Individualising blood glucose control in ICU patients: evidence and practical implementations Authors: Jan Gunst, James S. Krinsley, Michael J. Lanspa, Jean-Charles Preiser, Mervyn Singer, Greet Van den Berghe Published: 2026 ABSTRACT Many intensive care unit patients develop hyperglycaemia, and this is associated with an increased risk of adverse outcomes. Landmark randomised controlled trials (RCTs) suggest the impact of lowering blood glucose to the healthy, age-adjusted fasting range, hereafter called tight glucose control, depends on the accuracy of the glucose control protocol and the nutritional strategy. In the context of early parenteral nutrition, a condition that induces severe, iatrogenic hyperglycaemia, tight glucose control significantly reduced morbidity and mortality compared to tolerating severe hyperglycaemia. In the absence of early parenteral nutrition, hyperglycaemia was less severe; tight glucose control, applied with a computerised protocol that avoided severe hypoglycaemia, did not impact mortality nor the duration of ICU dependency as primary endpoint in a large multicentre RCT, but was associated with a less pronounced benefit on selected morbidity outcomes. Conversely, tight glucose control, using a protocol that considerably increased the risk of severe hypoglycaemic events, concomitantly increased mortality; nevertheless, the cause of excess mortality in this trial remains unproven and other mechanisms may have contributed as well. Altogether, evidence suggests that severe and iatrogenic hyperglycaemia and hypoglycaemia should both be avoided. From a pathophysiologic point of view, tight glucose control may be superior in selected settings if it can be applied with a validated protocol that avoids severe hypoglycaemia. In the absence of such a protocol, the increased risk of severe hypoglycaemia may offset any benefit of preventing hyperglycaemia, and tolerating intermediate blood glucose targets is likely the default setting. In the paper, we suggest specific blood glucose targets that differ according to the clinical context. Future research should aim at strategies that further optimise the quality of blood glucose control and study whether these improve patient-centred outcomes, and investigate whether specific patient populations benefit from individualised targets. The stress response to severe illness, major trauma and surgery induces insulin resistance and hyperglycaemia. Blood glucose concentrations in intensive care unit (ICU) patients show a U-shaped association with mortality, with the lowest mortality risk associated with concentrations in the age-adjusted fasting range [1]. As such associations are confounded by illness severity, proof of causality requires well-designed, adequately powered, randomised controlled trials (RCTs). In the last 25 years, several large RCTs have considerably changed our insights into the optimal blood glucose management strategy for ICU patients, and suggest that a one-size-fits-all approach is not valid [2–14]. In this narrative expert review, we will summarise insights derived from landmark RCTs and mechanistic studies, implications for clinical practice, and areas of future research, focusing on adapting blood glucose targets according to the individual patient context. RCTs revealed context-dependent effects of tight glucose control on outcome Landmark RCTs indicate the preferred blood glucose target is context-dependent (Table 1). In three RCTs (N = 3448) performed in critically ill patients in Leuven, Belgium, lowering blood glucose concentrations to the healthy, age-adjusted fasting range (50–80 mg/dL [2.8–4.4 mmol/L] for infants, 70–100 mg/dL [3.9–5.6 mmol/L] for children older than one year, and 80–110 mg/dL [4.4–6.1 mmol/L] for adults), hereafter labeled tight glucose control, significantly reduced short- and long-term morbidity and mortality as compared with tolerating hyperglycaemia up to the renal threshold (215 mg/dL [11.9 mmol/L]) [2–4]. The blood glucose targets in the intervention group were lower in infants and children than in adults, as healthy infants and children have lower normal fasting blood glucose concentrations than healthy adults [4]. Mechanistic studies associate the benefit of the intervention to prevention of glucose toxicity in vital cells with insulin-independent glucose uptake, and not to glucose-independent effects of insulin [16–19]. Patients included in the Leuven RCTs however received early full nutritional support as part of standard care at that time, which included early parenteral nutrition (PN) [2–4]. This nutritional strategy, which aimed to avoid any fasting, likely increases the risk of severe hyperglycaemia (>215 mg/dL [11.9 mmol/L]). Subsequent large RCTs demonstrated dose-dependent harm with early nutritional support, including harm from early PN [20–22]. These findings led to recommendations for more restrictive nutritional support in the acute phase of critical illness [23–25].Table 1Landmark RCTs on tight glucose control in ICULeuven RCTsNICE-SUGARTGC-FastPatient populationAdult, surgical ICUAdult, medical ICUPaediatric ICUAdult, mixed ICUAdult, mixed ICUN3448 (single centre)6104 (multicentre)9230 (multicentre)BG target Control group<215 mg/dL<180 mg/dL<215 mg/dL Intervention groupAge-adjusted(80–110 mg/dL for adults)80–110 mg/dL80–110 mg/dLBG measurement Measurement devicePredominantly blood gas analyserNot standardised. Inaccurate glucometers allowedPredominantly blood gas analyser Measurement sitePredominantly arterialNot standardised. Arterial, venous, capillaryPredominantly arterialInsulin protocol Administration modeContinuous IVContinuous + bolusContinuous IV Titration protocolTrained ICU nursesUnvalidated if-then protocolLOGIC-Insulin algorithmFeeding strategyEarly PNNo early PNNo early PNGlucose metrics Achieved difference in BG1±50 mg/dL in adults (105 vs. 152 mg/dL)±30 mg/dL(115 vs. 144 mg/dL)±30 mg/dL(107 vs. 140 mg/dL) Severe hypoglycaemiaIncreased 6-fold in adultsIncreased 14-foldUnaffectedOutcome Morbidity2DecreasedUnaffectedDecreased for selected outcomes Mortality3DecreasedIncreasedUnaffectedTo convert blood glucose in mg/dL to mmol/L, multiply by 0.055511 The reported achieved blood glucose concentrations are the mean blood glucose concentrations in ICU for the adult Leuven RCTs [15], the mean time-weighted blood glucose concentration in ICU for the NICE-SUGAR RCT, and the median morning blood glucose concentration in ICU for the TGC-Fast RCT2 The morbidity benefit in the adult Leuven RCTs was observed on secondary endpoints that included, among others, acute kidney injury, need for renal replacement therapy, cholestatic liver dysfunction in ICU, and duration of ICU and hospital dependency. The morbidity benefit in the paediatric Leuven RCT was observed on the primary endpoint (duration of ICU dependency and inflammation). The morbidity benefit in TGC-Fast was observed on selected secondary endpoints (severe acute kidney injury, need for renal replacement therapy, and cholestatic liver dysfunction in ICU); there was no impact on the duration of ICU dependency (primary endpoint of the trial) and other secondary endpoints3 In the Leuven RCTs, the mortality benefit was observed on the primary and secondary endpoints of the adult surgical and paediatric trials, respectively. In NICE-SUGAR, there was harm on the primary endpoint (90-day mortality). In TGC-Fast, 90-day mortality was a safety endpointAbbreviations: BG: blood glucose: ICU: intensive care unit; IV: intravenous; PN: parenteral nutrition; RCT: randomised controlled trial Table 1Landmark RCTs on tight glucose control in ICULeuven RCTsNICE-SUGARTGC-FastPatient populationAdult, surgical ICUAdult, medical ICUPaediatric ICUAdult, mixed ICUAdult, mixed ICUN3448 (single centre)6104 (multicentre)9230 (multicentre)BG target Control group<215 mg/dL<180 mg/dL<215 mg/dL Intervention groupAge-adjusted(80–110 mg/dL for adults)80–110 mg/dL80–110 mg/dLBG measurement Measurement devicePredominantly blood gas analyserNot standardised. Inaccurate glucometers allowedPredominantly blood gas analyser Measurement sitePredominantly arterialNot standardised. Arterial, venous, capillaryPredominantly arterialInsulin protocol Administration modeContinuous IVContinuous + bolusContinuous IV Titration protocolTrained ICU nursesUnvalidated if-then protocolLOGIC-Insulin algorithmFeeding strategyEarly PNNo early PNNo early PNGlucose metrics Achieved difference in BG1±50 mg/dL in adults (105 vs. 152 mg/dL)±30 mg/dL(115 vs. 144 mg/dL)±30 mg/dL(107 vs. 140 mg/dL) Severe hypoglycaemiaIncreased 6-fold in adultsIncreased 14-foldUnaffectedOutcome Morbidity2DecreasedUnaffectedDecreased for selected outcomes Mortality3DecreasedIncreasedUnaffectedTo convert blood glucose in mg/dL to mmol/L, multiply by 0.055511 The reported achieved blood glucose concentrations are the mean blood glucose concentrations in ICU for the adult Leuven RCTs [15], the mean time-weighted blood glucose concentration in ICU for the NICE-SUGAR RCT, and the median morning blood glucose concentration in ICU for the TGC-Fast RCT2 The morbidity benefit in the adult Leuven RCTs was observed on secondary endpoints that included, among others, acute kidney injury, need for renal replacement therapy, cholestatic liver dysfunction in ICU, and duration of ICU and hospital dependency. The morbidity benefit in the paediatric Leuven RCT was observed on the primary endpoint (duration of ICU dependency and inflammation). The morbidity benefit in TGC-Fast was observed on selected secondary endpoints (severe acute kidney injury, need for renal replacement therapy, and cholestatic liver dysfunction in ICU); there was no impact on the duration of ICU dependency (primary endpoint of the trial) and other secondary endpoints3 In the Leuven RCTs, the mortality benefit was observed on the primary and secondary endpoints of the adult surgical and paediatric trials, respectively. In NICE-SUGAR, there was harm on the primary endpoint (90-day mortality). In TGC-Fast, 90-day mortality was a safety endpointAbbreviations: BG: blood glucose: ICU: intensive care unit; IV: intravenous; PN: parenteral nutrition; RCT: randomised controlled trial Landmark RCTs on tight glucose control in ICU Adult, surgical ICU Adult, medical ICU Paediatric ICU Age-adjusted (80–110 mg/dL for adults) ±30 mg/dL (115 vs. 144 mg/dL) ±30 mg/dL (107 vs. 140 mg/dL) To convert blood glucose in mg/dL to mmol/L, multiply by 0.05551 1 The reported achieved blood glucose concentrations are the mean blood glucose concentrations in ICU for the adult Leuven RCTs [15], the mean time-weighted blood glucose concentration in ICU for the NICE-SUGAR RCT, and the median morning blood glucose concentration in ICU for the TGC-Fast RCT 2 The morbidity benefit in the adult Leuven RCTs was observed on secondary endpoints that included, among others, acute kidney injury, need for renal replacement therapy, cholestatic liver dysfunction in ICU, and duration of ICU and hospital dependency. The morbidity benefit in the paediatric Leuven RCT was observed on the primary endpoint (duration of ICU dependency and inflammation). The morbidity benefit in TGC-Fast was observed on selected secondary endpoints (severe acute kidney injury, need for renal replacement therapy, and cholestatic liver dysfunction in ICU); there was no impact on the duration of ICU dependency (primary endpoint of the trial) and other secondary endpoints 3 In the Leuven RCTs, the mortality benefit was observed on the primary and secondary endpoints of the adult surgical and paediatric trials, respectively. In NICE-SUGAR, there was harm on the primary endpoint (90-day mortality). In TGC-Fast, 90-day mortality was a safety endpoint Abbreviations: BG: blood glucose: ICU: intensive care unit; IV: intravenous; PN: parenteral nutrition; RCT: randomised controlled trial The current recommended nutritional strategy of withholding early high-dose PN will decrease the severity of hyperglycaemia, but may potentially increase the risk of harmful severe hypoglycaemia (<40 mg/dL [2.2 mmol/L]) if tight glucose control is applied [26]. In this regard, the multicentre NICE-SUGAR trial, comprising 6104 patients, found that tight glucose control in the absence of early PN significantly increased mortality as compared with tolerating hyperglycaemia up to 180 mg/dL (10.0 mmol/L) [5]. As morbidity was unaffected in NICE-SUGAR, the exact reason for the increased mortality remains unclear; the documented 14-fold increase in severe hypoglycaemic events and the absence of a protocol to frequently measure potassium in order to detect insulin-induced hypokalaemia have been hypothesised to possibly contribute [27, 28]. As before-after implementation studies initiated after the Leuven RCTs had shown benefit through adoption of tight, or close-to-tight blood glucose targets [29–31], standard care had changed in many centres, and the control group target in NICE-SUGAR was set slightly lower as compared with the control group target in the Leuven RCTs [5]. Therefore, the results from NICE-SUGAR suggest that such intermediate control may be preferred in patients not receiving early PN [32]. However, the relatively high incidence in hypoglycaemic events in the NICE-SUGAR intervention group may be partially explained by the allowance of potentially inaccurate blood glucose measurements and the use of insulin boluses which may have induced prolonged and undetected episodes of hypoglycaemia [28, 33]. This contrasted with the standardised Leuven protocol that recommended accurate arterial blood glucose (and potassium) measurements by a blood gas analyser, and avoided insulin boluses [28]. Following the Leuven RCTs, the glucose control protocol was further optimised by development and validation of a computer algorithm, the LOGIC-Insulin algorithm, which advises on insulin dose, glucose dose in case of a hypoglycaemic event, and the timing of the next blood glucose measurement [34, 35]. Use of the LOGIC-Insulin algorithm led to a high time-in-target range while avoiding severe hypoglycaemia [34, 35]. The most recent and largest (9230 patients) multicentre TGC-Fast RCT, in patients not receiving early PN, found that LOGIC-Insulin-guided tight glucose control did not reduce the duration of ICU stay (the primary endpoint of the trial), but was associated with a reduced incidence of severe acute kidney injury, less need for renal replacement therapy and less cholestatic liver dysfunction as compared with tolerating hyperglycaemia up to 215 mg/dL (11.9 mmol/L) [14]. Other secondary outcomes were not affected. An individual patient data meta-analysis found no benefit from tight glucose control, which appeared to be consistent among preplanned subgroups [36]. However, apart from only including individual patient data for less than half of all eligible patients, the meta-analysis found tight glucose control significantly increased severe hypoglycaemic events, which may have counteracted any potential benefit from preventing hyperglycaemia. The most recent systematic review and meta-analysis identified 45 RCTs including altogether 32,215 patients [37]. In the total population, tight glucose control was not associated with an altered mortality risk as compared with more liberal control, but it was associated with a reduced ICU length of stay, less infections, and a lower incidence of critical illness polyneuropathy, while increasing the risk of severe hypoglycaemia. In subgroup analyses, tight glucose control was associated with favourable neurologic outcomes in neurointensive care patients, a lower ICU length of stay in mixed ICU patients, and a lower ICU mortality in cardiac surgery patients. Nevertheless, given the considerable heterogeneity in design between the included studies, one should be cautious to infer strong conclusions based on these findings. Individualised blood glucose control based on patient characteristics Observational evidence suggests that blood glucose targets may need to be individualised. Indeed, the U-shaped association between blood glucose concentrations in the ICU and mortality was shown to be flattened in patients with diabetes, with the nadir shifted to the right [38–42]. However, in NICE-SUGAR and TGC-Fast, there was no differential effect of the intervention on the primary endpoint according to diabetic status [5, 14]. The individual patient data meta-analysis also did not reveal any heterogeneity of treatment effect in patients with or without diabetes [36]. Therefore, we do not recommend differentiating blood glucose targets based on diabetes status. The above analyses did not take into account the level of blood glucose control pre-ICU admission. Recent investigations have utilised the glycaemic ratio, calculated as the ratio of the mean blood glucose concentration in ICU and the estimated average pre-admission blood glucose concentration based on the haemoglobin A1c (HbA1c) level [43]. These studies found that a glycaemic ratio below 0.7–0.8 or above 1.1 (corresponding to blood glucose concentrations 20–30% lower or 10% higher than the average pre-admission blood glucose) were associated with increased mortality rates [44, 45]. This U-shaped association between the glycaemic ratio and mortality was observed in both patients with normal and high HbA1c [44]. Thus, apart from severe hypoglycaemia (<40 mg/dL [2.2 mmol/L]), also relative hypoglycaemia (defined as a glycaemic ratio < 0.7 in patients with a raised HbA1c level [≥8%]) has been associated with increased mortality [46–48]. Nevertheless, causality remains unclear [40]. Only one before-after trial and one large RCT have tested blood glucose targets based on pre-admission blood glucose control. The before-after implementation study (N = 1979) found that implementation of looser blood glucose targets for patients with diabetes and HbA1c ≥ 7% (110–160 mg/dL [6.1–8.9 mmol/L]) was associated with decreased mortality compared against a historical control cohort in whom blood glucose was targeted at 90–120 mg/dL (5.0–6.7 mmol/L) [49]. The largest RCT (N = 2075) investigating individualised, HbA1c-guided blood glucose control as compared with conventional control (<180 mg/dL [10.0 mmol/L]) in ICU patients however found no benefit from individualised treatment [50]. Of note, there was substantial overlap in achieved blood glucose levels between groups, because of delayed randomisation, and a considerable number of patients with mid-range HbA1c values. Moreover, the protocol was not validated to achieve individualised targets while avoiding hypoglycaemia and large blood glucose variability [50]. Therefore, individualised, HbA1c-guided blood glucose control requires further study to evaluate efficacy and safety, before it can be widely recommended for routine clinical practice [51]. While awaiting further evidence, very liberal blood glucose control should likely be avoided in all ICU patients. The LUCID RCT, conducted in 419 ICU patients with type 2 diabetes, found that very liberal blood glucose control (180–252 mg/dL [10.0–14.0 mmol/L]), compared against standard care (targeting blood glucose between 108 and 180 mg/dL [6.0–10.0 mmol/L]), reduced the incidence of mild hypoglycaemia (<72 mg/dL [4.0 mmol/L]; the primary endpoint), but it was not associated with improved patient-centred outcomes, and was associated with a non-significant 4.6% increase in mortality [52]. In this trial, severe hypoglycaemic events were rare and not different between groups. A second subgroup that may respond differently to blood-glucose-lowering strategies is the subgroup of neurointensive care patients [53–56]. Indeed, the brain may be particularly vulnerable to both hyperglycaemic and hypoglycaemic damage [57–59]. The TGC-Fast RCT revealed significant heterogeneity of treatment effects for neurointensive care patients, suggesting a potential mortality benefit by tight glucose control in this subgroup (N = 1053) [14]. Importantly, these results, which should be considered hypothesis-generating and require confirmation, were obtained within the context of avoiding severe hypoglycaemia [14]. In the absence of such a protocol, the increased risk of hypoglycaemia could offset any benefit from preventing hyperglycaemia [60]. Secondary analyses of two paediatric RCTs have suggested that patients with severe inflammation at baseline would particularly benefit from tight glucose control [61, 62]. However, in these patients, the intervention was also associated with a greater decrease in blood glucose concentrations as compared with patients without hyperinflammation, likely because of a higher illness severity in the hyperinflamed patients [61]. Hence, it remains unclear whether hyperinflammation per se increases vulnerability to hyperglycaemic damage, or whether hyperinflammation is merely a proxy for more severe hyperglycaemia. In this context, landmark RCTs have not shown a differential response to tight glucose control in patients with or without sepsis [5, 14, 36]. What to do in clinical practice? Taken together, RCT data suggest that both severe hyperglycaemia and hypoglycaemia should be prevented [63]. Large blood glucose variability should likely also be avoided, as this has been associated with harm [38, 64]. The mortality benefit achieved through tight glucose control has only been shown in a context of iatrogenic hyperglycaemia induced by early PN [2–4]. The less pronounced potential benefit of tight glucose control in the absence of early PN, even with a protocol that avoided severe hypoglycaemia, supports the avoidance of interventions that induce iatrogenic hyperglycaemia, such as administration of early high doses of glucose and PN, or corticosteroid use without evidence-based indication. [14, 63]. Tight glucose control performed with a protocol that induces severe hypoglycaemia could be harmful and should be discouraged [5, 27]. After avoidance of unnecessary iatrogenic causes of hyperglycaemia such as early high-dose glucose or PN, it thus appears advisable to adapt the blood glucose target according to the clinical context (Fig. 1) [63]. Safe blood glucose control requires a protocol that avoids hyperglycaemia, hypoglycaemia and large blood glucose fluctuations, and this can be achieved by frequent and accurate blood glucose measurements, and avoidance of insulin boluses. Arterial measurements are preferred, as capillary measurements may be unreliable in patients with poor tissue perfusion and patients receiving vasopressors, and venous measurements may be unreliable in patients with concurrent glucose infusions [26]. Blood gas analysers are preferred, as these provide a fast, accurate measurement of blood glucose, and a simultaneous measurement of potassium, to detect and treat insulin-induced hypokalaemia [63]. If glucometers are used, measurement accuracy needs to be confirmed [33]. Insulin is preferably administered by continuous intravenous infusion through a central venous line. If such an accurate protocol is not available or non-feasible, if the patient is able to eat, or does not need arterial and central venous access, and/or if the team is not experienced with blood glucose control, we suggest a liberal blood glucose target, initiating insulin only when blood glucose exceeds 185–215 mg/dL (10.3–11.9 mmol/L) [63]. If all above mentioned conditions are met, then we advise at least intermediate blood glucose control, initiating insulin when blood glucose exceeds 150–180 mg/dL (8.3–10.0 mmol/L) [63]. As indicated, we do not suggest one upper limit of liberal and intermediate control, but rather a range of possible upper limits. This is a pragmatic choice, and reflects the overall uncertainty of evidence that does not clearly favour one particular target. We suggest that clinicians select an appropriate upper limit depending on their specific context; the more experienced the team and the more safeguards in place to prevent severe hypoglycaemia and large blood glucose fluctuations, the lower the upper target can possibly be. Tight glucose control (80–110 mg/dL [4.4–6.1 mmol/L]) should only be considered if this is possible without inducing severe hypoglycaemic events, which requires a validated protocol [63]. As such validated glucose control protocols are often unavailable, intermediate blood glucose control should likely be the default setting for most patients in the majority of ICUs. Available data from RCTs do not support the use of specific targets for particular subgroups of patients, including patients with diabetes, neurointensive care patients, and patients with severe inflammation. If insulin is still needed to prevent severe hyperglycaemia at ICU discharge or upon resumption of oral feeding, maintenance antidiabetic treatment should be initiated in consultation with a diabetes specialist [63]. These suggested targets align well with current guidelines from the American Diabetes Association and the Society of Critical Care Medicine that recommend at least an intermediate blood glucose target (keeping blood glucose below 180 mg/dL [10.0 mmol/L]) for ICU patients, and leave open the possibility for tighter control, if achievable without inducing severe hypoglycaemia [65, 66]. These guidelines however do not consider more liberal control if not all criteria for safe blood glucose lowering by insulin treatment are fulfilled. Here we suggest clear criteria as to how to differentiate the target according to the specific context, allowing better individualisation of treatment.Fig. 1Proposed expert-based algorithm for blood glucose control in the intensive care unit. The proposed thresholds are pragmatic, context-based recommendations derived from trial interpretation and current guidelines; these have not directly been validated as an algorithm in a prospective RCT. Liberal use of corticosteroids means corticosteroid administration in the absence of a clear evidence-based indication. * As explained in the text, we suggest a range of possible upper limits for intermediate and liberal blood glucose control, reflecting the overall uncertainty of evidence that does not clearly favour one target for all patients. We suggest that clinicians select an appropriate upper limit depending on their specific context; the more experienced the team and the more safeguards in place to prevent severe hypoglycaemia and large blood glucose fluctuations, the lower the upper target can possibly be. Abbreviations: PN: parenteral nutrition; TGC: tight glucose control Proposed expert-based algorithm for blood glucose control in the intensive care unit. The proposed thresholds are pragmatic, context-based recommendations derived from trial interpretation and current guidelines; these have not directly been validated as an algorithm in a prospective RCT. Liberal use of corticosteroids means corticosteroid administration in the absence of a clear evidence-based indication. * As explained in the text, we suggest a range of possible upper limits for intermediate and liberal blood glucose control, reflecting the overall uncertainty of evidence that does not clearly favour one target for all patients. We suggest that clinicians select an appropriate upper limit depending on their specific context; the more experienced the team and the more safeguards in place to prevent severe hypoglycaemia and large blood glucose fluctuations, the lower the upper target can possibly be. Abbreviations: PN: parenteral nutrition; TGC: tight glucose control Areas of future research Strategies to optimise and individualise blood glucose control As recent evidence suggests that tight glucose control may be preferred, provided that it can be applied without inducing severe hypoglycaemia, future research should aim at optimising strategies that further improve the quality of blood glucose control by improving the time-in-target range, and study whether implementation of such strategy improves patient-centred outcomes. Potential strategies include continuous glucose monitoring (CGM), implementation of validated glucose control algorithms, and closed-loop blood glucose control [53, 67]. Although CGM is widely used in people with diabetes in the community, implementation in critical care is hampered by technological and financial constraints [53]. A main drawback of interstitial blood glucose measurements is the inherent lag time, and potential inaccuracy in patients with poor tissue perfusion. Although this constraint could be overcome by intravascular glucose sensors, such sensors are not widely available [68]. Theoretically, combination of CGM with a validated glucose control protocol into a closed-loop system could lead to optimal blood glucose control and improve patient outcomes. A single-centre RCT of 447 surgical ICU patients revealed that tight glucose control (80–110 mg/dL [4.4–6.1 mmol/L]) performed using a closed-loop system with continuous low-volume blood sampling was superior to an intermediate glucose control regimen (139–180 mg/dL [7.7–10.0 mmol/L]), with a reduction in surgical site infections (the primary endpoint of the trial) [69]. These patients however received early PN, and these results have not been confirmed in other contexts. Hence, the results require confirmation in RCTs adequately powered for patient-centred outcomes. Apart from that, outside Japan, closed-loop glucose control systems are not readily available for use in ICU patients. Future research could implement artificial intelligence, not only to optimise insulin titration, but also to individualise blood glucose targets and nutrition therapy. The preferred blood glucose target may depend not only on context, but also the optimal timing of nutrition support initiation [23, 70]. Aggregated evidence from the glucose control RCTs suggests that early nutrition support may be harmful in part by aggravating hyperglycaemia, although other mechanisms have been implicated as well [23, 26]. Consequently, a hyperglycaemic response to nutrition may indicate a metabolic unreadiness to tolerate nutrition support, although this hypothesis requires validation [23]. Theoretically, artificial intelligence could assist in identifying patterns that predict the blood glucose response to nutrition and the preferred time point to initiate and titrate nutrition support and insulin, which requires further study. Future research should also investigate to what extent in-ICU glucose control could prevent long-term diabetes mellitus. Recent large observational studies have shown that severe stress hyperglycaemia is independently associated with an increased risk of long-term diabetes [71, 72]. This association has also been observed in patients with a normal HbA1c upon ICU admission [73], hence decreasing the possibility of unknown diabetes that manifests months or years later. These data suggest that the development of (severe) stress hyperglycaemia may indicate a propensity to (severe) insulin resistance, and to diabetes mellitus, as is the case for gestational diabetes mellitus [74]. As severe and prolonged hyperglycaemia could damage pancreatic beta-cells, any potential long-term effect of tight glucose control in ICU on the development of subsequent diabetes mellitus requires further investigation [75]. Alternative blood-glucose-lowering strategies Apart from advanced monitoring and treatment algorithms, another way to improve blood glucose control and reduce the risk of hypoglycaemia is to study alternative glucose-lowering strategies, including alternative feeding strategies and alternative glucose-lowering agents [63]. Theoretically, strategies that have an insulin-sparing effect could also be beneficial via stimulation of cellular repair [26]. Insulin has suppressive effects on autophagy and ketogenesis which have been identified as potentially crucial processes in cellular repair (Fig. 2) [70, 76]. Apart from harm through aggravating hyperglycaemia, suppression of autophagy and ketogenesis are further mechanisms that may explain harm from early full nutrition support [76–79]. Whereas insulin could be protective by prevention of glucose overload and associated toxicity in vital tissues, its beneficial effects could in part be counteracted by suppressive effects on cellular repair [26].Fig. 2Simplified summary of pathophysiological mechanisms that may explain the effects of early parenteral nutrition and tight glucose control. This figure provides a simplified framework of hypothesised mechanisms that are supported by observational patient data; direct proof of causality is mainly derived from animal studies, which cannot necessarily be extrapolated to the human condition. Moreover, other mechanisms may play a role. The beneficial effects of tight glucose control have been attributed to prevention of glucose overload and associated toxicity in vital cells with insulin-independent glucose uptake. Harm from early parenteral nutrition has been attributed to aggravation of hyperglycaemia-induced cell damage, and to suppression of beneficial repair processes that are activated by fasting, including autophagy and ketogenesis. As with nutrition, insulin has suppressive effects on autophagy and ketogenesis. The red arrow indicates an aggravating effect; dashed lines ending with a perpendicular bar indicate a potentially suppressive effect. The green colour indicates a potentially protective effect, red a potentially harmful effect Simplified summary of pathophysiological mechanisms that may explain the effects of early parenteral nutrition and tight glucose control. This figure provides a simplified framework of hypothesised mechanisms that are supported by observational patient data; direct proof of causality is mainly derived from animal studies, which cannot necessarily be extrapolated to the human condition. Moreover, other mechanisms may play a role. The beneficial effects of tight glucose control have been attributed to prevention of glucose overload and associated toxicity in vital cells with insulin-independent glucose uptake. Harm from early parenteral nutrition has been attributed to aggravation of hyperglycaemia-induced cell damage, and to suppression of beneficial repair processes that are activated by fasting, including autophagy and ketogenesis. As with nutrition, insulin has suppressive effects on autophagy and ketogenesis. The red arrow indicates an aggravating effect; dashed lines ending with a perpendicular bar indicate a potentially suppressive effect. The green colour indicates a potentially protective effect, red a potentially harmful effect Alternative feeding strategies that may improve glucose tolerance and reduce insulin need include intermittent fasting strategies and ketogenic diets [80–84]. Although a fasting-mimicking effect associated with such interventions could be beneficial [83, 84], the efficacy and safety of such feeding strategies in critically ill patients need to be confirmed. With regard to alternative glucose-lowering agents, sodium-glucose cotransporter-2 inhibitors require further investigation. The DEFENDER RCT (N = 507) found, compared to standard care, that dapagliflozin treatment in critically ill patients did not improve the hierarchical composite primary endpoint of mortality, renal replacement therapy and ICU length of stay, while it was associated with a potential reduction in the need for renal replacement therapy [85, 86]. Although these results require confirmation, this potential effect was comparable to the potential renoprotective effect of tight glucose control seen in TGC-Fast [14]. However, the mechanism for potential renal protection in DEFENDER (if any) remains unclear, and alternative pathways apart from improved glucose control may have played a role. Moreover, given the potential risk of severe adverse effects including euglycaemic ketoacidosis, use of dapagliflozin and other alternative glucose-lowering agents should not be used outside the context of a clinical trial. Conclusion The optimal blood glucose target likely depends on the accuracy of the blood glucose measurement, insulin titration protocol, and the nutritional strategy. Current evidence suggests to avoid iatrogenic causes of hyperglycaemia, and to avoid severe hyperglycaemia and hypoglycaemia. Safe blood glucose control in ICU patients requires accurate and frequent blood glucose measurements, and avoidance of insulin boluses. Tight glucose control may be considered in selected settings if it can be applied with a protocol that avoids severe hypoglycaemia; in the absence of such protocol, the risks outweigh any potential benefit, and intermediate blood glucose control is likely the default strategy for most patients. Future research should focus on strategies that could further optimise blood glucose control. This includes studying patient-specific factors that could lead to further individualisation of blood glucose control, as well as investigating glucose-lowering strategies that have an insulin-sparing effect and that do not increase the risk of hypoglycaemia.