Software-guided versus nurse-directed blood glucose control in critically ill patients: the LOGIC-2 multicenter randomized controlled clinical trial

Elevated blood glucose levels are very common in critically ill patients, independent of pre-existing diabetes mellitus. This hyperglycemia has been associated with an increased risk of morbidity and mortality [1]. Tight blood glucose control (BGC), targeting blood glucose levels below 110 mg/dL, improved the outcome of critically ill patients only in well-controlled single-center trials and in implementation studies [2‐6]. In large multicenter trials, however, this beneficial effect was not reproduced and in the NICE-SUGAR trial mortality even increased in the intensive BGC group [7‐9]. While excessive hyperglycemia above 180 mg/dL is no longer accepted, the target range for BGC is still controversial and variable [10]. The American Diabetes Association recommends targeting blood glucose levels below 180 mg/dL, while in Europe most centers use stricter target ranges (below 145 mg/dL), despite a lack of evidence [11, 12].

Excess mortality in patients undergoing BGC has been attributed to the increased incidence of hypoglycemia and unnecessary blood glucose variability [13, 14]. Inaccurate blood glucose measurements and the use of insulin dosing protocols that have not been clinically validated play an important role herein [15]. BGC requires not only frequent measurements of blood glucose, but also difficult calculations of the insulin doses, depending on the protocol. Moreover, BGC has to be done in severely ill patients, who may undergo rapid changes in their insulin sensitivity due to the underlying illness and medication such as parenteral nutrition and steroids. To obtain better (less hyperglycemia) and safer (less hypoglycemia) BGC it has been advocated to use computerized protocols for insulin dosing and timing of blood glucose measurements [16, 17]. However, the current evidence for these computer algorithms consists primarily of implementation studies or small randomized controlled studies [18‐22]. Therefore, their use has not yet been widely adopted in general clinical practice although a growing number of ICUs are starting to use validated computer algorithms with a beneficial effect on blood glucose control, such as EndoTool and STAR [23, 24].

The LOGIC-Insulin algorithm, developed and based on the Leuven guideline and practice, had already been clinically validated in a single-center randomized trial, showing improved and safer BGC (below 110 mg/dL), compared to that performed by nurses who were highly experienced in BGC [25]. However, performance of the algorithm in a setting of different blood glucose target ranges, in a wide-ranging patient population, by nurses with less experience in BGC has not yet been evaluated in a pragmatic, multicenter setting. Therefore, we hypothesized that the LOGIC-Insulin algorithm improves the quality of BGC, in comparison with nurse-directed BGC, using two different glucose target ranges (80–110 mg/dL or 90–145 mg/dL) in a heterogeneous population of critically ill adults and in the setting of a nursing staff with variable levels of expertise.

Using the LOGIC-Insulin algorithm improved the quality of BGC reflected by a reduction in the GPI, an increase in time-in-target range and a reduction in blood glucose variability, without increasing the incidence of hypoglycemia.

The beneficial effects of software-guided BGC were independent of the chosen target blood glucose range and of the center’s expertise in BGC. Morbidity and mortality did not differ between patients in whom BGC was done either by expert nurses or aided by the LOGIC-Insulin algorithm in the context of similar blood glucose target ranges.

These data confirm the findings from our single center trial, in which LOGIC-Insulin improved tight BGC and lowered the incidence of mild hypoglycemia [25]. However, the treatment effect in the pragmatic multicenter LOGIC-2 trial was twofold that in the LOGIC-1 trial and that for which it was statistically powered. This was explained by more pronounced benefits of algorithm-guided BGC in centers with looser BGC. The performance of the Leuven nurses in BGC, incorporated in the LOGIC-Insulin algorithm, may also have blunted the effect in Leuven. Similar blood glucose profiles in the software groups in Leuven and Hasselt, using the same target range, demonstrate that the LOGIC-Insulin software can be generalized outside the context of centers with extensive BGC experience. The fact that more than 98% of all insulin dosing advice provided by the LOGIC-Insulin algorithm were followed by the nurses in all centers in the algorithm-guided BGC group indicates high protocol compliance and indicates that the results are protocol-induced and not a representation of the nurses’ BGC performance in the intervention group. This further underpins the generalizability of the study findings and the potential for implementation of the LOGIC-Insulin software, independent of the chosen blood glucose target range. This is important as the interaction between the BGC algorithm and patient glucose dynamics may be influenced by the performance of the bedsides nurses, who are the ultimate “controllers” in open-loop BGC software.

The larger treatment effect in the broader target range of 90–145 mg/dL (delta GPI 15.1) than in the narrow target range of 80–110 mg/dL (delta GPI 5.7) was therefore not anticipated. One might have expected that BGC is easier for nurses when the target is broader. This indicates that nurses, in the context of a broader target range, tend to be more lenient towards the upper target limit. The higher incidence of extended hyperglycemia above 180 mg/dL reflects this, in spite of the consensus in all guidelines on BGC in the ICU that excessive hyperglycemia should be avoided. Nevertheless, the higher incidence of overrules in Amsterdam, possible practical differences in the insulin administration, and greater statistical uncertainty due to the smaller patient numbers may have contributed as well. Most likely, BGC will be even looser outside study settings. In contrast, use of the narrower target range results in greater incidence of mild hypoglycemia. Algorithm-guided BGC did not increase the incidence of potentially harmful, severe hypoglycemia, which should always be strictly avoided. A recent randomized controlled trial used computerized insulin-dosing algorithms in the intensive BGC group, but did not show a clinical difference with more conservative, nurse-directed BGC [31].

Algorithm-guided BGC increased the workload, similar to other algorithms [32]. Ten instead of eight blood glucose measurements per day were needed to improve BGC in one patient by software use. The workload of blood glucose measurements may be offset by lowering of the cognitive burden for the nurses in deciding on the right insulin dose [33]. Hence, this slight increase in workload may be inevitable in obtaining safe and effective blood glucose control, especially in the initial phase, aiming for a short time to reach the chosen target range combined with a high time-in-target range, low glucose variability, and avoidance of hypoglycemia, and this in a setting with rapid changes in patients insulin sensitivity and in the presence of external confounding factors such as use of steroids and parenteral nutrition. There is now a consensus that a minimal measurement frequency is needed to provide safe glycemic control without adverse effects on the outcome [28]. Whether the increased workload resulted in more attention paid to the patient (collateral benefit) or less attention to the patient (added risk) cannot be delineated from this study.

The improvement in BGC provided by the software was comparable in all patient populations, such as cardiac surgery, pre-existing diabetes, sepsis, and medical admissions. However, the clinical effects of improved BGC should be the ultimate determinants to evaluate medical interventions. Although the LOGIC-2 trial was not set up to test differences in clinical outcomes, it was monitored. As expected, no clinical differences were seen in the overall patient population. Patients with diabetes mellitus did not benefit more from algorithm-guided BGC, as was seen in a recent trial of BGC in patients after coronary artery bypass graft surgery [34]. However, the level of pre-admission control of the diabetes mellitus may be the most important factor in determining whether patients benefit from tight BGC.

In patients with infectious problems on admission, improved BGC resulted in lowering of the incidence of new infections. This may be explained by the fact that patients with infectious problems on admission are more severely ill, have greater risk of more severe hyperglycemia, which is more difficult to control, and have greater risk of new infections.

The LOGIC-2 trial has its limitations though. Blood glucose dynamics in the ICU are determined by patient characteristics, by the BGC protocol, and by the experience of the bedside nurses in executing the BGC. Therefore, the differences in the quality of BGC in the Nurse-C group between the centers are most likely multifactorial. The variability in the “baseline” quality of BGC may have affected the performance using the LOGIC-Insulin software. It can also not be excluded that nursing practice in performing BGC changed over time during the study, under the influence of the algorithm. However, participating in a study on BGC had most likely improved the performance of BGC by the nurses over time due to a training effect. The LOGIC-Insulin software was now only tested when using glucose readings from arterial blood measured in accurate blood gas analyzers. To gain further applicability the software should be able to cope with blood glucose measurements from point-of-care, handheld blood glucose meters. Other computerized protocols however, have already used these point-of-care glucometers showing a safe and efficient level of blood glucose control [23, 24]. Efficiency or cost-effectiveness of the LOGIC-Insulin software is still unproven, despite that the present pragmatic multicenter RCT showed that the software works in real-life circumstances.

In a randomized controlled trial of a mixed critically ill patient population, the use of the LOGIC-Insulin blood glucose control algorithm, compared with blood glucose control by expert nurses, improved the quality of blood glucose control without increasing hypoglycemia. This is demonstrated by an important improvement in all domains of glycemic control: a decrease in GPI, an increase in the time-in-target range, and a decrease in glucose variability. Moreover, the LOGIC-Insulin algorithm was shown to be efficient and safe when used in three different ICUs with local variations in clinical practice, with different blood glucose target ranges, and with different levels of experience in blood glucose control. The slight increase in workload generated by the LOGIC-insulin algorithm should be considered an inevitable trade-off when providing safe and efficient blood glucose control.