial (RCT) of intensive insulin therapy (IIT) that reported TIR [
3]. Only 27.8% of the values obtained in patients in the experimental arm were within the targeted BG range of 80 to 110 mg/dl. Subsequent analysis of these data demonstrated that, for patients in either the intensively treated or moderate arm, with a BG target of 140 to 180 mg/dl, TIR >50% was independently associated with an increased rate of survival [
4]. Chase and coinvestigators have published a series of studies that assessed the association of TIR (referred to as
cumulative time in band with organ failure mortality in a 784-patient, before-and-after, single-center cohort evaluation of their Specialized Relative Insulin Nutrition Tables (SPRINT) protocol for IIT [
5-
7]. They concluded that TIR ≥50% was independently associated with less organ failure, as quantified by reduction in Sequential Organ Failure Assessment (SOFA) score [
6], and that TIR ≥70% was independently associated with increased survival compared with lower thresholds of TIR (≥30% and ≥ 50%) [
7]. Recently, Okabayashi and colleagues published a single-center RCT demonstrating reduction in surgical site infection with intensive vs. moderate BG targets, notable for the very high TIR achieved in the two groups with use of a closed-loop BG monitoring and insulin treatment system [
8]. In contrast, another recent multicenter study in which computerized glucose control was used failed to show any clinical benefit when TIR was low [
9]. These data raise the possibility that low TIR may have confounded the results of the major RCT of IIT and may explain their uneven outcomes [
1-
3,
10,
11].
A robust literature has demonstrated that hyperglycemia, hypoglycemia and increased glucose variability are independently associated with mortality in diverse cohorts of critically ill patients [
12-
22] and that diabetic status modulates these relationships [
23-
25]. The outcome of patients with diabetes mellitus (DM) may be less influenced by dysglycemia than it is among patients without diabetes (non-diabetic (NON)). Nevertheless, there is no firm consensus about how to manage glycemia in the critically ill, and some current guidelines have promoted BG targets in the hyperglycemic range to mitigate the occurrence of hypoglycemia [
26-
28].
We hypothesize that a high TIR is the key element of glycemic control needed to effect optimal outcome and may, in fact, blunt the deleterious impact of transient excursions into the hypoglycemic and hyperglycemic ranges. Accordingly, we evaluated the impact of a very high TIR, such as >80%, a level not generally evaluated in previous investigations, in a diverse population of critically ill patients. In addition, we assessed the interrelationship of TIR and diabetic status. To test these hypotheses, we performed a retrospective analysis of a large cohort of patients at a single mixed medical-surgical intensive care unit (ICU).
Discussion
In this retrospective analysis of a large, heterogeneous cohort of critically ill patients, we investigated the relationship between TIR 70 to 140 mg/dl and mortality. The salient finding of the study is that TIR was strongly associated with survival among patients without diabetes, with a 61% increase in the odds of death when we compared patients below and above the median value for TIR. The association between TIR and mortality was independent of severity of illness, with the greatest difference demonstrated in the subpopulation with the highest severity of illness. In addition, the association between TIR and mortality was independent of ICU LOS. Notably, TIR-hi was strongly associated with survival among patients with ICU LOS of 1 to 3 days; earlier data derived from interventional trials suggested no benefit of glycemic control in a short-stay cohort [
1,
10].
In contrast, among patients with DM in the present study, there was no consistent relationship between TIR and mortality. Glycemic control differed substantially among patients with vs. without DM. Although BG targets and glycemic control guidelines in the ICU did not differ based on diabetic status, patients with DM had higher mean BG, higher glucose variability, higher rates of hypoglycemia and a much lower TIR than did those without DM. These divergent findings are consistent with recent literature describing differences between patients with vs. without diabetes regarding the relationships of hyperglycemia, hypoglycemia and glucose variability to mortality [
23-
25].
These data expand on the limited literature that has explored the relationship between TIR and mortality in critically ill patients. Chase and colleagues evaluated the clinical impact of a computerized glycemic control protocol requiring BG measurement every 1 to 2 hours that controlled insulin infusion and nutrition input concomitantly [
5]. Using a before-and-after experimental design, they demonstrated reduced mortality in the interventional group among those staying in the ICU for 3 to 5 days [
6]. Subsequent analysis included the entire cohort of 784 patients to investigate the relationship between TIR 72 to 126 mg/dl and clinical endpoints [
6,
7]. TIR ≥50% was independently associated with reduced organ failure, as shown by fewer patients with SOFA scores ≤5 after day 5 in the ICU [
6]. In addition, TIR ≥70% was independently associated with increased odds of survival compared with TIR ≥50% or TIR ≥30% [
7]. These investigations did not stratify patients by diabetic status (17% of the entire cohort).
Our findings shed light on the varied results of the interventional trials of IIT. Intensive monitoring of BG levels and treatment of hyperglycemia became a worldwide standard of care following publication of a single-center study conducted in a population of mechanically ventilated surgical patients that demonstrated marked reductions in mortality and morbidity in patients treated with intravenous insulin and a BG target of 80 to 110 mg/dl [
1]. Subsequent interventional trials failed to reproduce these impressive findings [
2,
3,
10,
11]. Only one of these studies explicitly reported TIR [
3]; in fact, the Data Safety Monitoring Board for that study forced the premature termination of the study because of the low TIR. Estimates of TIR in the intensive insulin arm for the major interventional trials, inferred using the reported morning BG values and the standard normal distribution table, range from 31% [
2] to 53% [
1]. However, the comparison of TIR of these studies is confounded by the differences in the number of BG values used and in BG monitoring, including the use of arterial blood and blood gas analyzers [
1] or combinations of arterial blood, capillary blood, blood gas analyzers and bedside glucometers [
2,
3,
10,
11]. None of these studies reported BG monitoring frequency.
In contrast to these findings, Okabayashi and coworkers performed a single-center RCT in 447 patients undergoing hepatobiliary or pancreatic surgery to compare infection rates in those with BG targets of 80 to 110 mg/dl and 140 to 180 mg/dl, the same targets used in the Normoglycemia in Intensive Care Evaluation and Surviving Using Glucose Algorithm Regulation trial (referred to as NICE-SUGAR) [
8]. They used a computerized, closed-loop BG monitoring and insulin delivery system to achieve very high TIR for the two study groups—85.8% and 96.8%, respectively—without significant differences between patients with and without DM. Notably, surgical site infection, the primary outcome, occurred in 9.8% of patients treated with the intermediate BG target and 4.1% of those treated with the “tight” BG target (
P =0.028). This intriguing investigation is, in fact, hypothesis-generating. Although a high rate of hypoglycemia in the intensively treated patients of the major interventional trials of IIT has been independently associated with increased risk of death in all of the studies [
1-
3,
10,
11,
14,
15], and though elevated glucose variability has been independently associated with increased risk of death in two of them [
14], it is quite possible that the low TIR contributed to the finding of lack of benefit of the intervention. This conclusion is supported by a recently published
post hoc analysis of the multicenter Glucocontrol trial of IIT [
4]. The Glucocontrol trial was terminated prematurely because of the low rate of achieving the targeted BG for the patients in the interventional arm. The new analysis concludes that, though there was no difference in the development of organ failure based on intention to treat to different glycemic targets (BG 80–110 mg/dl in the intensive arm and 140 to 180 mg/dl in the conventional arm), achieving the BG range of 72 to 136 mg/dl was independently associated with increased survival for patients in either arm of the trial.
The present study has several strengths. The large dataset is robust, containing a comprehensive set of clinical parameters. The patients were treated consistently over the period of time they were included in the study, with a high frequency of BG measurements exceeding that commonly seen in clinical practice [
22]. The analysis includes an explicit detailing of the relationship of TIR to survival, as well as its interaction with the other domains of glycemic control: hyperglycemia, hypoglycemia and glucose variability. Finally, the investigation stratifies patients by diabetic status, a factor increasingly recognized in the literature regarding glycemic control in critically ill patients [
23-
25].
There are several limitations of the present study. First, the major limitation is its observational nature. Consequently, its conclusions must be considered hypothesis-generating rather than as proof of causality. Nevertheless, it is unlikely that a randomized trial will ever be conducted that stratifies patients into a TIR-lo cohort. Second, we conducted a single-center study, raising questions about its generalizability. However, the ICU treats a heterogeneous population of critically ill adults admitted with a wide array of medical and surgical diagnoses. Third, there is significant mathematical coupling between TIR and the other established domains of glycemic control hyperglycemia, hypoglycemia and glucose variability. A patient with low TIR is certainly more likely to have higher rates of dysglycemia reflected by the other three glucose metrics. Nevertheless, the data in Table
3 demonstrate, especially for non-diabetics, that concomitant hypoglycemia, hyperglycemia or increased glucose variability, regardless of TIR, was associated with increased risk of death. Fourth, we do not have hemoglobin A1c values for the patients. Some patients characterized as non-diabetic may, in fact, have more properly been characterized as previously undiagnosed diabetics. Fifth, we do not provide data about nutritional support or insulin therapy; subsequent work may help delineate the interplay between these factors and glycemic control. Finally, the ICU staff employed glucometers for BG analysis, using capillary blood in the majority of tests and venous or arterial blood in the remaining. Although this method of analysis has been associated with analytic inaccuracies [
31-
33], it remains the standard of care in many ICUs, especially in the United States.
These findings have implications relating to current clinical practice, to the assessment of quality of care and for the design of future interventional trials using new methods to improve glucose control. We have shown that TIR 70 to 140 mg/dl is independently associated with survival in a heterogeneous cohort of critically ill patients without diabetes. Nevertheless, one of the lessons of the interventional trials of glycemic control is that high TIR is an elusive target that is very difficult to achieve. Computer-guided insulin administration algorithms may be one way to increase TIR and consequently improve glycemic control. The Leuven investigators have demonstrated that use of a computer-guided algorithm increased TIR from 60.1% to 68.8% compared with the BG control efforts of their well-trained nurses [
34]. Similarly, Juneja and coworkers used a software program to guide insulin therapy targeting 80 to 110 mg/dl in 4,488 ICU patients, achieving TIR of 73.4% [
35]. Nevertheless, 4.2% of patients sustained at least one episode of severe hypoglycemia (BG <40 mg/day). The investigators concluded that measurement delay (mean of only 12 minutes, with a protocol requiring hourly monitoring) was contributory in two-thirds of these episodes.
Importantly, even hourly BG monitoring is associated with a high likelihood of missed excursions into the hyperglycemic and hypoglycemic ranges. This has been clearly demonstrated in studies in which researchers investigated continuous or near-continuous BG monitors [
36,
37], suggesting that increased BG monitoring frequency may provide the best path to achieving high TIR. Indeed, in a recently published Monte Carlo simulation of the impact of monitoring frequency on the occurrence of hyperglycemia and hypoglycemia, investigators determined that rates of dysglycemia decreased substantially when monitoring frequency was increased from every 2 hours to every 1 hour, with further modest improvement resulting from an increase in frequency to every 15 minutes [
38].
These data suggest the possibility that TIR may be a useful quality-of-care metric for assessing glucose control practice in an ICU in view of its ease of measurement and strong association with risk of death. Finally, we believe that the findings of the present study have important implications for the design of future interventional trials of IIT in critically ill patients. First, our data confirm that the optimal target for BG control in critically ill patients with diabetes has not yet been established, a finding that is consistent with recent observational literature [
23-
25]. Clinical equipoise may now exist for the implementation of an interventional trial incorporating multiple BG targets, based on patient characteristics, including diabetic status and, for patients with diabetes, preexisting glycemic control [
39,
40] Most importantly, achieving a high TIR, such as >80%, must be considered as integral to the success of the trial. Notably, the number of patients needed for a trial will likely be much lower than the numbers participating in the previously published interventional trials if TIR is very high, as demonstrated by a recently published study using a closed-loop insulin delivery system that in which researchers found a reduction in surgical site infection when they compared intensive and moderate BG targets, with just over 200 patients in each arm of the study [
8]. These findings cannot be achieved with the monitoring technologies and the monitoring frequencies used in the published interventional trials of IIT [
41]. In future trials, investigators must consider use of the new emerging technologies that allow continuous or near-continuous BG monitoring.
Competing interests
JSK has received consultant fees from Medtronic, Edwards Lifesciences, Roche Diagnostics, OptiScan Biomedical and Alere and research support from OptiScan Biomedical. J-CP has received consultant fees from Medtronic, Edwards Lifesciences, Maquet and OptiScan Biomedical.
Authors’ contributions
JSK had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. JSK was responsible for the study concept and design, acquisition of the data, drafting of the manuscript and statistical analysis. JSK and J-CP were responsible for analysis and interpretation of the data and critical revision of the manuscript for important intellectual content. Both authors read and approved the final manuscript.